Expert Clouds and Applications: Proceedings of ICOECA 2022 (Lecture Notes in Networks and Systems, 444) 9811924996, 9789811924996

The book features original papers from International Conference on Expert Clouds and Applications (ICOECA 2022), organiz

273 27 23MB

English Pages 823 [785] Year 2022

Report DMCA / Copyright

DOWNLOAD PDF FILE

Table of contents :
Foreword
Preface
Acknowledgments
Contents
Editors and Contributors
An Approach for Summarizing Text Using Sentence Scoring with Key Optimizer
1 Introduction
2 Related Works
3 Methodology
3.1 Data Acquisition
3.2 Essential Pre-processing Steps
3.3 Similarity Computation with Weighted Features
3.4 Feature Weight Scoring
3.5 Sentence Scoring with KEY Optimizer
3.6 Summarization Process
4 Numerical Results and Discussion
5 Conclusion
References
Image Classification of Indian Rural Development Projects Using Transfer Learning and CNN
1 Introduction
1.1 Dataset
2 Background
3 Literature Review
3.1 Works Related to Image Classification of Rural Projects
3.2 Works Related to Data Augmentation Techniques
3.3 Gaps Identified in the Related Works
3.4 Contributions of This Paper
4 Proposed Work
4.1 Experimental Setup and Analysis
4.2 DenseNet121
4.3 Remote-Sensing (RS) Architecture
4.4 Comparative Analysis
5 Conclusion
References
Design of a Mobile Application to Deal with Bullying
1 Introduction
2 Literature Review
3 Proposed Work
4 Results and Discussion
5 Conclusion
References
Selection of Human Resources Prospective Student Using SAW and AHP Methods
1 Introduction
2 Current and Previous Similar Research Papers
3 Simple Additive Weighting Implementation
3.1 List of Weighting Criteria, Including the Range of Criteria and Normalization of Criteria Range
3.2 Scoring of Sample Data Using a Normalized Range of Criteria
3.3 Normalization of Performance Rating Score
3.4 Ranking Score
4 Analytical Hierarchy Process Implementation
4.1 Processing Criteria
4.2 Processing Alternatives
5 Calculation Result of SAW and AHP Method
6 Conclusions
References
Implementation of the Weighted Product Method to Specify scholarship’s Receiver
1 Introduction
2 Current and Previous Similar Research Papers
3 Result and Discussion
3.1 System Modeling
3.2 Weighted Product Implementation
3.3 System Implementation
4 Conclusion
References
A Survey on E-Commerce Sentiment Analysis
1 Introduction
2 Related Works
3 Conclusion
References
Secured Cloud Computing for Medical Database Monitoring Using Machine Learning Techniques
1 Introduction
2 Literature Survey
3 Materials and Methodology
4 Implementation
5 Conclusion
References
Real-Time Big Data Analytics for Improving Sales in the Retail Industry via the Use of Internet of Things Beacons
1 Introduction
2 Literature Survey
3 Materials and Methodology
3.1 Apache Spark for Real-Time Data Analytics:
3.2 Proposed Ideal Beacon Selection
3.3 Analysis and Results
4 Conclusion
References
Healthcare Application System with Cyber-Security Using Machine Learning Techniques
1 Introduction
2 Related Work
3 Materials and Methods
3.1 Proposed Methodology
4 Implementation
4.1 Camera Specifications
4.2 Comparison with Existing Algorithm
5 Conclusion
References
Energy Efficient Data Accumulation Scheme Based on ABC Algorithm with Mobile Sink for IWSN
1 Introduction
2 Literature Survey
3 Materials and Methodology
3.1 ABC Algorithm
3.2 Proposed Ideal Path Selection by ABC Algorithm
3.3 Analysis and Results
4 Conclusion
References
IoT Based Automatic Medicine Reminder
1 Introduction
2 Background
2.1 Cloud Computing and IoT
3 Proposed System
4 System Reqirements
5 Hardware Requirements
6 Hardware Used
6.1 Arduino Board
6.2 GSM (Worldwide Framework for Portable)
6.3 LCD (Liquid Crystal Display)
6.4 RTC (Real-Time Clock)
7 Results
7.1 Unit Testing
8 Conclusion
References
IoT Based Framework for the Detection of Face Mask and Temperature, to Restrict the Spread of Covids
1 Introduction
2 Related Background
3 Proposed Work
4 Results
5 Conclusion
References
Efficient Data Partitioning and Retrieval Using Modified ReDDE Technique
1 Introduction
2 Literature Survey
3 Methodology
3.1 Data Node
3.2 Shard
4 Implementation
4.1 Pre-Processing of Data
4.2 Clustering of Data
4.3 Data Loading
4.4 Input Query
4.5 Shard Selection
4.6 Data Retrieval
5 Results and Discussion
6 Conclusion
References
Stock Price Prediction Using Data Mining with Soft Computing Technique
1 Introduction
2 Related Work
3 Proposed Work
3.1 Procedure
3.2 Proposed Model
3.3 Clustering Algorithm
3.4 Clustering Technique
3.5 Proposed Algorithm
3.6 Implementation of Novel LSTM Approach
3.7 Data Set
4 Results
5 Conclusion
References
A Complete Analysis of Communication Protocols Used in IoT
1 Introduction
2 Characteristics of IoT
3 Background and Related Work
4 IoT Architecture
5 IoT Protocols
6 Conclusion
References
Image Captioning Using Deep Learning Model
1 Introduction
2 Review of Literature
3 Literature Table
4 Proposed work
5 Observation and Result
6 Conclusion
References
Recommender System Using Knowledge Graph and Ontology: A Survey
1 Introduction
2 Background and Related Work
2.1 Concept of Knowledge Graph
2.2 Literature Work
3 Observation and Discussion
3.1 Ontology-Based
3.2 Challenges in Recommender System
3.3 Challenges in Knowledge Graph
3.4 Available Tools and Dataset
4 Proposed Idea
5 Conclusion and Future Work
References
A Review of the Multistage Algorithm
1 Introduction
2 Exploration of Multistage Algorithm Publication
3 Classification of Multistage Algorithm Publication
4 Review of Multistage Algorithm Publication
5 Discussion and Opinion
6 Conclusion
References
Technology for Disabled with Smartphone Apps for Blind People
1 Introduction
2 Existing Works
3 Proposed Smartphone Application
4 Conclusion
References
Mobile Apps for Musician Community in Indonesia
1 Introduction
2 Existing Works
3 Proposed Idea
4 Conclusion
References
A Genetic-Based Virtual Machine Placement Algorithm for Cloud Datacenter
1 Introduction
2 Related Work
3 Placement Algorithms
4 Performance Evaluation
4.1 Experimental Setup
4.2 Evaluation of FF and BF Algorithm
4.3 Evaluation of IGA Algorithm
5 Conclusion
References
Visual Attention-Based Optic Disc Detection System Using Machine Learning Algorithms
1 Introduction
2 Related Work
3 Proposed Visual Attention-Based Optic Disc Detection System
3.1 Pre-processing
3.2 Processing
3.3 Post-processing
4 Results
5 Conclusion
References
An Overview of Blue Eye Constitution and Applications
1 Introduction
2 Technologies Used in Blue Eye
2.1 Manual and Gaze Input Cascaded (MAGIC) Pointing
2.2 Artificial Intelligence Speech Recognition
2.3 Simple User Interest Tracker (SUITOR)
2.4 Emotion Mouse
3 Construction of Blue Eye
3.1 Software Required
3.2 Hardware Required
4 Applications of Blue Eye
5 Emotion Computing Using Blue Eye Technology
6 Results
7 Conclusion
References
Enhancement of Smart Contact on Blockchain Security by Integrating Advanced Hashing Mechanism
1 Introduction
1.1 Blockchain
1.2 Smart Contracts
1.3 Blockchain Hash Function
1.4 Hashing Techniques
1.5 Encryption Algorithm Based on RSA
2 Literature Review
3 Problem Statements
4 Proposed Work
5 Results and Discussion
6 Conclusions
7 Future Scopes
References
Evaluation of Covid-19 Ontologies Through OntoMetrics and OOPS! Tools
1 Introduction
2 Literature
3 Ontology at a Glance
3.1 Ontology Development Methodologies
3.2 Available Covid-19 Ontologies
4 Evaluation of Covid-19 Ontologies
4.1 OntoMetrics Tool
4.2 Ontology Pitfall Scanner! (OOPS!)
5 Conclusion and Future Work
References
Recognition of the Multioriented Text Based on Deep Learning
1 Introduction
2 Related Work
3 Methodology
4 Results and Discussions
5 Conclusion
References
Fruit and Leaf Disease Detection Based on Image Processing and Machine Learning Algorithms
1 Introduction
2 Background
3 Methodology
4 Results and Discussion
5 Conclusion
References
A Survey for Determining Patterns in the Severity of COVID Patients Using Machine Learning Algorithm
1 Introduction
2 Related Works
3 Conclusion
References
Relation Extraction Between Entities on Textual News Data
1 Introduction
2 Literature Survey
2.1 Recurrent Neural Networks
2.2 Gated Recurrent Units
2.3 Conditional Random Fields
2.4 Named Entity Recognition
3 Methodology
3.1 Dataset
3.2 Word Embedding
3.3 Bi-Directional Gated Neural Networks
3.4 Conditional Random Fields
3.5 Multi-Head Selection
4 Results and Discussion
5 Conclusion
References
A Comprehensive Survey on Compact MIMO Antenna Systems for Ultra Wideband Applications
1 Introduction
2 UWB MIMO Antenna, Classifications, and Its Performance Enhancement Techniques
2.1 UWB MIMO Antennas
2.2 Suitable Antennas for 5G Applications
2.3 Antenna Performance Enhancement Techniques
3 UWB MIMO Antenna
3.1 Printed UWB MIMO Antenna
3.2 High-Isolation Compact MIMO Antenna
3.3 Compact UWB Four Antenna Element MIMO Antenna Using CPW Fed
3.4 8-Element Based UWB MIMO Antenna with Band-Notch and Reduced Mutual Coupling
3.5 WLAN Band Rejected 8-Element Based Compact UWB MIMO Antenna
3.6 Comparison of UWB MIMO Wideband Antennas Structures
4 Conclusion
References
A Review of Blockchain Consensus Algorithm
1 Introduction
1.1 Blockchain Technology
2 Related Work
2.1 Proof of Work Consensus Algorithm
2.2 Proof of Stake Consensus Algorithm
2.3 Proof of Elapsed Time
2.4 Byzantine Fault Tolerance
3 Challenges/Findings
3.1 Proof of Work
3.2 Proof of Stake
3.3 Proof of Elapsed Time
3.4 Byzantine Fault Tolerance
4 Conclusion
References
A Short Systematic Survey on Precision Agriculture
1 Introduction
1.1 Precision Agriculture
1.2 Machine Learning Approaches
1.3 Review of Machine Learning Approaches
1.4 Analysis of Learning Models
2 Related Work
2.1 Flora Test in Data Acquisition
2.2 Production Management System
2.3 Disease Detection Techniques
2.4 Soil Classification
2.5 Production Phase
3 Precision Agriculture System
3.1 Precision Agriculture Routing Protocols Using WSN
3.2 IoT Computing Architecture in Precision Agriculture
4 Comparison Accuracy
5 of soil accessibility in agro zone districtsConclusion
References
Improving System Performance Using Distributed Time Slot Assignment and Scheduling Algorithm in Wireless Mesh Networks
1 Introduction
2 Related Work
3 Effective Distributed Time-Based Schedule for Channel Assignment
4 Performance Evaluation
5 Conclusion
References
Comparison of PRESENT and KLEIN Ciphers Using Block RAMs of FPGA
1 Introduction
2 Related Work
3 The PRESENT Cipher
4 The KLEIN Cipher
5 Implementation
6 Results and Conclusion
6.1 Results
6.2 Conclusion
References
Remote Controlled Patrolling Robot
1 Introduction
1.1 Objective
1.2 Synopsis
2 Related Work—Literature Survey
2.1 Overview
3 Methodology- Proposed Work
3.1 Design Components
3.2 Overview
3.3 Movement of Robot
3.4 Designing the Live Stream
3.5 ThingSpeak Cloud Application in Raspberry Pi
4 Implementation and Result
4.1 Implementation
4.2 Result
5 Conclusion
6 Future Scope
References
Novel Modeling of Efficient Data Deduplication for Effective Redundancy Management in Cloud Environment
1 Introduction
2 Related Work
3 Problem Description
4 Proposed Methodology
5 System Design
5.1 Strategies of System Design
5.2 Assumptions and Dependencies
5.3 Algorithm Design
6 Results Discussion
6.1 Experimental Parameters
6.2 Test Environment
6.3 Result Analysis
7 Conclusion
References
A Survey on Patients Privacy Protection with Steganography and Visual Encryption
1 Introduction
1.1 Motivation
1.2 Survey Strategy and Evaluation
1.3 Paper Organization
2 Spatial Hiding Models
2.1 Least Significant Bit (LSB)
2.2 Most Significant Bit (MSB)
2.3 Other Spatial Hiding Models
3 Frequency Hiding Models
3.1 Discrete Cosine Transforms (DCT)
3.2 Discrete Wavelet Transform (DWT)
4 Visual Encryption Models
5 Analysis and Evaluatio
6 Conclusion
References
Survey on Channel Estimation Schemes for mmWave Massive MIMO Systems – Future Directions and Challenges
1 Introduction
2 mmWave MIMO Systems and Its Characteristics
2.1 Propagation Characteristics
2.2 mmWave Technology
3 Existing Architectures for Channel Estimation for Massive MIMO Systems
3.1 Tensor Based Channel Estimation Scheme
3.2 Beam Squint Based Channel Estimation Scheme
3.3 Frequency Domain Compressive Channel Estimation Scheme
3.4 Hybrid Beamforming Based Channel Estimation Schemes via Random Spatial Sampling
3.5 Time-Domain Based Channel Estimation with Hybrid Architecture
3.6 Comparison of Existing Channel Estimation Schemes
4 Conclusion
References
Fake News Detection: Fact or Cap
1 Introduction
2 Related Work
2.1 Pre-trained Linguistic Models
2.2 Clickbait Detection
2.3 Fake News Categorization
3 Methodology
3.1 Data Collection
3.2 Data Pre-processing
3.3 Feature Extraction
4 Results and Discussion
5 Conclusion
References
The Future of Hiring Through Artificial Intelligence by Human Resource Managers in India
1 Introduction
2 Literature Review
3 Research Methodology
4 Findings and Discussions
5 Conclusion
References
A Computer Vision Model for Detection of Water Pollutants Using Deep Learning Frameworks
1 Introduction
2 Related Work
3 Architecture Details
3.1 Dataset Development
3.2 Detection Algorithms
3.3 Custom Dataset Training
4 Experimental Details and Results
4.1 Training Environment
4.2 Hardware Specifications
5 Conclusion and Future Work
References
Medical IoT Data Analytics for Post-COVID Patient Monitoring
1 Introduction
2 Literature Review
3 IoT Based Medical Data Analytics
4 Data Collection
5 ML Approaches for Medical IoT
6 Post-COVID-19 Patient Monitoring System
7 Some of the Existing Models
8 Proposed System
9 Evaluation
10 Conclusion
References
SNAP—A Secured E Exchange Platform
1 Introduction
2 Literature Survey
3 Proposed System
4 System Architecture
5 Expected Results and Discussion
5.1 Experimental Setup
5.2 Third-Party Services
5.3 Expected Results
6 Conclusion and Future Scope
References
An AES-Based Efficient and Valid QR Code for Message Sharing Framework for Steganography
1 Introduction
2 Related Work
3 Proposed System
3.1 AES Algorithm
3.2 Encryption Process
4 Results and Discussion
4.1 Parameters Evaluation
4.2 Outcomes
4.3 Robustness Against Attacks
5 Conclusion
6 Future Scope
References
Categorization of Cardiac Arrhythmia from ECG Waveform by Using Super Vector Regression Method
1 Introduction
1.1 Electrocardiogram (ECG) Electrodes
2 Review of the Existing Literature
3 Methodology
3.1 Regression with Support Vectors
4 Results
5 Conclusion
References
Performance Analysis of Classification Algorithm Using Stacking and Ensemble Techniques
1 Introduction
2 Literature Survey
3 Methodology
3.1 Data Pre-processing
3.2 Feature Selection
3.3 Support Vector Machine (SVM)
3.4 Logistic Regression (LR)
3.5 Naive Bayes (NB)
3.6 Stacking Algorithms
4 Implementation
4.1 Data Pre-processing
4.2 Feature Selection
4.3 Data Partitioning
4.4 Support Vector Machine
4.5 Naive Bayes
4.6 Logistic Regression
5 Proposed Hybrid Algorithm
6 Results and Discussions
6.1 Performance Analysis of Existing Classification Algorithms
6.2 Performance Analysis of Proposed Stacking Technique
6.3 Comparative Study Analysis of Existing and Proposed Stacking Technique
7 Conclusion
References
Conceptual Study of Prevalent Methods for Cyber-Attack Prediction
1 Introduction
2 Background Work
2.1 Identifying the Attack Scenario
2.2 Prediction of Attacks
2.3 Various Techniques for Attack Prediction
3 Attack Graph Analysis
4 Attack Prediction with Deep Learning Approach
5 Hidden Markov Model Approach
6 Proposed Model
7 Conclusion and Future Work
References
Performance Analysis of Type-2 Diabetes Mellitus Prediction Using Machine Learning Algorithms: A Survey
1 Introduction
2 Implementation Techniques
2.1 Traditional Methods
2.2 Machine Learning Classifiers
2.3 Decision Tree Algorithm
2.4 Deep Learning
2.5 Hybrid Models
3 Conclusion and Future Scope
References
Dynamic Updating of Signatures for Improving the Performance of IDS
1 Introduction
2 Background of IDS
3 Proposed Framework
4 Experiment Environment and Result
5 Conclusion
References
M-mode Carotid Artery Image Classification and Risk Analysis Based on Machine Learning and Deep Learning Techniques
1 Introduction
2 Materials and Methods
2.1 Elastic Modulus
2.2 Stiffness Index
2.3 Arterial Distensibilty
2.4 Arterial Compliance
3 Results and Discussion
4 Conclusion
References
Analyzing a Chess Engine Based on Alpha–Beta Pruning, Enhanced with Iterative Deepening
1 Introduction
1.1 Background
1.2 Research Work
2 Game Tree of the System
3 Proposed Work
3.1 Move Generator
3.2 Evaluation Functions and Pieve Values
3.3 Search Algorithm
4 Findings
5 Future and Scalability
6 Conclusion
References
Intelligent Miniature Autonomous Vehicle
1 Introduction
2 Literature Survey
3 Methodology
3.1 Data Collection
3.2 Pre-processing of the Data
4 Results and Discussion of Model Training and Testing
5 Conclusion
References
Energy and Trust Efficient Cluster Head Selection in Wireless Sensor Networks Under Meta-Heuristic Model
1 Introduction
2 Literature Review
2.1 Related Works
3 Proposed Optimal CH Selection Model with Multi-objective Decision Making
3.1 Assumptions
3.2 Overview of the Proposed Work
3.3 Proposed Scheme
3.4 Proposed HCSO
4 Results and Discussions
4.1 Simulation Procedure
4.2 Analysis on the Count of Alive Node
4.3 Analysis on Normalized Network Energy
4.4 Analysis on Trust
4.5 Statistical Analysis on Count of Alive Nodes
4.6 Statistical Analysis on Normalized Energy
4.7 Statistical Analysis on Trust
4.8 Statistical Analysis on Separation Distance Between CHs
4.9 Computational Complexity
5 Conclusion
References
Performance Analysis of OTFS Scheme for TDL and CDL 3GPP Channel Models
1 Introduction
2 Orthogonal Time Frequency Space Modulation
2.1 OTFS Modulation
2.2 OFTS Demodulation
2.3 Equalization and Channel Estimation
3 Results and Discussion
3.1 TDL Channel Models
3.2 CDL Channel Models
4 Conclusion
References
Design of Smart Super Market Assistance for the Visually Impaired People Using YOLO Algorithm
1 Introduction
1.1 Inspiration Towards Research Work
1.2 Facts About Blind People [4]
2 Related Works
2.1 Assistive Clothing Pattern Recognition for Visually Impaired People
2.2 From Smart Objects to Social Objects: The Next Evolutionary Step of the Internet of Things
2.3 Accessible Shopping Systems for Blind and Visually Impaired Individuals: Design Requirements and the State of the Art
2.4 Touch Access for People with Disabilities
2.5 Graphene-Based Web Framework for Energy Efficient IoT Applications
2.6 Design of Deep Learning Algorithm for IoT Application by Image-Based Recognition
2.7 Computer Vision on IoT Based Patient Preference Management System
2.8 Study of Retail Applications with Virtual and Augmented Reality Technologies
3 The Proposed Work
3.1 Steps of the Research Work
3.2 Smart Glass with Their Nature of Working
3.3 Working Principle of YOLO Algorithm
3.4 Classification
4 Investigative Results
4.1 Building a Product Recognition System
4.2 Output Image
4.3 Graphical Representation
5 Conclusion
6 Future Work
References
Investigating the Effectiveness of Zero-Rated Website on Students on E-Learning
1 Introduction
2 Literature Review
2.1 The Nature of Zero-Rating of Websites
2.2 Content and Access Requirements for E-Learning
2.3 Content Accessibility on Zero-Rated Website
3 Research Methodology
4 Findings and Analysis
4.1 Nature of Zero-Rating
4.2 Content and Access Requirement for E-Learning
4.3 Levels of Content Accessibility on Zero-Rated University Website
5 Conclusion
References
Role of Machine Learning Algorithms on Alzheimer Disease Prediction
1 Introduction
2 Literature Survey
3 Proposed System
4 Results
5 Conclusion
References
Author Index
Recommend Papers

Expert Clouds and Applications: Proceedings of ICOECA 2022 (Lecture Notes in Networks and Systems, 444)
 9811924996, 9789811924996

  • 0 0 0
  • Like this paper and download? You can publish your own PDF file online for free in a few minutes! Sign Up
File loading please wait...
Citation preview

Lecture Notes in Networks and Systems 444

I. Jeena Jacob Selvanayaki Kolandapalayam Shanmugam Robert Bestak   Editors

Expert Clouds and Applications Proceedings of ICOECA 2022

Lecture Notes in Networks and Systems Volume 444

Series Editor Janusz Kacprzyk, Systems Research Institute, Polish Academy of Sciences, Warsaw, Poland Advisory Editors Fernando Gomide, Department of Computer Engineering and Automation—DCA, School of Electrical and Computer Engineering—FEEC, University of Campinas— UNICAMP, São Paulo, Brazil Okyay Kaynak, Department of Electrical and Electronic Engineering, Bogazici University, Istanbul, Turkey Derong Liu, Department of Electrical and Computer Engineering, University of Illinois at Chicago, Chicago, USA Institute of Automation, Chinese Academy of Sciences, Beijing, China Witold Pedrycz, Department of Electrical and Computer Engineering, University of Alberta, Alberta, Canada Systems Research Institute, Polish Academy of Sciences, Warsaw, Poland Marios M. Polycarpou, Department of Electrical and Computer Engineering, KIOS Research Center for Intelligent Systems and Networks, University of Cyprus, Nicosia, Cyprus Imre J. Rudas, Óbuda University, Budapest, Hungary Jun Wang, Department of Computer Science, City University of Hong Kong, Kowloon, Hong Kong

The series “Lecture Notes in Networks and Systems” publishes the latest developments in Networks and Systems—quickly, informally and with high quality. Original research reported in proceedings and post-proceedings represents the core of LNNS. Volumes published in LNNS embrace all aspects and subfields of, as well as new challenges in, Networks and Systems. The series contains proceedings and edited volumes in systems and networks, spanning the areas of Cyber-Physical Systems, Autonomous Systems, Sensor Networks, Control Systems, Energy Systems, Automotive Systems, Biological Systems, Vehicular Networking and Connected Vehicles, Aerospace Systems, Automation, Manufacturing, Smart Grids, Nonlinear Systems, Power Systems, Robotics, Social Systems, Economic Systems and other. Of particular value to both the contributors and the readership are the short publication timeframe and the world-wide distribution and exposure which enable both a wide and rapid dissemination of research output. The series covers the theory, applications, and perspectives on the state of the art and future developments relevant to systems and networks, decision making, control, complex processes and related areas, as embedded in the fields of interdisciplinary and applied sciences, engineering, computer science, physics, economics, social, and life sciences, as well as the paradigms and methodologies behind them. Indexed by SCOPUS, INSPEC, WTI Frankfurt eG, zbMATH, SCImago. All books published in the series are submitted for consideration in Web of Science. For proposals from Asia please contact Aninda Bose ([email protected]).

I. Jeena Jacob · Selvanayaki Kolandapalayam Shanmugam · Robert Bestak Editors

Expert Clouds and Applications Proceedings of ICOECA 2022

Editors I. Jeena Jacob Department of Computer Science and Engineering GITAM University Bangalore, India

Selvanayaki Kolandapalayam Shanmugam Department of Mathematics and Computer Science Ashland University Ashland, OH, USA

Robert Bestak Department of Telecommunication Engineering Czech Technical University in Prague Prague, Czech Republic

ISSN 2367-3370 ISSN 2367-3389 (electronic) Lecture Notes in Networks and Systems ISBN 978-981-19-2499-6 ISBN 978-981-19-2500-9 (eBook) https://doi.org/10.1007/978-981-19-2500-9 © The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed. The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. The publisher, the authors, and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty, expressed or implied, with respect to the material contained herein or for any errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. This Springer imprint is published by the registered company Springer Nature Singapore Pte Ltd. The registered company address is: 152 Beach Road, #21-01/04 Gateway East, Singapore 189721, Singapore

We are honored to dedicate the proceedings of ICOECA 2022 to all the participants, organizers and editors of ICOECA 2022.

Foreword

RV Institute of Technology and Management was honored to host the Second International Conference on Expert Clouds and Applications (ICOECA 2022), which was held at Bangalore, India, from February 3 to 4, 2022. The main purpose of this conference series is to provide a research forum for establishing a communication between the researchers, academicians and industrialists/users of computing technologies and applications. It is equally gratifying that the first conference series of ICOECA has received an ample number of research submissions. The conference program has included a keynote session by two keynote speakers—Dr. Joy Iong-Zong Chen, Professor, Electrical Engineering, Dayeh University, Taiwan, and Dr. Archana Patel, Department of Software Engineering, School of Computing and Information Technology, Eastern International University, Vietnam. The conference event was attended by 57 participants and RV Institute of Technology and Management is so glad to acknowledge the conference organizing committee composed of faculty and non-faculty members and student volunteers for their tremendous and generous support in making it all happen. A special note of gratitude for reviewers, who have contributed much in behind the scenes for many months in the preparation of conference and in the final delivery of high-quality research papers in the conference proceedings. Dr. J. Anitha HOD/CSE RV Institute of Technology and Management Bengaluru, India

vii

Preface

It is a great privilege for us to present the proceedings of the First International Conference on Expert Clouds and Applications (ICOECA 2022) to the readers, delegates and authors of the conference event. We greatly hope that all the readers will find it more useful and resourceful for their future research endeavor. International Conference on Expert Clouds and Applications (ICOECA 2022) was held in Bangalore, India, from February 3 to 4, 2022, with an aim to provide a platform for researchers, academicians and industrialists to discuss the state-of-the-art research opportunities, challenges and issues in intelligent computing applications. The revolutionizing scope and rapid development of computing technologies will also create new research questions and challenges, which in turn results in the need to create and share new research ideas and encourage significant awareness in this futuristic research domain. The proceedings of ICOECA 2022 provides a limelight on creating a new research landscape for intelligent computing applications with the support received and the research enthusiasm that has truly exceeded our expectations. This made us to be more satisfied and delighted to present the proceedings with a high level of satisfaction. The responses from the researchers to the conference had been overwhelming from India and overseas countries. We have received 282 manuscripts from the prestigious universities/institutions across the globe. Furthermore, 57 manuscripts are shortlisted based on the reviewing outcomes and conference capacity constraints. Nevertheless, we would like to express our deep gratitude and commendation for the entire conference review team, who helped us to select the high-quality research works that are included in the ICOECA 2022 proceedings published by Springer. Also, we would like to extend our appreciation to organizing committee member for their continual support. We are pleased to thank Springer publications for publishing the proceedings of ICOECA 2022 and maximizing the popularity of the research manuscript across the globe.

ix

x

Preface

At last, we wish all the authors and participants of the conference event a best success in their future research endeavors. Bengaluru, India Ashland, USA Prague, Czech Republic

Dr. I. Jeena Jacob Dr. Selvanayaki Kolandapalayam Shanmugam Dr. Robert Bestak

Acknowledgments

We wish to express our gratitude and appreciation to our beloved President Dr. M. K. Panduranga Setty and dynamic Secretary A. V. S. Murthy for their constant encouragement and guidance to successfully organize the conference in this series. They have strongly encouraged us in conducting the conferences at RV Institute of Technology and Management Bangalore, India. Also, we like thank Principal Dr. R. Jayapal and all the board of directors for their perpetual support during the conduct of the ICOECA 2022 from which the conference proceedings book has evolved into existence. We like to thank all our review board members, who have assured the research novelty and quality from the initial to final selection phase of the conference. Also, we are thankful especially to our eminent speakers, reviewers and guest editors. Furthermore, we like to acknowledge all the session chairs of the conference event for the seamless contribution in evaluating the oral presentation of the conference participants. We would like to mention the hard work put up by the authors to revise and update their manuscripts according to the review comments to meet the conference and publication standards. We like to acknowledge the support of Springer publications for their constant and timely support throughout the publication process.

xi

Contents

An Approach for Summarizing Text Using Sentence Scoring with Key Optimizer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . G. Malarselvi and A. Pandian

1

Image Classification of Indian Rural Development Projects Using Transfer Learning and CNN . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Aditya Mangla, J. Briskilal, and D. Senthil Kumar

17

Design of a Mobile Application to Deal with Bullying . . . . . . . . . . . . . . . . . Vania Margarita, Agung A. Pramudji, Benaya Oktavianus, Randy Dwi, and Harco Leslie Hendric Spits Warnars Selection of Human Resources Prospective Student Using SAW and AHP Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Ahmad Rufai, Diana Teresia Spits Warnars, Harco Leslie Hendric Spits Warnars, and Antoine Doucet Implementation of the Weighted Product Method to Specify scholarship’s Receiver . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chintia Ananda, Diana Teresia Spits Warnars, and Harco Leslie Hendric Spits Warnars A Survey on E-Commerce Sentiment Analysis . . . . . . . . . . . . . . . . . . . . . . . . Astha Patel, Ankit Chauhan, and Madhuri Vaghasia Secured Cloud Computing for Medical Database Monitoring Using Machine Learning Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . M. Balamurugan, M. Kumaresan, V. Haripriya, S. Annamalai, and J. Bhuvana

31

43

65

81

91

Real-Time Big Data Analytics for Improving Sales in the Retail Industry via the Use of Internet of Things Beacons . . . . . . . . . . . . . . . . . . . 111 V. Arulkumar, S. Sridhar, G. Kalpana, and K. S. Guruprakash

xiii

xiv

Contents

Healthcare Application System with Cyber-Security Using Machine Learning Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127 C. Selvan, C. Jenifer Grace Giftlin, M. Aruna, and S. Sridhar Energy Efficient Data Accumulation Scheme Based on ABC Algorithm with Mobile Sink for IWSN . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143 S. Senthil Kumar, C. Naveeth Babu, B. Arthi, M. Aruna, and G. Charlyn Pushpa Latha IoT Based Automatic Medicine Reminder . . . . . . . . . . . . . . . . . . . . . . . . . . . 157 Ramya Srikanteswara, C. J. Rahul, Guru Sainath, M. H. Jaswanth, and Varun N. Sharma IoT Based Framework for the Detection of Face Mask and Temperature, to Restrict the Spread of Covids . . . . . . . . . . . . . . . . . . . 173 Ramya Srikanteswara, Anantashayana S. Hegde, K. Abhishek, R. Dilip Sai, and M. V. Gnanadeep Efficient Data Partitioning and Retrieval Using Modified ReDDE Technique . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187 Praveen M. Dhulavvagol, S. G. Totad, Nandan Bhandage, and Pradyumna Bilagi Stock Price Prediction Using Data Mining with Soft Computing Technique . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199 R. Suganya and S. Sathya A Complete Analysis of Communication Protocols Used in IoT . . . . . . . . 211 Priya Matta, Sanjeev Kukreti, and Sonal Sharma Image Captioning Using Deep Learning Model . . . . . . . . . . . . . . . . . . . . . . . 225 Disha Patel, Ankita Gandhi, and Zubin Bhaidasna Recommender System Using Knowledge Graph and Ontology: A Survey . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237 Warisahmed Bunglawala, Jaimeel Shah, and Darshna Parmar A Review of the Multistage Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253 Velia Nayelita Kurniadi, Vincenzo, Frandico Joan Nathanael, and Harco Leslie Hendric Spits Warnars Technology for Disabled with Smartphone Apps for Blind People . . . . . . 271 Hartato, Riandy Juan Albert Yoshua, Husein, Agelius Garetta, and Harco Leslie Hendric Spits Warnars Mobile Apps for Musician Community in Indonesia . . . . . . . . . . . . . . . . . . 283 Amadeus Darren Leander, Jeconiah Yohanes Jayani, and Harco Leslie Hendric Spits Warnars

Contents

xv

A Genetic-Based Virtual Machine Placement Algorithm for Cloud Datacenter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 301 C. Pandiselvi and S. Sivakumar Visual Attention-Based Optic Disc Detection System Using Machine Learning Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 317 A. Geetha Devi, N. Krishnamoorthy, Karim Ishtiaque Ahmed, Syed Imran Patel, Imran Khan, and Rabinarayan Satpathy An Overview of Blue Eye Constitution and Applications . . . . . . . . . . . . . . 327 Jadapalli Sreedhar, T. Anuradha, N. Mahesha, P. Bindu, M. Kathiravan, and Ibrahim Patel Enhancement of Smart Contact on Blockchain Security by Integrating Advanced Hashing Mechanism . . . . . . . . . . . . . . . . . . . . . . . 337 Bharat Kumar Aggarwal, Ankur Gupta, Deepak Goyal, Pankaj Gupta, Bijender Bansal, and Dheer Dhwaj Barak Evaluation of Covid-19 Ontologies Through OntoMetrics and OOPS! Tools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 351 Narayan C. Debnath, Archana Patel, Debarshi Mazumder, Phuc Nguyen Manh, and Ngoc Ha Minh Recognition of the Multioriented Text Based on Deep Learning . . . . . . . . 367 K. Priyadarsini, Senthil Kumar Janahan, S. Thirumal, P. Bindu, T. Ajith Bosco Raj, and Sankararao Majji Fruit and Leaf Disease Detection Based on Image Processing and Machine Learning Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 377 S. Naresh Kumar, Sankararao Majji, Tulasi Radhika Patnala, C. B. Jagadeesh, K. Ezhilarasan, and S. John Pimo A Survey for Determining Patterns in the Severity of COVID Patients Using Machine Learning Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . 385 Prachi Raol, Brijesh Vala, and Nitin Kumar Pandya Relation Extraction Between Entities on Textual News Data . . . . . . . . . . . 393 Saarthak Mehta, C. Sindhu, and C. Ajay A Comprehensive Survey on Compact MIMO Antenna Systems for Ultra Wideband Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 403 V. Baranidharan, S. Subash, R. Harshni, V. Akalya, T. Susmitha, S. Shubhashree, and V. Titiksha A Review of Blockchain Consensus Algorithm . . . . . . . . . . . . . . . . . . . . . . . . 415 Manas Borse, Parth Shendkar, Yash Undre, Atharva Mahadik, and Rachana Yogesh Patil A Short Systematic Survey on Precision Agriculture . . . . . . . . . . . . . . . . . . 427 S. Sakthipriya and R. Naresh

xvi

Contents

Improving System Performance Using Distributed Time Slot Assignment and Scheduling Algorithm in Wireless Mesh Networks . . . . 441 K. S. Mathad and M. M. Math Comparison of PRESENT and KLEIN Ciphers Using Block RAMs of FPGA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 453 S. Santhameena, Edwil Winston Fernandes, and Surabhi Puttaraju Remote Controlled Patrolling Robot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 467 Samarla Puneeth Vijay Krishna, Vritika Tuteja, Chatna Sai Hithesh, A. Rahul, and M. Ananda Novel Modeling of Efficient Data Deduplication for Effective Redundancy Management in Cloud Environment . . . . . . . . . . . . . . . . . . . . 479 G. Anil Kumar and C. P. Shantala A Survey on Patients Privacy Protection with Steganography and Visual Encryption . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 491 Hussein K. Alzubaidy, Dhiah Al-Shammary, and Mohammed Hamzah Abed Survey on Channel Estimation Schemes for mmWave Massive MIMO Systems – Future Directions and Challenges . . . . . . . . . . . . . . . . . . 505 V. Baranidharan, B. Moulieshwaran, V. Karthick, M. K. Munavvar Hasan, R. Deepak, and A. Venkatesh Fake News Detection: Fact or Cap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 517 C. Sindhu, Sachin Singh, and Govind Kumar The Future of Hiring Through Artificial Intelligence by Human Resource Managers in India . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 529 Ankita Arora, Vaibhav Aggarwal, and Adesh Doifode A Computer Vision Model for Detection of Water Pollutants Using Deep Learning Frameworks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 543 Anaya Bodas, Shubhankar Hardikar, Rujuta Sarlashkar, Atharva Joglekar, and Neeta Shirsat Medical IoT Data Analytics for Post-COVID Patient Monitoring . . . . . . . 555 Salka Rahman, Suraiya Parveen, and Shabir Ahmad Sofi SNAP—A Secured E Exchange Platform . . . . . . . . . . . . . . . . . . . . . . . . . . . . 569 Neeta Shirsat, Pradnya Kulkarni, Shubham Balkawade, and Akash Kasbe An AES-Based Efficient and Valid QR Code for Message Sharing Framework for Steganography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 581 Abhinav Agarwal and Sandeep Malik

Contents

xvii

Categorization of Cardiac Arrhythmia from ECG Waveform by Using Super Vector Regression Method . . . . . . . . . . . . . . . . . . . . . . . . . . . 599 S. T. Sanamdikar, N. M. Karajanagi, K. H. Kowdiki, and S. B. Kamble Performance Analysis of Classification Algorithm Using Stacking and Ensemble Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 615 Praveen M. Dhulavvagol, S. G. Totad, Ashwin Shirodkar, Amulya Hiremath, Apoorva Bansode, and J. Divya Conceptual Study of Prevalent Methods for Cyber-Attack Prediction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 631 S. P. Sharmila and Narendra S. Chaudhari Performance Analysis of Type-2 Diabetes Mellitus Prediction Using Machine Learning Algorithms: A Survey . . . . . . . . . . . . . . . . . . . . . . 643 B. Shamreen Ahamed, Meenakshi Sumeet Arya, and V. Auxilia Osvin Nancy Dynamic Updating of Signatures for Improving the Performance of IDS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 659 Asma Shaikh and Preeti Gupta M-mode Carotid Artery Image Classification and Risk Analysis Based on Machine Learning and Deep Learning Techniques . . . . . . . . . . . 675 P. Lakshmi Prabha, A. K. Jayanthy, and Kumar Janardanan Analyzing a Chess Engine Based on Alpha–Beta Pruning, Enhanced with Iterative Deepening . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 691 Aayush Parashar, Aayush Kumar Jha, and Manoj Kumar Intelligent Miniature Autonomous Vehicle . . . . . . . . . . . . . . . . . . . . . . . . . . . 701 Dileep Reddy Bolla, Siddharth Singh, and H. Sarojadevi Energy and Trust Efficient Cluster Head Selection in Wireless Sensor Networks Under Meta-Heuristic Model . . . . . . . . . . . . . . . . . . . . . . . 715 Kale Navnath Dattatraya and S Ananthakumaran Performance Analysis of OTFS Scheme for TDL and CDL 3GPP Channel Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 737 I. S. Akila, S. Uma, and P. S. Poojitha Design of Smart Super Market Assistance for the Visually Impaired People Using YOLO Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . 749 D. Jebakumar Immanuel, P. Poovizhi, F. Margret Sharmila, D. Selvapandian, Aby K. Thomas, and C. K. Shankar Investigating the Effectiveness of Zero-Rated Website on Students on E-Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 765 Asemahle Bridget Dyongo and Gardner Mwansa

xviii

Contents

Role of Machine Learning Algorithms on Alzheimer Disease Prediction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 779 V. Krishna Kumar, M. S. Geetha Devasena, G. Gopu, and N. Sivakumaran Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 791

Editors and Contributors

About the Editors I. Jeena Jacob is working as a Professor in Computer Science and Engineering department at GITAM University, Bengaluru, India. She actively participates on the development of the research field by conducting international conferences, workshops and seminars. She has published many articles in referred journals. She has guest edited an issue for International Journal of Mobile Learning and Organisation. Her research interests include mobile learning and computing. Selvanayaki Kolandapalayam Shanmugam is currently working as a Professor in the Department of Mathematics and Computer Science, Ashland University, Ashland, OH, 44805. She had overall 15+ years of Lecturing sessions for theoretical subjects, experimental and instructional procedures for laboratory subjects. She presented more research articles in national and international conferences and journals. Her research interest includes image processing, video processing, soft computing techniques, intelligent computing, Web application development, object-oriented programming like C++, Java, scripting languages like VBScript and JavaScript, data science, algorithms, data warehousing and data mining, neural networks, genetic algorithms, software engineering, software project management software quality assurance, enterprise resource planning, information systems, database management systems. Robert Bestak obtained Ph.D. degree in Computer Science from ENST Paris, France (2003), and M.Sc. degree in Telecommunications from Czech Technical University in Prague, CTU, Czech Republic (1999). Since 2004, he has been an Assistant Professor at Department of Telecommunication Engineering, Faculty of Electrical Engineering, CTU. He participated in several national, EU, and third-party research projects. He is the Czech representative in the IFIP TC6 organization and chair of working group TC6 WG6.8. He annually serves as Steering and Technical Program Committee Member of numerous IEEE/IFIP conferences (networking, WMNC,

xix

xx

Editors and Contributors

NGMAST, etc.), and he is Member of editorial board of several international journals (Computers and Electrical Engineering, Electronic Commerce Research Journal, etc.). His research interests include 5G networks, spectrum management and big data in mobile networks.

Contributors Abed Mohammed Hamzah College of Computer Science and Information Technology, Al-Qadisiyah University, Al-Dewaniyah, Iraq Abhishek K. Nitte Meenakshi Institute of Technology, Bengaluru, India Agarwal Abhinav Department of Computer Science, Singhania University, Jhunjhunu, India Aggarwal Bharat Kumar Deperment of Computer Science and Engineering, Vaish College of Engineering, Rohtak, Haryana, India Aggarwal Vaibhav O.P. Jindal Global University, Sonipat, India Ahmed Karim Ishtiaque Computer Science, Bahrain Training Institute, Higher Education Council, Ministry of Education, Manama, Bahrain Ajay C. Department of Computing Technologies, SRM Institute of Science and Technology Kattankulathur, Kattankulathur, India Akalya V. Department of Electronics and Communication Engineering, Bannari Amman Institute of Technology, Sathy, India Akila I. S. Department of ECE, Coimbatore Institute of Technology, Coimbatore, India Al-Shammary Dhiah College of Computer Science and Information Technology, Al-Qadisiyah University, Al-Dewaniyah, Iraq Alzubaidy Hussein K. College of Computer Science and Information Technology, Al-Qadisiyah University, Al-Dewaniyah, Iraq Ananda Chintia Information Systems Department, School of Information Systems, Bina Nusantara University, Jakarta, Indonesia Ananda M. Department of Electronics and Communication (UGC), PES University Electronic City Campus (UGC), Bangalore, India Ananthakumaran S School of CSE, VIT Bhopal University, Madhya Pradesh, India Anil Kumar G. Department of Computer Science and Engineering, Channabasaveshwara Institute of Technology Gubbi, Tumkur, India; Visvesvaraya Technological University, Belagavi, Karnataka, India

Editors and Contributors

xxi

Annamalai S. School of Computing Science and Engineering, Galgotias University, Greater Noida, India Anuradha T. Department of Electrical and Electronics Engineering, KCG College of Technology, Chennai, India Arora Ankita Apeejay School of Management, Delhi, India Arthi B. Department of Computer Science and Engineering, Faculty of Engineering and Technology, SRM Institute of Science and Technology, SRM Nagar, Kattankulathur, Kanchipuram, Chennai, Tamil Nadu, India Arulkumar V. School of Computer Science and Engineering, Vellore Institute of Technology, Vellore, Tamilnadu, India Aruna M. Department of Computer Science and Engineering, Faculty of Engineering and Technology, SRM Institute of Science and Technology, SRM Nagar, Kattankulathur, Kanchipuram, Chennai, Tamil Nadu, India Arya Meenakshi Sumeet SRM Institute of Science and Technology, Vadapalani Campus, Vadapalani, TN, India Auxilia Osvin Nancy V. SRM Institute of Science and Technology, Vadapalani Campus, Vadapalani, TN, India Balamurugan M. Department of Computer Science and Engineering, School of Engineering and Technology, Christ (Deemed to Be University), Bangalore, India Balkawade Shubham Department of Information Technology, PVG’s College of Engineering and Technology, Pune, India Bansal Bijender Deperment of Computer Science and Engineering, Vaish College of Engineering, Rohtak, Haryana, India Bansode Apoorva School of Computer Science and Engineering, KLE Technological University, Hubballi, India Barak Dheer Dhwaj Deperment of Computer Science and Engineering, Vaish College of Engineering, Rohtak, Haryana, India Baranidharan V. Department of Electronics and Communication Engineering, Bannari Amman Institute of Technology, Sathy, India Bhaidasna Zubin Parul Institute of Engineering and Technology, Vadodara, India Bhandage Nandan School of Computer Science and Engineering, KLE Technological University, Hubballi, India Bhuvana J. School of Computer Science and IT, Jain (Deemed to Be) University, Bangalore, India Bilagi Pradyumna School of Computer Science and Engineering, KLE Technological University, Hubballi, India

xxii

Editors and Contributors

Bindu P. Department of Mathematics, Koneru Lakshmaiah Education Foundation, Vaddeswaram, AP, India Bodas Anaya Department of Information Technology, PVG’s College of Engineering and Technology, Pune, India Bolla Dileep Reddy Department of CSE, Nitte Meenakshi Institute of Technology, Yehahanka, Bangalore, India Borse Manas Pimpri Chinchwad College of Engineering, Pune, India Briskilal J. SRM Institute of Science and Technology, Chennai, Tamil Nadu, India Bunglawala Warisahmed Parul Institute of Engineering and Technology, Vadodara, India Charlyn Pushpa Latha G. Department of IT, Saveetha School of Engineering, Saveetha Institute of Medical and Technical Sciences, Chennai, India Chaudhari Narendra S. Department of Computer Science and Engineering, Indian Institute of Technology, Indore, India Chauhan Ankit Parul Institute of Engineering and Technology, Vadodara, India Dattatraya Kale Navnath Department of Computer Science and Engineering, Koneru Lakshmaiah Education Foundation, Vaddeswaram, Andhra Pradesh, India Debnath Narayan C. Department of Software Engineering, Eastern International University, Binh Duong, Vietnam Deepak R. Department of Electronics and Communication Engineering, Bannari Amman Institute of Technology, Sathy, India Dhulavvagol Praveen M. School of Computer Science and Engineering, KLE Technological University, Hubballi, India Divya J. School of Computer Science and Engineering, KLE Technological University, Hubballi, India Doifode Adesh Institute for Future Education, Entrepreneurship and Leadership, Pune, India Doucet Antoine Laboratoire L3i, Université de La Rochelle, La Rochelle, France Dwi Randy Information Systems Department, School of Information Systems, Bina Nusantara University, Jakarta, Indonesia Dyongo Asemahle Bridget Walter Sisulu University, East London, South Africa Ezhilarasan K. Department of ECE, CMR University, Bangalore, Karnataka, India Fernandes Edwil Winston Department of Electronics and Communication Engineering, PES University, Bangalore, India Gandhi Ankita Parul Institute of Engineering and Technology, Vadodara, India

Editors and Contributors

xxiii

Garetta Agelius Information Systems Department, School of Information Systems, Bina Nusantara University, Jakarta, Indonesia Geetha Devasena M. S. Department of Computer Science and Engineering, Sri Ramakrishna Engineering College, Coimbatore, India Geetha Devi A. Department of ECE, PVP Siddhartha Institute of Technology, Vijayawada, Andhra Pradesh, India Gnanadeep M. V. Nitte Meenakshi Institute of Technology, Bengaluru, India Gopu G. Department of Electronics and Communication Engineering, Sri Ramakrishna Engineering College, Coimbatore, India Goyal Deepak Deperment of Computer Science and Engineering, Vaish College of Engineering, Rohtak, Haryana, India Gupta Ankur Deperment of Computer Science and Engineering, Vaish College of Engineering, Rohtak, Haryana, India Gupta Pankaj Deperment of Computer Science and Engineering, Vaish College of Engineering, Rohtak, Haryana, India Gupta Preeti Amity University Maharashtra, Mumbai, Maharashtra, India Guruprakash K. S. Department of Computer Science and Engineering, K. Ramakrishnan College of Engineering, Trichy, India Hardikar Shubhankar Department of Information Technology, PVG’s College of Engineering and Technology, Pune, India Haripriya V. School of Computer Science and IT, Jain (Deemed to Be) University, Bangalore, India Harshni R. Department of Electronics and Communication Engineering, Bannari Amman Institute of Technology, Sathy, India Hartato Information Systems Department, School of Information Systems, Bina Nusantara University, Jakarta, Indonesia Hasan M. K. Munavvar Department of Electronics and Communication Engineering, Bannari Amman Institute of Technology, Sathy, India Hegde Anantashayana S. Nitte Meenakshi Institute of Technology, Bengaluru, India Hiremath Amulya School of Computer Science and Engineering, KLE Technological University, Hubballi, India Hithesh Chatna Sai Department of Electronics and Communication (UGC), PES University Electronic City Campus (UGC), Bangalore, India Husein Information Systems Department, School of Information Systems, Bina Nusantara University, Jakarta, Indonesia

xxiv

Editors and Contributors

Jagadeesh C. B. New Horizon College of Engineering, Bangalore, India Janahan Senthil Kumar Department of CSE, Lovely Professional University, Phagwara, Punjab, India Janardanan Kumar Department of General Medicine, SRM Medical College Hospital and Research Centre, Chennai, India Jaswanth M. H. Nitte Meenakshi Institute of Technology, Bengaluru, India Jayani Jeconiah Yohanes Computer Science Department, School of Computer Science, Bina Nusantara University, Jakarta, Indonesia Jayanthy A. K. Department of Biomedical Engineering, SRM Institute of Science and Technology, Chennai, Tamil Nadu, India Jebakumar Immanuel D. Department of Computer Science and Engineering, SNS College of Engineering, Coimbatore, TamilNadu, India Jenifer Grace Giftlin C. Department of Information Technology, Sri Krishna College of Engineering and Technology, Coimbatore, TN, India Jha Aayush Kumar Delhi Technological University, Delhi, India Joglekar Atharva Department of Information Technology, PVG’s College of Engineering and Technology, Pune, India Kalpana G. Rajalakshmi Institute of Technology, Chennai, India Kamble S. B. Electronics Department, PDEA’s College of Engineering, Manjari, Pune, India Karajanagi N. M. Instrumentation Department, Government College of Engineering and Research, Awasari Khurd, Pune, India Karthick V. Department of Electronics and Communication Engineering, Bannari Amman Institute of Technology, Sathy, India Kasbe Akash Department of Information Technology, PVG’s College of Engineering and Technology, Pune, India Kathiravan M. Department of Computer Science and Engineering, Hindustan Institute of Technology and Science, Padur, Kelambaakkam, Chengalpattu, India Khan Imran Computer Science, Bahrain Training Institute, Higher Education Council, Ministry of Education, Manama, Bahrain Kowdiki K. H. Instrumentation Department, Government College of Engineering and Research, Awasari Khurd, Pune, India Krishna Kumar V. Department of Computer Science and Engineering, Sri Ramakrishna Engineering College, Coimbatore, India

Editors and Contributors

xxv

Krishna Samarla Puneeth Vijay Department of Electronics and Communication (UGC), PES University Electronic City Campus (UGC), Bangalore, India Krishnamoorthy N. MCA Department, SRM Institute of Science and Technology, Chennai, India Kukreti Sanjeev Department of Department of Computer Science and Engineering, Graphic Era University, Dehradun, India Kulkarni Pradnya Department of Information Technology, PVG’s College of Engineering and Technology, Pune, India Kumar Govind Department of Computing Technologies, School of Computing, SRM Institute of Science and Technology, Kattankulathur, India Kumar Manoj Delhi Technological University, Delhi, India Kumaresan M. School of Computer Science and Engineering, Jain (Deemed to Be) University, Bangalore, India Kurniadi Velia Nayelita Computer Science Department, School of Computer Science, Bina Nusantara University, Jakarta, Indonesia Leander Amadeus Darren Computer Science Department, School of Computer Science, Bina Nusantara University, Jakarta, Indonesia Mahadik Atharva Pimpri Chinchwad College of Engineering, Pune, India Mahesha N. Department of Civil Engineering, New Horizon College of Engineering, Bangalore, India Majji Sankararao Department of Electronics and Communication Engineering, GRIET, Hyderabad, India Malarselvi G. SRM Institute of Science and Technology, Chennai, India Malik Sandeep Department of Computer Science, Oriental University, Indore, India Mangla Aditya SRM Institute of Science and Technology, Chennai, Tamil Nadu, India Manh Phuc Nguyen Department of Software Engineering, Eastern International University, Binh Duong, Vietnam Margarita Vania Information Systems Department, School of Information Systems, Bina Nusantara University, Jakarta, Indonesia Margret Sharmila F. Department of Computer Science and Engineering, SNS College of Engineering, Coimbatore, TamilNadu, India Math M. M. KLS Gogte Institute of Technology, Belagavi, Karnataka, India Mathad K. S. KLS Gogte Institute of Technology, Belagavi, Karnataka, India

xxvi

Editors and Contributors

Matta Priya Department of Department of Computer Science and Engineering, Graphic Era University, Dehradun, India Mazumder Debarshi Department of Software Engineering, Eastern International University, Binh Duong, Vietnam Mehta Saarthak Department of Computing Technologies, SRM Institute of Science and Technology Kattankulathur, Kattankulathur, India Minh Ngoc Ha Department of Software Engineering, Eastern International University, Binh Duong, Vietnam Moulieshwaran B. Department of Electronics and Communication Engineering, Bannari Amman Institute of Technology, Sathy, India Mwansa Gardner Walter Sisulu University, East London, South Africa Naresh Kumar S. School of Computer Science and Artificial Intelligence, SR University, Waragal, Telangana, India Naresh R. Department of Computer Science and Engineering, SRM Institute of Science and Technology(SRMIST), Kattankulathur, Chennai, India Nathanael Frandico Joan Computer Science Department, School of Computer Science, Bina Nusantara University, Jakarta, Indonesia Naveeth Babu C. Department of Computer Science, Kristu Jayanti College, (Autonomous), Bangaluru, Karnataka, India Oktavianus Benaya Information Systems Department, School of Information Systems, Bina Nusantara University, Jakarta, Indonesia Pandian A. SRM Institute of Science and Technology, Chennai, India Pandiselvi C. Department of Computer Science, Cardamom Planters’ Association College, Bodinayakanur, India Pandya Nitin Kumar Parul Institute of Engineering and Technology, Vadodara, India Parashar Aayush Delhi Technological University, Delhi, India Parmar Darshna Parul Institute of Engineering and Technology, Vadodara, India Parveen Suraiya Department of Computer Science and Engineerin, Jamia Hamdard University New Delhi, Delhi, India Patel Archana Department of Software Engineering, Eastern International University, Binh Duong, Vietnam Patel Astha Parul Institute of Engineering and Technology, Vadodara, India Patel Disha Parul Institute of Engineering and Technology, Vadodara, India

Editors and Contributors

xxvii

Patel Ibrahim Department of ECE, B V Raju Institute of Technology, Narsapur, Medak, India Patel Syed Imran Computer Science, Bahrain Training Institute, Higher Education Council, Ministry of Education, Manama, Bahrain Patil Rachana Yogesh Pimpri Chinchwad College of Engineering, Pune, India Patnala Tulasi Radhika Department of Electronics and Communication Engineering, GITAM University, Hyderabad, India Pimo S. John St. Xavier’s Catholic College of Engineering, Nagercoil, India Poojitha P. S. Department of ECE, Coimbatore Institute of Technology, Coimbatore, India Poovizhi P. Department of Information Technology, Dr. N.G.P. Institute of Technology, Coimbatore, TamilNadu, India Prabha P. Lakshmi Department of Biomedical Engineering, SRM Institute of Science and Technology, Chennai, Tamil Nadu, India Pramudji Agung A. Information Systems Department, School of Information Systems, Bina Nusantara University, Jakarta, Indonesia Priyadarsini K. Department of Data Science and Business Systems, School of Computing, College of Engineering and Technology, SRM Institute of Science and Technology, Kattankulathur, Chennai, India Puttaraju Surabhi Department of Electronics and Communication Engineering, PES University, Bangalore, India Rahman Salka Department of Computer Science and Engineerin, Jamia Hamdard University New Delhi, Delhi, India Rahul A. Department of Electronics and Communication (UGC), PES University Electronic City Campus (UGC), Bangalore, India Rahul C. J. Nitte Meenakshi Institute of Technology, Bengaluru, India Raj T. Ajith Bosco Department of Electronics and Communication Engineering, PSN College of Engineering and Technology, Tirunelveli, Tamil Nadu, India Raol Prachi Parul Institute of Engineering and Technology, Vadodara, India Rufai Ahmad Information Systems Department, School of Information Systems, Bina Nusantara University, Jakarta, Indonesia Sai R. Dilip Nitte Meenakshi Institute of Technology, Bengaluru, India Sainath Guru Nitte Meenakshi Institute of Technology, Bengaluru, India Sakthipriya S. Department of Computer Science and Engineering, SRM Institute of Science and Technology(SRMIST), Kattankulathur, Chennai, India

xxviii

Editors and Contributors

Sanamdikar S. T. Instrumentation Department, PDEA’s College of Engineering, Manjari, Pune, India Santhameena S. Department of Electronics and Communication Engineering, PES University, Bangalore, India Sarlashkar Rujuta Department of Information Technology, PVG’s College of Engineering and Technology, Pune, India Sarojadevi H. Department of CSE, Nitte Meenakshi Institute of Technology, Yehahanka, Bangalore, India Sathya S. Department of Information Technology, School of Computing Sciences, Vels Institute of Science, Technology and Advanced Studies, (VISTAS), Chennai, Tamilnadu, India Satpathy Rabinarayan CSE (FET), Sri Sri University, Cuttack, Odisha, India Selvan C. Department of Computer Science & Engineering, New Horizon College of Engineering, Bangalore, Karnataka, India Selvapandian D. Department of Computer Science and Engineering, Karpagam Academy of Higher Education, Coimbatore, TamilNadu, India Senthil Kumar D. Sri Sairam Engineering College, Chennai, India Senthil Kumar S. Department of Computer Science, Sri Ramakrishna Mission Vidyalaya College of Arts and Science (Autonomous), Coimbatore, India Shah Jaimeel Parul Institute of Engineering and Technology, Vadodara, India Shaikh Asma Amity University Maharashtra, Mumbai, Maharashtra, India; Marathwada Mitra Mandal College of Engineering, Pune, India Shamreen Ahamed B. SRM Institute of Science and Technology, Vadapalani Campus, Vadapalani, TN, India Shankar C. K. Department of Electrical and Electronics Engineering, Sri Ramakrishna Polytechnic College, Coimbatore, TamilNadu, India Shantala C. P. Department of Computer Science and Engineering, Channabasaveshwara Institute of Technology Gubbi, Tumkur, India; Visvesvaraya Technological University, Belagavi, Karnataka, India Sharma Sonal Uttaranchal University, Dehradun, India Sharma Varun N. Nitte Meenakshi Institute of Technology, Bengaluru, India Sharmila S. P. Department of Information Science and Engineering, Siddaganga Institute of Technology, Tumakuru, India Shendkar Parth Pimpri Chinchwad College of Engineering, Pune, India

Editors and Contributors

xxix

Shirodkar Ashwin School of Computer Science and Engineering, KLE Technological University, Hubballi, India Shirsat Neeta Department of Information Technology, PVG’s College of Engineering and Technology, Pune, India Shubhashree S. Department of Electronics and Communication Engineering, Bannari Amman Institute of Technology, Sathy, India Sindhu C. Department of Computing Technologies, School of Computing, SRM Institute of Science and Technology, Kattankulathur, India Singh Sachin Department of Computing Technologies, School of Computing, SRM Institute of Science and Technology, Kattankulathur, India Singh Siddharth Department of CSE, Nitte Meenakshi Institute of Technology, Yehahanka, Bangalore, India Sivakumar S. Cardamom Planters’ Association College, Bodinayakanur, India Sivakumaran N. Instrumentation and Control Engineering, NIT, Trichy, India Sofi Shabir Ahmad Department of Information Technology, National Institute of Technology Srinagar, Jammu and Kashmir, India Sreedhar Jadapalli EEE Department, Vignana Bharathi Institute of Technology, Hyderabad, India Sridhar S. Department of Computer Science and Engineering, Saveetha School of Engineering, Saveetha Institute of Medical and Technical Sciences, Chennai, India Srikanteswara Ramya Nitte Meenakshi Institute of Technology, Bengaluru, India Subash S. Department of Electronics and Communication Engineering, Bannari Amman Institute of Technology, Sathy, India Suganya R. Research Scholar, Department of Information Technology, School of Computing Sciences, Vels Institute of Science, Technology and Advanced Studies (VISTAS), Chennai, Tamilnadu, India Susmitha T. Department of Electronics and Communication Engineering, Bannari Amman Institute of Technology, Sathy, India Thirumal S. Department of Computer Science and Engineering, Vels Institute of Science Technology and Advanced Studies, Chennai, India Thomas Aby K. Department of Electronics and Communication Engineering, Alliance College of Engineering and Design, Alliance University, Bangalore, India Titiksha V. Department of Electronics and Communication Engineering, Bannari Amman Institute of Technology, Sathy, India Totad S. G. School of Computer Science and Engineering, KLE Technological University, Hubballi, India

xxx

Editors and Contributors

Tuteja Vritika Department of Electronics and Communication (UGC), PES University Electronic City Campus (UGC), Bangalore, India Uma S. Department of ECE, Coimbatore Institute of Technology, Coimbatore, India Undre Yash Pimpri Chinchwad College of Engineering, Pune, India Vaghasia Madhuri Parul Institute of Engineering and Technology, Vadodara, India Vala Brijesh Parul Institute of Engineering and Technology, Vadodara, India Venkatesh A. Department of Electronics and Communication Engineering, Bannari Amman Institute of Technology, Sathy, India Vincenzo Computer Science Department, School of Computer Science, Bina Nusantara University, Jakarta, Indonesia Warnars Diana Teresia Spits Information Systems Department, School of Information Systems, Bina Nusantara University, Jakarta, Indonesia Warnars Harco Leslie Hendric Spits Computer Science Department, Binus Graduate Program-Doctor of Computer Science, Bina Nusantara University, Jakarta, Indonesia Yoshua Riandy Juan Albert Information Systems Department, School of Information Systems, Bina Nusantara University, Jakarta, Indonesia

An Approach for Summarizing Text Using Sentence Scoring with Key Optimizer G. Malarselvi and A. Pandian

Abstract There is an enormous amount of textual content rolled out over the web, which performs automatic text summarization efficiently. Specifically, extracting the multi-keywords from the textual content produces the summary from the source document by reducing the isolating text. In recent research, these summarization approaches and the problems related to this process are easily addressed with the optimization approaches. In existing research, most investigators concentrate on single-objective solutions; however, multi-objective approaches provide solutions to various issues during summarization. This work adopts a Keyword-based Elephant Yard Optimization (KEY) approach that improves the summarization process. In KEY, the analysis of the elephant movement is performed based on the group (cluster) of elephants. The significance of the movement relies on the priority given to the head. Accordingly, the textual contents are optimized based on the clustering priority. The analysis is performed over the available online datasets for text summarization to provide multiple solutions by handling multi-objective problems. Some of the predominant metrics like ROUGE-1 and ROUGE-2 score and kappa coefficient are evaluated to attain superior outcomes. Keywords Textual content · Text summarization · Keyword-based elephant yard optimization · Multi-objective problems · Multi-keywords

1 Introduction The growth of textual information is improved due to the expansion of internet applications and smartphones in our everyday routine. The explosion of generated data makes it impossible for humans to summarize it, and machines also struggle to manage the huge data from various applications, technologies, and firms [1]. The G. Malarselvi · A. Pandian (B) SRM Institute of Science and Technology, Chennai, India e-mail: [email protected] G. Malarselvi e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 I. J. Jacob et al. (eds.), Expert Clouds and Applications, Lecture Notes in Networks and Systems 444, https://doi.org/10.1007/978-981-19-2500-9_1

1

2

G. Malarselvi and A. Pandian

most challenging task is to analyze the huge amount of unstructured data, and also it is impossible to manage it. These massive information documents require automated document summarization to generate more concise documents to maintain important data [2]. Hence, the summarization technique is important to acquire awareness from data and to make decisions. Therefore, in today’s world, text summarization is one of the best methods. Many social media applications like Facebook and Twitter are now used for marketing and political and personal use. Today, social media is used for political campaigns to reach their respective supporters [3]. For outstanding marketing and political scenario, textual information that is extracted plays an essential role. To attain better marketing or political campaigns, the real-time applications of automatic text summarization are unrestricted. To attain the observation from data and ease of making the decision, summarization is used. Text summarization is used to get the shorter illustration of the search results over the search engines and in keywordbased subscriptions [4]. Based on user navigation between various contents [5], text summarization of social media facilitates the users’ trust effectively. The different categorization methods are available for the summarization of documents. Moreover, automatic summarization models are classified into two types. They are: (i) abstractive; and (ii) extractive. Abstractive technique recognizes the text deeply to exhibit the text deeply in a shorter alternate form. Contrarily, an extractive technique aims to choose the most substantial parts. Although the machine finds this method is difficult to generate, a summary, i.e., understandable by humans, mostly extractive techniques are utilized practically. In the proposed summarization task, the suggested technique eliminates the requirement of feature engineering. Since machine learning algorithms have the most important feature extraction phase, they mostly concentrate on sentence selection. For the summarization process, few efforts have been taken recently to identify an optimum feature set. This technique examines the importance of every feature as a binary problem that includes each feature in the feature set or not. The practical outcome provides that KEY can catch the basic design of features. A group of similar samples is found as the outcome and features are weighted locally in every group. These features weights explain the significance of every feature in sub-spaces (clusters). These spaces show not being or being in summary in the summarization issue. The major benefaction can be summarized as follows: • The KEY optimization acquires general performance. To illustrate this, we give some proof as experiments where the model is trained, and the discrimination takes place on the significance of all features of diverse classes on different datasets are informed. • This work executes human evaluation experiments to evaluate summaries from the human point of view, which provides ground truth. This experiment shows that the KEY approach’s summaries are unnecessary and descriptive than summaries generated by challenging techniques. • The KEY approach has an added benefit that is being more understandable to be a state-of-the-art performer. In the optimization process, the separated terms

An Approach for Summarizing Text Using Sentence …

3

permit to trace of the output summary. It is effective to describe decisions that are produced with the help of the system to the end-user.

2 Related Works To get the most needed sentences, most studies can compute the score for each sentence [6]. To evaluate the score for every sentence, there are two steps. They are (i) sentence scoring [7] and (ii) word scoring [8]. Based on statistical principles [9], these steps are summary generation models. The features should evaluate the score for each word, and also sentence score is the summation of scores of each word[10]. To evaluate the significance of words like word frequency [11] and TF [12]. Sentence position [13] and sentence length [14] are two essential elements in a sentence to compute the sentence’s score. Nengroo et al. [15] utilize the similarity of title and sentence to enhance the correctness of summary sentences, sentence length, and sentence position. Mikolov et al. [16] suggested hybrid and single-document text summarization has three methods. They are: (a) domain knowledge approach, (b) statistical features approach, and (c) genetic algorithms. The domain knowledge approach organizes the knowledge base domain corpus manually, and based on the domain keywords; this method summarizes the text. Statistical features method use heuristic methods to concentrate on ranking and segment extraction, which allocates weighted scores to text segmentation. The genetic algorithm-based method considers the automated text summary work as a classification issue. Categorization utilizes GA’s ML approach to generate a cohesive, readable, and better summary identical to the document title. The description of documents depends on the set of attributes. Single-text summary models are presented in this model [2]. A combination of three news features (person, time, and location) and a new text feature selection method are suggested with the news text features. A combination of fuzzy logic systems and the genetic algorithm is established. The genetic algorithm weights the text features. These features are turned twice to obtain more accurate summary sentences using the fuzzy logic system to produce a high-quality summary [17]. Table 1 depicts the comparison of various prevailing text summarization approaches.

3 Methodology The anticipated model is composed of three-pre-dominant stages to perform the summarization process. The redundant contents from the available document are filtered, and the successive stage is the optimization process. The weights of available document features are optimized to highlight the important sentence in the provided document. Finally, the text summary is presented by summarizing the basic sentences based on connectivity and similarity. Consider, a document set represented as doc =

4

G. Malarselvi and A. Pandian

Table 1 Comparison of various text summarization approaches Topic

Problems

Techniques

Methods

Extractive

Semantic Optimization Hybrid Clustering Topic modeling

Fuzzy-based statistics Machine learning Machine learning Graphical Machine learning Topic

Fuzzy hypergraph LSA and NMF LSA + ANN DL CGS DQN LDA

Abstractive

Extraction Ambiguity Clustering Sentence scoring

Machine learning

NAT-SUM TF-IDF NeuFuse Markov

Unsupervised learning

Optimization Noise Ambiguity Sematic

Machine learning

Bi-RNN Round robin and POS FROM UAE and Deep AE

Single document

Similarity Redundancy

Graphical Machine learning

NLP parser AE, NN

Multi-document

Extraction Redundancy Hybrid Clustering

Statistics Fuzzy-based Machine learning Machine learning

TEDU + COMP ANFISA ExpQuery RA-MDS

Optimization

Optimization Similarity Ambiguity

Machine learning Machine learning Machine learning

ABC and MOABC Shark smell optimization MMI diversity

Real-time

Similarity Redundancy Keywords Feature and real-time Semantic clustering

Fuzzy-based Machine learning Machine learning Statistics Machine learning

Fuzzy formal concept IncreSTS DTM RSE IS, NS IncreSTS

Domain

Extraction Similarity Clustering Keyword Feature Semantic analysis Redundancy Latent Selection

Fuzzy-based Statistics Fuzzy-based Machine learning Statistics Statistics Machine learning Statistics Statistics

Fuzzy AHP WordNet + common sub-summer CUBS Term frequency Sentence scoring LSA SVD MWI-sum

  {doc1 , doc2 , . . . , docn } with a sentence set doci = sen1 , sen2 , . . . , sen|doci | where |doci | specifies the total sentence over the document. The available document is provided for pre-processing steps like stop word removal, stemming, and sentence separation. As an outcome, the sentence is transformed into a word set   provided   S j = wd1 , wd2 , . . . , wd|sen j | where sen j specifies the complete word sentence. The redundancies over the provided sentence set are removed, and the word similarity

An Approach for Summarizing Text Using Sentence …

5

Fig. 1 Block of KEY model

is measured with the Keyword-based Elephant Yard Optimization (KEY). Here, the less similar sentence from one document is taken, and the document formation   Doc n = sen1 sen2 . . . sen L contains several sentences. It is expressed as L ≤ n = i=1 |doci |. The summarized text is provided with the sentence of length 1, i.e., C = 1, where C specifies compression value. The summarized text is extracted doc∗ 100 from the weighted features. The optimal weight is measured using the KEY concept. The flow diagram of the anticipated model is shown in Fig. 1

3.1 Data Acquisition The text summarization process begins with the selection of documents from online resources. The PubMed medical dataset is considered for coherent text segments beneficial for human comprehension and information retrieval. It is an open phrase for searching biomedical literature Word vectors (word2vec binary format), trained on all PubMed abstracts.

3.2 Essential Pre-processing Steps To initialize the process, the input is taken from the dataset. Some of the essential pre-processing steps are sentence separation, stemming, and stop word removal. (i)

Sentence separation: The provided document is partitioned into the sentence, and it assists in predicting the initialization of sentences.

6

(ii)

(iii)

G. Malarselvi and A. Pandian

Stemming: Here, the process maps the provided words into common words, and this work adopts Porter stemming algorithm. The process shows some deviation, i.e., negative or insignificant influence over the system related to semantic analysis. It is an optional process, with or without stems. Stop word removal: Here, the repetitive words over the document are removed, i.e., common words with no important information like prepositions, queries, helping verbs, conjunction, articles and so on. The text content is limited to the essential word summary.

3.3 Similarity Computation with Weighted Features The summary from the provided input dataset is merged where the redundant content over the document is filtered via maximal content coverage. The summaries are examined using semantic similarity measure among the sentences, and it is expressed as in Eq. (1):     sim s j , s j  =∝ ∗simweighted measure s j , s j 

  + (1− ∝) ∗ simnormalized weighted measure s j , s j 

(1)

The weighted parameter ∝∈ [0, 1] specifies the relationship among the sentence information from the weighted feature measure and the sentence provided is provided as in Eq. (2):   simweighted measure s j , s j      W p ∈s j W p ∈s  simweighted measure W p , W p j = s j . s j 

(2)

The sentence similarity is determined with the weighted measure, and the outcomes specify the summary execution. The objective is to reduce redundancy and maximize coverage. The summary should hold the essential sentence information to maximize the content coverage. The superior capacity measure is related to the special nearness of summary sentences with the actual content of the gathered document. Thus, it needs to be maximized. The summary has to be away from the sentence replication of the document set to reduce the redundancy. It is measured as the non-redundant where the similarity among two sentences relies on the provided threshold T ∈ (0, 1). At last, the difference among the non-redundant sentences is provided as in Eq. (3):   sim s j , s j ≤ T

(3)

An Approach for Summarizing Text Using Sentence …

7

3.4 Feature Weight Scoring Here, the feature chosen for text summarization is sentence similarity, sentence position, numerical data, title word, proper noun, numerical data, proper noun, frequent word, sentence significance, and sentence length. Consider, T f x specifies xth text feature, s j specifies sentence set, ST f x specifies text feature score for every sentence. The text feature characteristics are discussed below: (i)

Sentence position: It is depicted as the essential feature for phrase extraction. It is expressed as in Eq. (4): i   doc − ST f 1 S j 2 doci 2

(ii)

j

(4)

Sentence length (T f 2 ): It is considered to eliminate longer or shorter sentences, and it is expressed as in Eq. (5), and the average distance sentences are expressed as in Eq. (6): AL(s) − s j     ST f 2 S j = 1 − max s j     min s j + max s j Al(S) = 2

(iii)

(6)

Title words (T f 3 ): It is known as the essential phrases from the available summary. The sentences are composed of words and represent the essential content document. These are known as the baseline of the document, and it is provided with a high score. It is expressed as in Eq. (7):   ST f 3 S j =

(iv)

(5)





 W p ∈s j

W p ∈tw

W p ∩ W p

 2

S j .|tw|

Here, |tw| specifies the total amount of title words. Numerical data (T f 4 ): It has some numerical data that specifies the important information from the summary, and it is expressed as in Eq. (8):   ST f 4 s j = number of available numerical data in s j s j

(v)

(7)

(8)

Proper noun: It includes organization, place, or person in the provided sentences is determined more weight, and it is evaluated as in Eq. (9):   number of proper nouns in s j ST f 5 s j = s j

(9)

8

G. Malarselvi and A. Pandian

3.5 Sentence Scoring with KEY Optimizer Generally, elephants are measured as extroverted creatures with a huge population (multi-document), and it is partitioned into various clans. Every individual lives within the clan under the female matriarch leadership, and when the individual reaches adulthood, some number of males leave the clan it belongs to. It has some characteristics by partitioning the searching agent behavior known as updation and separation operator. (i)

Updation operator: The individual elephant position  j over the clan Ci is expressed as in Eq. (10):   xnew,ci , j = xci, j + α ∗ xbest,ci − xci, j ∗ r

(ii)

Here, xnew,ci , j and xci, j specifies the original and new position of every individual j over clan Ci , α and r specify both random numbers within [0, 1]. Separation operator: It is provided to update the worst individual over the clan, and it is expressed in Eq. (11): xworst,ci = xmin + (xmax − xmin + 1) ∗ r

(iii)



| log(r )|

(12)

Cyclone-based foraging strategy: It performs swin forwarding in spiral pattern and every individual approaches its prey and adjusts the current position, prior best agents, and maintains swim forwarding. The weighted coefficient β is expressed as in Eq. (13):

β = 2e

r1

(v)

(11)

Here, xmax and xmin specify the upper and lower bounds of every clan’s position, r specifies the random number of individuals [0, 1]. The novelty of the work relies on the foraging strategies with initialization and agent updation via iteration sequentially to predict the superior (best) solution. Chain-based foraging strategy: This chain process is performed sequentially from the up-to-down manner. The optimal solution updates the agent position, and the attained solutions are produced during prior iterations. The weighted coefficients ∝ expressed as in Eq. (12): ∝= 2 ∗ r ∗

(iv)

(10)

T −t +1 T

 ∗ sin(2πr1 )

(13)

Here, r1 specifies the random number r1 ∈ [0, 1], T specifies the maximal amount of iterations, the update position by searching the random position to improve exploration abilities. Flip-flop foraging strategies: This strategy considers the food as the central point (hub). Every agent performs flipping motion over the central point.

An Approach for Summarizing Text Using Sentence …

9

The agent updates its position over the present optimal solution to improve the exploitation. Algorithm 1 shows the explanation of Keyword-based optimization. Here, S specifies the flipping factor and specifies the ranges of the flip-flop motion. Here, the parameter is set as one where r2 and r3 specify the random number [0, 1]. Algorithm 1 KEY optimizer Input: Initialize population size N , maximal no. of generations tmax , upper and lower bounds xmax and xmin ; Output: Best optimal solutions xbest ; 1. Initialize parameters like S, α, β; 2. Compute fitness of every agent based on the fitness value; 3. While t < tmax do; 4. if r and < 0.5 and

t tmax

, then

5. Update xi ; //as in Eq. (18); 6. else if 7. Update xi ; //as in Eq. (19); 8. end if 9. Compute fitness value based on updation strategy; 10. for i = 1 : N , do 11. Update the position; 12. if f |xt (t + 1)| < f (xbest ) then 13. Substitute xbest with xi (t + 1) 14. end if 15. end for 16. Compute fitness value based on position updaiton; 17. Sort new population-based on every updation; 18. t = t + 1; 19. end while 20. return xbest ;

3.6 Summarization Process Finally, text summarization is performed to extract the summarized outcomes. It provides optimal outcomes with the weighted coefficients of β. The weighted factors of the summarized text and the sentence score are based on the individual text summary. It is expressed as in Eq. (14):

10

G. Malarselvi and A. Pandian dim       Score s j = aq x ∗ Summarized text s j ;

(14)

q=0,x=1

At last, the highly scored sentences are extracted from the provided document and considered for the summarization process. It is performed based on the generated similarities, and it possesses a significant position over the main sentence. The deleted sentence is like the first sentence, the separated sentence looks like of successive sentence and so on.

4 Numerical Results and Discussion The simulation is done in MATLAB 2016b environment, and the experimental settings and the metrics analysis are performed with the PubMed medical dataset. Here, 70% of data is used for training and 30% is used for testing purposes. The ROUGE toolbox is used for the computation of estimation scores, and the efficacy is well-formed. One word is evaluated with ROUGE-1 to fix with system and referral summarization. ROUGE-2 evaluates the successive summarization to check with the baseline summarization. Table 2 depicts the ROUGE-1 and ROUGE-2 score computations. Some metrics like precision, recall, and F1-score is evaluated for various methods like Secure Squirrel Optimizer (SSO), Particle Swarm Optimization (PSO), and Elephant Herd Optimization (EHO). In Table 2, the similarity measure of KEY is 0.325, SSO is 0.305, and EHO is 0.312, respectively. This analysis proves that the model works more effectively than the other approaches like SSO and EHO models. Here, three different metrics are observed with precision (P), recall (R), and F1score (F). The evaluation is based on the summarization process where P specifies how the summary information is nearer to reference summary; R specifies the amount of information during system summaries nearer to reference summary; F specifies similarity among reference and system offering weights equal to P and R score. It is expressed as in Eq. (15) F=

2∗P∗R P+R

(15)

Table 2 ROUGE-1 and ROUGE-2 computation Techniques

ROUGE-1

ROUGE-2

Summarization process

P

R

F1

P

R

F1

Similarity measure

SSO

0.345

0.125

0.195

0.155

0.065

0.085

0.305

EHO

0.352

0.128

0.202

0.161

0.070

0.092

0.312

KEY

0.364

0.130

0.210

0.168

0.085

0.095

0.325

An Approach for Summarizing Text Using Sentence …

11

Fig. 2 Similarity measure computation of KEY with SSO and EHO

Table 3 Similarity measure computation Techniques

ROUGE-1

ROUGE-2

Summarization process

P

R

F1

P

R

F1

Similarity measure

PSO

0.330

0.127

0.183

0.186

0.039

0.064

0.250

SSO

0.347

0.146

0.205

0.165

0.062

0.089

0.235

EHO

0.350

0.148

0.208

0.167

0.063

0.090

0.240

KEY

0.364

0.130

0.210

0.168

0.085

0.095

0.325

Table 3 depicts the similarity measure computation of the anticipated model with various existing approaches like PSO, SSO, and EHO. The similarity measure of KEY is 0.325, PSO is 0.250, SSO is 0.235, and EHO is 0.240, which shows that the KEY model is superior to other approaches. The performance of the anticipated model is superior to other approaches like PSO, SSO, and EHO. The proposed and the existing approaches are provided in a readable manner (See Figs. 3, 4, 5 and 6) (Table 4). The kappa coefficient of this process is expressed as in Eq. (16): Kappa coefficient =

p(c) − p(s) 1 − p(s)

(16)

Here, p(c) specifies the proportion of time concurred, and p(s) specifies the proportion of time agreed upon by a certain coincidence. The evaluation of the kappa coefficient over an interval [2/3, 1] is provided acceptably.

12

G. Malarselvi and A. Pandian

Fig. 3 ROUGE-1 computation of KEY with others

Fig. 4 ROUGE-2 computation of KEY with others

5 Conclusion This research concentrates on modeling and efficient Keyword-based Elephant Yard Optimization (KEY) approaches for the automatic text summarization process. This model performs three preliminary phases: redundancy elimination contents by filtering, text features optimization, and text summarization based on keyword connectivity and similarity. Here, Keyword-based Elephant Yard Optimization (KEY) optimizes weights of the enormous text functionality, and this optimizer intends to predict the proper sentences over a document. The summarization process

An Approach for Summarizing Text Using Sentence …

13

Fig. 5 Similarity measure of KEY with others

Fig. 6 Kappa coefficient computation of KEY with others

Table 4 Kappa coefficient evaluation

Techniques

Judgment mode

Kappa coefficient

PSO

Readable

0.723

SO

Readable

0.690

WHO

Readable

0.802

KEY

Readable

0.990

14

G. Malarselvi and A. Pandian

is generated by summarizing appropriate sentences based on certain similarities. The anticipated model is evaluated with various existing approaches like PSO, SSO, and EHO using PubMed dataset. The experimental outcomes revealed that the anticipated model provides superior performance to the general approaches. The performances of the summaries are measured with metrics like recall, precision, and F1-score over PubMed dataset. Some features like redundancy and coverage give better performance than other approaches. The major research constraints are the selection of a dataset for the summarization process; however, in the future, the similarity methods are examined with the deep network concept to attain superior generalization over the comprehensible summaries.

References 1. X. Xia, D. Lo, E. Shihab, X. Wang, Automated bug report field reassignment and refinement prediction. IEEE Trans. Rel. 65(3), 1094–1113 (2016) 2. A.M. Rush, S. Chopra, J. Weston, A neural attention model for abstractive sentence summarization, in Proceedings of Conference Empirical Methods Natural Language Processing (2015), pp. 379–389 3. H. Jiang, N. Nazar, J. Zhang, T. Zhang, Z. Ren, PRST: a pagerankbased summarization technique for summarizing bug reports with duplicates. Int. J. Softw. Eng. Knowl. Eng. 27(6), 869–896 (2017) 4. H. Jiang, X. Li, Z. Ren, J. Xuan, Z. Jin, Toward better summarizing bug reports with crowdsourcing elicited attributes. IEEE Trans. Rel. 68(1), 2–22 (2019) 5. R. Nithya, A. Arunkumar, Summarization of bug reports using feature extraction. Int. J. Comput. Sci. Mob. Comput. 52(2), 268–273 (2016) 6. E. Vázquez, R.A. García-Hernández, Y. Ledeneva, Sentence features relevance for extractive text summarization using genetic algorithms. J. Intell. Fuzzy Syst. 35(1), 353–365 (2018) 7. S. Charitha, N.B. Chittaragi, S.G. Koolagudi, Extractive document summarization using a supervised learning approach, in Proceedings of IEEE Distributed Computing, VLSI, Electrical Circuits and Robotics (DISCOVER) (2018), pp. 1–6 8. E. Cardinaels, S. Hollander, B.J. White, Automatic summarization of earnings releases: attributes and effects on investors’ judgments. Rev. Accounting Stud. 24(3), 860–890 (2019) 9. M. Afsharizadeh, H. Ebrahimpour-Komleh, A. Bagheri, Queryoriented text summarization ˙ using sentence extraction technique, in Proceedings of 4th International Conference on Web Research (ICWR) (2018), pp. 128–132 10. S. Narayan, S.B. Cohen, M. Lapata, Ranking sentences for extractive summarization with reinforcement learning (2018). arXiv:1802.08636. (Online). Available: http://arxiv.org/abs/1802. 08636 11. S. Chopra, M. Auli, A.M. Rush, Abstractive sentence summarization with attentive recurrent neural networks, in Proceedings of the NAACL-HLT (San Diego, CA, USA, 2016), pp. 93–98 12. R. Nallapati, B. Zhou, C.N. dos Santos, C. Gulcehre, B. Xiang, Abstractive text summarization using sequence-to-sequence RNNs and beyond,in Proceedings of the EMNLP (2016), pp. 1–12 13. Z. Lin, M. Feng, C.N. dos Santos, M. Yu, B. Xiang, B. Zhou, Y. Bengio, A structured selfattentive sentence embedding (2017), pp. 1–15. arXiv:1703.03130. (Online). Available: https:// arxiv.org/abs/1703.03130 14. T. Shen, T. Zhou, G. Long, J. Jiang, S. Pan, C. Zhang, Disan: directional self-attention network for RNN/CNN-free language understanding (2017), pp. 1–11. arXiv:1709.04696. (Online). Available: https://arxiv.org/abs/1709.04696

An Approach for Summarizing Text Using Sentence …

15

15. A. Shaqoor Nengroo, K.S. Kuppusamy, Machine learning-based heterogeneous web advertisements detection using a diverse feature set. Future Gener. Comput. Syst. 89, 68–77 (2018) 16. T. Mikolov, I. Sutskever, K. Chen, G. Corrado, J. Dean, Distributed representations of words and phrases and their compositionality (2013). arXiv:1310.4546. (Online). Available: http:// arxiv.org/abs/1310.4546 17. Z. Tu, Z. Lu, Y. Liu, X. Liu, H. Li, Modeling coverage for neural machine translation, in Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (2016), pp. 76–85 18. A. Sinha, A. Yadav, A. Gahlot, Extractive text summarization using neural networks (2018). arXiv:1802.10137. (Online). Available: https://arxiv.org/abs/1802.10137 19. C. Yadav, A. Sharan, A new LSA and entropy-based approach for automatic text document summarization. Int. J. Semantic Web Inf. Syst. 14(4), 1–32 (2018) 20. R. Rattray, R.C. Balabantaray, Cat swarm optimization based evolutionary framework for multi-document summarization. Phys. A Stat. Mech. Appl. 477, 174–186 (2017) 21. W. Song, L.C. Choi, S.C. Park, X.F. Ding, Fuzzy evolutionary optimization modelling and its applications to unsupervised categorization and extractive summarization. Expert Syst. Appl. 38(8), 9112–9121 (2011) 22. A. Sungheetha, R. Sharma, Transcapsule model for sentiment classification. J. Artif. Intell. 2(03), 163–169 (2020) 23. S. Smys, J.I.Z. Chen, S. Shakya, Survey on neural network architectures with deep learning. J. Soft Comput. Paradigm (JSCP) 2(03), 186–194 (2020) 24. J.S. Manoharan, J. Samuel, Capsule network algorithm for performance optimization of text classification. J. Soft Comput. Paradigm (JSCP) 3(01), 1–9 (2021) 25. J.S. Raj, J. Vijitha Ananthi, Recurrent neural networks and nonlinear prediction in support vector machines. J. Soft Comput. Paradigm (JSCP) 1(01), 33–40 (2019) 26. S.R. Mugunthan, T. Vijayakumar, Design of improved version of sigmoidal function with biases for classification task in ELM domain. J. Soft Comput. Paradigm (JSCP) 3(02), 70–82 (2021)

Image Classification of Indian Rural Development Projects Using Transfer Learning and CNN Aditya Mangla, J. Briskilal, and D. Senthil Kumar

Abstract In recent years, the convolutional neural network has been demonstrated to be effective in classifying data from animals to objects and as well as the human hand signs. Convolutional Neural Network (CNN) shows high performance on image classification. Recent trends in CNN that are used extensively are transfer learning and data augmentation. In this paper, the classification of Indian Government rural projects such as check dams, farm ponds, soak pits, etc. that promote agricultural activities are classified based on image. These projects built in the rural parts of India are of similar features and hence a challenging task to classify them. Remote-Sensing (RS) model has been proposed for the classification of these projects which is further compared with DenseNet-121 model over the same task on the basis of different data sizes and number of layers. Moreover, checking their influence on classifications of these images has been performed. Using this proposed RS model, a test accuracy of 0.9150 has been achieved. Keywords Convolutional neural network · Deep learning · Image augmentation · DenseNet121 · Image classification · Transfer learning

1 Introduction The research and development of remote-sensing images are of significant importance in many areas. Remote-sensing agriculture-related projects lead to precision agriculture and rural planning. Since the past decade, data available for A. Mangla · J. Briskilal (B) SRM Institute of Science and Technology, Potheri, Kattankulathur, Chengalpattu District, Chennai, Tamil Nadu 603203, India e-mail: [email protected] A. Mangla e-mail: [email protected] D. Senthil Kumar Sri Sairam Engineering College, Chennai, India e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 I. J. Jacob et al. (eds.), Expert Clouds and Applications, Lecture Notes in Networks and Systems 444, https://doi.org/10.1007/978-981-19-2500-9_2

17

18

A. Mangla et al.

remote-sensing research has increased drastically which leads to ease of training convolutional neural networks model to achieve decent accuracy. Convolutional Neural Network has shown great levels of efficiency and accuracy when it comes to image classification tasks and is breaking boundaries day after day. From the work of Razavian et al. [1], it is determined that along with CNN, Transfer learning has also gained huge significance due to the reduced amount of training time and sufficient dataset required to achieve decent results [2]. Rather than training model with the random initialization of weights, if weights that are trained on the network architecture have been used for the same or another dataset, and used that as pre-training task and then transfer to a new task, the proposed model will progress faster in a lesser time [3].

1.1 Dataset The dataset used in this research is acquired from ISRO’s Geoportal- Bhuvan. 6 k images for training for six classes, 1.5 k images for validation with 450 × 450 × 3 input size, and 1250 pictures per class for six different classes are obtained. To experiment with effectiveness of augmentation techniques, a restricted data of 550 images per class for six classes are also trained, with and without DenseNet. The projects constructed under Mahatma Gandhi National Rural Employment Act include cement roads in rural villages, check dams, farm ponds, soak pits, horticulture for agricultural activities, Latrines for sanitization, and even more. The images of all these projects are maintained by ISRO’s Geoportal—Bhuvan. The data is initially categorized according to the regions and then according to the projects built in that region. Intentionally, decrement of the size of the dataset is performed, in order to apply augmentation to the images and evaluate how better it works with less dataset plus data augmentation, and with larger dataset plus without augmentation. While there are some advanced data augmentation techniques such as GANs (Generative Adversial Networks), as given by Luis Perez et al. [4], traditional data augmentation transformations are used in this work. The model used to experiment with is “DenseNet”, given by Liu et al. [5]. By Zhang et al. [6], Transfer learning is widely used for COVID-19 diagnosis, since an already trained model on chest x-rays tends to perform better on covid x-rays than a model from scratch. Whereas in some cases such as COVID-Net, it has been proved to perform better on chest X-ray images than transfer learning models. Here, RS model which outperforms the transfer learning model DenseNet-121 is proposed.

Image Classification of Indian Rural Development …

19

2 Background Object recognition with convolutional neural network was first proposed by LeCunn et al. [7], in which, how learning the right features can produce a better accuracy and good performance has been elaborated. Convolutional network’s architecture showed, how a pixel of an image can be mapped to feature, which helps in prediction [8]. One of the state-of-the-art models is DenseNet-121 proposed by Huang et al. [9]. The work done in past decade has shown that convolutional neural networks can be denser along with being more accurate and efficient. In vanilla convolutional neural network, the input image is passed through the neurons of the neural network to get the result as predicted, where the forward pass is really direct. However, DenseNet engineering is tied in with changing this standard CNN design. In a DenseNet engineering, a five thick layer block that has a development pace of k = 4 has been used [10]. DenseNet design is in such a way that each layer is related to all the other single layers, as per the name.

3 Literature Review This literature review examines the works related to image classification of rural projects using remote-sensing data or image datasets. Additionally, it illustrates how data augmentation was performed in traditional ways, along with transfer learning that provides decent results even on a small dataset.

3.1 Works Related to Image Classification of Rural Projects The task of image classification of rural projects is used for village formation which leads to better rural planning and resource allocation. Previous research by Singh et al. [11] proposed the formation of town data arrangement of a Moga area in Punjab, utilizing Geoinformatics. The work focused on Land Use which included Built Up, Pond, River, Wasteland, Plantation, etc. information of the Moga district was extracted from the Satellite Imagery with the help of supervised classification technique. Support Vector Machine was used in order to achieve the former on the IRS-P6 LISS-III data acquired in 2007 and used information of visually interpreted land use of 2002 to generate the set for the classifier. The disaster-prone areas and amenities required for its management were mapped along with the identification of poor infrastructure. Al-doski et al. [12] looked into the image classification process and procedures, and image classification techniques which are K-means Classifier and Support Vector

20

A. Mangla et al.

Machine, respectively. The supervised and unsupervised methods were used to obtain accuracy over the dataset. The multiple remote-sensing features, including spectral, multitemporal, and spatial were utilized. Varshney et al. [13] proposed a model to recognize metal rooftops from covered rooftops based on the fact that the metal rooftops are dazzling white or dark, whereas the covered rooftops are caramel with less fresh and somewhat less straight edges. Arbitrary woods relapses were prepared for the absolute number of rooftops in the picture fix. The satellite symbolism and the prepared model were used on that. An outright blunder of 1.95 on the quantity of rooftops and 0.162 on the extent of metal was accomplished. The above research mainly used Support Vector Machine on the satellite data to classify images. However, a convolutional neural network model can provide much more efficiency than an SVM is argued here since it captures different features of an image and applies different filters to the image other than making prediction based on distance from the hyperplane as in SVM. In this paper, RS model which is able to extract features at a much better level, along with strengthening feature propagation has been proposed. In the model, feature propagation is dynamic which benefits in seamless flow of information. Furthermore, computationally it requires lesser resources compared to ResNet [14].

3.2 Works Related to Data Augmentation Techniques Wang et al. [4] explored various data augmentation solutions in image classification. The information was compelled falsely to a little subset of the ImageNet dataset and considered the execution of every information expansion procedure. The tinyimagenet-200 information and MNIST were utilized. Initially, customary changes were applied on the dataset, for example, producing a copy picture that is zoomed in/out, turned, flipped, mutilated, and so on. For the N size dataset, a 2 N size dataset was used. Further, Generative Adversial Networks was applied for style change utilizing six distinct styles: Cezanne, Enhance, Monet, Ukiyo-e, Van Gogh, and Winter. To test the viability of different expansions, 10 investigations were run on picture net information. On characterizing canines versus goldfish, the customary performed better compared to the GANs. However, the best result was given by Neural+. Yet, if there should arise an occurrence of Dogs versus Cat, customary gave 0.775 approval exactness which was the most elevated. Traditional transformations method proved to be more useful in data augmentation than GANs. Geometric Transformation, flipping, color space, cropping, etc., prove to fill up the gap for the less number of images. Many such traditional transformations were explored by Shorten et al. [15]. A baseline accuracy of 48.13 was increased to 61.95 using different augmentation techniques. Baseline results from Wong et al. [16] show that CNN performed better from additional training samples, as it improves both training and validation error. Applying

Image Classification of Indian Rural Development …

21

data augmentation, provides the RS model, images with different transformations such as distortion and zoom which helps the model to learn better even with less number of images along with those additional training samples.

3.3 Gaps Identified in the Related Works In the above-related work, the major gap found is the use of traditional machine learning methods instead of advanced machine learning methods. Above researchers used SVM and K-means which are traditional machine learning methods. At present, there are much more advanced computer vision models available such as deep learning models, convolutional neural networks, etc. Furthermore, data augmentation was not used and this leads to low accuracy as if the same image is flipped or zoomed. Therefore, the model might fail to recognize the image. Moreover, the model was trained from scratch instead of using a pre-trained model, which leads to uses of a lot of computational power and is not efficient due to computational power usage. The above issues result in longer training time, inefficient use of computational resources, and lack of producing good efficiency.

3.4 Contributions of This Paper To overcome the above-mentioned problems, use of convolutional neural network to train the images, which leads to learn useful features instead of random numbers, is proposed. Along with this, it helps to train in less time. Dataset has been created from the beginning. The use of latest techniques such as transfer learning and data augmentation allows the model to be robust again with no cropped or blurred images or any other altered image. Moreover, with the use of transfer learning methods, it is not required to train the model from the start, instead previously trained model can be used on some other dataset for the classification task and can be trained further on the right dataset.

4 Proposed Work The classification of Indian Government rural projects to promote agricultural activities, based on image is proposed. These projects are built in the rural places with similar features. Thus, classifying them is difficult. In particular, at first training the convolutional neural network model to perform a rudimentary classification is done [17]. After which it is followed by data augmentation along with transfer learning. Then, DenseNet121 is used to pretrain weights to further train on the model. Later, exploration and proposal of image classification model to classify agricultural and

22

A. Mangla et al.

Fig. 1 Conceptual and architectural design

rural development projects for precision agriculture and rural planning are performed. Finally, training with RS model to out-bit DenseNet121 model accuracy is done. Below is the architectural diagram (Fig. 1). Experiments are implemented by RS model on the basis of different data sizes and number of layers. Constraining dataset to less number of images in order to apply image augmentation to these images and optimize the RS model accordingly are executed.

4.1 Experimental Setup and Analysis 4.1.1

Dataset Creation

The dataset is created by web-scrapping the images from the Bhuvan Indian Geo Platform of ISRO. The images of six classes to be classified are obtained from the portal. 1400–1500 images were downloaded for each class. After dataset cleaning, finally 1000 images for each class of different dimensions, all greater than 500 × 500 were gathered. In total, 6000 images for six classes, i.e., Cement roads in rural

Image Classification of Indian Rural Development …

23

areas, Check Dams in rural areas, Farm Ponds in rural areas, Soak pits in rural areas, Indian Household Latrines in rural areas, and Horticulture land in rural areas, taken from the state of Tamil Nadu, a southern part of India, were collected.

4.1.2

Dataset Validation and Evaluation

For the dataset validation, Cohen’s kappa score is used to get the interrelated agreement and reliability of the dataset. It is a quantifiable coefficient that tends to level the precision and dependability in a verifiable game plan. • • • •

A=> Both raters agree to include B=> Only the first rater wants to include C=> Only the second rater wants to include D=> Both raters agree to exclude

To work out the kappa esteem, there is a need to know the likelihood of the arrangement. This recipe is determined by adding the quantity of tests where the appointed authorities concur, then separating it by the complete number of tests. Utilizing the model would mean (A + D)/ (A + B + C + D). Po = Number in Agreement/Total P(correct) = (A + B/A + B + C + D) ∗ (A + C/A + B + C + D) P(incorrect) = (C + D/ A + B + C + D) ∗ (B + D/A + B + C + D) Pe = P(correct) + P(incorrect) The values for the A, B, C, and D for each class are given in Tables 1. Using these values in the formula the following results are obtained (Table 2). Table 1 Values of A. B. C. D of Cohen’s kappa Class

A

B

C

D

Cemented road

851

33

29

87

Check dam

837

34

28

101

Indian latrines

980

10

3

7

Farm pond

880

17

3

100

Horticulture

900

12

8

80

Soak pits

953

5

1

41

Total

5401

111

72

416

24 Table 2 Cohen’s Kappa result

A. Mangla et al. % of agreement

96.95%

Cohen’s k

0.8030722168974801

Agreement

Almost perfect agreement

4.2 DenseNet121 DenseNet comprises dense blocks connected by transition layer. The component guides a large number of previous layers that are utilized as wellsprings of data, and its own component graphs are used as commitment for each of the following layers [18]. DenseNet designing is isolated into various thick squares. Each plan includes four dense blocks with changing number of layers. The Densenet121has [19–22] layers in the four thick squares. A global spatial average pooling layer is added at the end of the base model of DenseNet-121, along with this a dense output layer with softmax activation function is attached. Softmax is a numerical capacity that changes a vector of numbers into a vector of probabilities, where the probabilities of each worth correspond to the general size of each worth in the vector. Softmax activation function has the given formula.

The input picture is addressed as x o , output of the ith layer as x i, and each convolutional module as capacity H. Contribution to the ith layer is yields of all past layers. 450 × 450 size of images are taken for training and then the data augmentation is applied on the dataset [23]. Batch size of 30 is used for training and 25 for validation. 1000 images for each class and 250 for validation. So in total 6000 images are considered to train the dataset and 1500 images for validation. It is then trained for 30 epochs where it takes 200 steps per epoch for training and 60 steps per epoch for validation. Categorical cross-entropy loss is used to calculate loss. RMSprop is utilized as the optimizer at a learning rate of 1e-4. Metrics used for evaluation are accuracy. After training for 30 epochs, the result of 82.40% accuracy is obtained on training dataset and 84.20% accuracy on validation dataset. The loss for training is 0.4962 and loss for validation is 0.4345. The following plot is received from plotting training accuracy and validation accuracy against epochs.

Image Classification of Indian Rural Development …

25

4.3 Remote-Sensing (RS) Architecture RS model architecture is proposed to further boost network’s performance. As an input layer, input of images of dimensions 350 × 350 × 3 are taken, where, three represent the RGB value of the image. Next, the images are passed through the convolutional 2D layer with 32 filters of dimension 3 × 3 with ReLU as an activation function [24]. With regards to artificial neural networks, the rectifier or ReLU activation function is an actuation work characterized as the positive piece of its contention: f (x) = x + = max where x is the input to a neuron. MaxPooling layer is used in between the convolutional layers. MaxPooling is a discretization process. The goal is to down-example an information portrayal (picture, stowed away layer yield lattice, and so on), diminishing its dimensionality and taking into account suspicions to be made with regards to the highlights contained in the sub-locales binned. When there is a 4 × 4 framework addressing underlying info and a 2 × 2 channel to run over feedback, there will be a step of 2 (which means the (dx, dy) for venturing over feedback (2, 2)) and will not cover areas. For every area addressed by the channel, the maximum of that locale is considered and a new yield lattice is made, where, every component is the maximum of a district in the first information. Total params: 26,458,822 Trainable params: 26,458,822 Non-trainable params: 0 Yield is inferred utilizing argmax work which reacts with the greatest likelihood among the two classes, as anticipated by the model. Probabilities utilizing softmax work are used. The softmax work is a limit that changes a vector of K authentic characteristics into a vector of K real characteristics that totals to 1. The data regards can be positive, negative, or at least zero conspicuous than one, but the softmax transforms them into values of someplace in the scope of 0 and 1, so that they can be interpreted as probabilities. On the off chance that one of the sources of info is little or negative, the softmax transforms it into a little likelihood, and assuming an information is huge, it transforms it into a huge likelihood, yet it will consistently stay somewhere in the range of 0 and 1. Confidence is acquired by ascertaining the softmax of the contributions at the softmax layer. The softmax work is mostly utilized as the last initiation work in the neural organizations for characterization issues. This capacity standardizes an info vector, into a reach that frequently prompts a probabilistic translation. While the facts really confirm that the yield of the softmax capacity can be treated as a likelihood vector, it is not difficult to fall into the snare of entrapping this data with proclamations about trust (in the measurable sense). Since the softmax work doesn’t change the requesting of the yield esteems, the biggest worth before standardization, will in any case be the biggest worth after the standardization. Regardless of whether the expectation is right, there may be a punishment related to yield esteem. Batch size of 30 for training and 25 for validation is considered. 1000 images for each class and 250 for validation. So in total 6000 images are considered to train

26 Table 3 Precision and recall for each class

A. Mangla et al. Class

Precision

Recall

Cement road

0.90

0.889

Check dam

0.92

0.889

Farm pond

0.89

0.923

Horticulture

0.907

0.900

Indian household latrines

0.91

0.892

Soak pits

0.887

0.921

and 1500 images for validation. It is trained the model for 30 epochs where it takes 200 steps per epoch for training and 60 steps per epoch for validation. Categorical cross-entropy loss is used to calculate loss. RMSprop is utilized as the optimizer at a learning rate of 1e-4. Metrics used for evaluation are accuracy. After training for 30 epochs, the result of 88.93% accuracy is obtained on training dataset and 90.00% accuracy on validation dataset. The loss for training is 0.2583 and loss for validation is 0.2541. The following plot is received from plotting training accuracy and validation accuracy against epochs. Evaluation accuracy on test dataset is obtained as 91.50% with a loss of 0.2272 on the test dataset with RS model (Table 3). The above precision and recall results were achieved for each class with RS model.

4.4 Comparative Analysis Figure 2 shows the result of image classification using DenseNet architecture. Using DenseNet-121 models, accuracy of 82.40% with a loss of 0.4962 and validation accuracy of 84.20% with validation loss of 0.4345 are achieved. Fig. 2 Plot for training and validation accuracy for DenseNet-121

Image Classification of Indian Rural Development … Table 4 Accuracy of models

Model

27 Train

Validation

DenseNet 121

0.8240

0.8420

RS Architecture

0.8893

0.9000

Fig. 3 Plot for training and validation accuracy for RS Model

Figure 3 shows the result of image classification using RS model architecture. Using RS model, accuracy of 88.93% with a loss of 0.2583 and validation accuracy of 90.00% with validation loss of 0.2541 are achieved (Table 4). On test dataset with RS model architecture, the accuracy of 0.9150 is obtained (Fig. 3).

5 Conclusion Data augmentation and transfer learning has been shown to enhance model accuracy significantly. Though traditional data augmentation has been used, yet it has proved to be very effective. Transfer learning using DenseNet-121 increases the accuracy. Pre-trained DenseNet-121 model proved to be of great help while training it further on dataset. The RS model consumes less time than the former model to train and hence is able to produce better performance with the help of data augmentation techniques and hyper tuning the model so that the model does not underfit or overfit on the dataset at hand. This shows, although DenseNet121 is a state-of-the-art model, the RS model on this specific task performs better in terms of computation time and accuracy. Experimenting with different hyperparameters to get the overall most efficient results is implemented. There has been previous research using SVM and K-means, but using convolutional neural networks has produced edge over those models. Moreover, using RS model, a better accuracy has been achieved which can further lead in the formation of village information systems and resource allocation.

28

A. Mangla et al.

Data augmentation techniques may be utilized not only to tackle lacking adequate information problems but also, to assist with working on the present state-of-theart classification algorithms. Besides, the above work can be relevant on satellite imagery to distinguish rural development projects, and check for resource allocation and utilization in a more fitting manner.

References 1. A.S. Razavian, H. Azizpour, J. Sullivan, S. Carlsson, et al., CNN features off-the-shelf: an astounding baseline for recognition (2014) 2. M. Tripathi, Analysis of convolutional neural network based image classification techniques. J. Innov. Image Process. (JIIP) 3(02), 100–117 (2021) 3. E.-Y. Huan, G.-H. Wen, Transfer learning with deep convolutional neural network for constitution classification with face image. Multimedia Tools Appl. 79(17–18), 11905–11919 (2020) 4. L. Perez, J. Wang, et al., The effectiveness of data augmentation in image classification using deep learning (2017) 5. G. Huang, Z. Liu, L. van der Maaten, K.Q. Weinberger, Densely connected convolutional networks, in 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) (2017) 6. Y.-D. Zhang, S.C. Satapathy, X. Zhang, S.-H. Wang, COVID-19 diagnosis via DenseNet and optimization of transfer learning setting. Cogn. Comput. (2021) 7. Y. LeCun, P. Haffner, L. Bottou, Yo. Bengio, et al., Object Recognition with gradient based learning (1999) 8. S. Smys, J.I.Z. Chen, S. Shakya, Survey on neural network architectures with deep learning. J. Soft Comput. Paradigm (JSCP) 2(03), 186–194 (2020) 9. T. Guo, J. Dong, H. Li, Y. Gao, Simple convolutional neural network on image classification, in 2017 IEEE 2nd International Conference on Big Data Analysis (ICBDA) (2017). https://doi. org/10.1109/icbda.2017.8078730 10. A. Bashar, Survey on evolving deep learning neural network architectures. J. Artif. Intell. 1(02), 73–82 (2019) 11. H. Singh, K. Krishan, P.K. Litoria, et al., Creation of village information system of Moga district in Punjab using Geoinformatics (2009) 12. J. Al-doski, SB. Mansor, H.Z.M. Shafri, et al., Image classification in remote sensing (2013). (Department of Civil Engineering, Faculty of Engineering, Universiti Putra Malaysia) 13. K.R. Varshney, G.H. Chen, B. Abelson, K. Nowocin, V. Sakhrani, L. Xu, B.L. Spatocco, Targeting villages for rural development using satellite image analysis. Big Data 3(1), 41–53 (2015) 14. T. Vijayakumar, Comparative study of capsule neural network in various applications. J. Artif. Intell. 1(01), 19–27 (2019) 15. C. Shorten, T.M. Khoshgoftaar, A survey on image data augmentation for deep learning. J. Big Data 6(1) (2019) 16. S.C. Wong, A. Gatt, V. Stamatescu, M.D. McDonnell, Understanding data augmentation for classification: when to warp? CoRR, abs/1609.08764, 2016. 1 17. E. Maggiori, Y. Tarabalka, G. Charpiat, P. Alliez, Convolutional neural networks for largescale remote-sensing image classification. IEEE Trans. Geosci. Remote Sens. 55(2), 645–657 (2017). https://doi.org/10.1109/tgrs.2016.2612821 18. W. Liu, K. Zeng, SparseNet: a sparse DenseNet for image classification (2018). Sun Yat-sen University

Image Classification of Indian Rural Development …

29

19. M. Miu, X. Zhang, M.A.A. Dewan, J. Wang, et al., Aggregation and visualization of spatial data with application to classification of land use and land cover (2017) 20. Y. Kubo, G. Tucker, S. Wiesler, Compacting neural network classifiers via dropout training. ArXiv e-prints (2016) 21. T. Wellmann, A. Lausch, E. Andersson, S. Knapp, C. Cortinovis, J. Jache, S. Scheuer, P. Kremer, A. Mascarenhas, R. Kraemer, A. Haase, Remote sensing in urban planning: contributions towards ecologically sound policies? Landscape Urban Plann. 204, 103921 (2020) 22. J. Briskilal, C.N. Subalalitha, An ensemble model for classifying idioms and literal texts using BERT and RoBERTa. Inf. Process. Manage. 59(1), 102756 (2022) 23. K. Simonyan, A. Zisserman, Very deep convolutional networks for large-scale image recognition. arXiv preprint arXiv:1409.1556 (2014) 24. L. Wang, Z.Q. Lin, A. Wong, . COVID-Net: a tailored deep convolutional neural network design for detection of COVID-19 cases from chest X-ray images (2020)

Design of a Mobile Application to Deal with Bullying Vania Margarita, Agung A. Pramudji, Benaya Oktavianus, Randy Dwi, and Harco Leslie Hendric Spits Warnars

Abstract Any type of bullying can make the victims traumatized and not all victims can overcome it without the need for professional help, particularly the teenagers who are in the vulnerable age for bullying. Hence, a mobile application named “Protect ur smile” has been created for victims who can consult with their choice of psychologists. Moreover, they can share their stories among other youngsters who are victimized to gain mental support. The proposed mobile application is designed using the use case diagram, class diagram, and user interface as a prototype showing design. Furthermore, the application has several phases such as registering, login, creating a bully report, consultation, forum, and the information column. Using this mobile application to gain control over bullying, will increase a victim’s confidentiality and decrease the number of suicides. Keywords Bullying mobile application · Anti-bully mobile application · Anti-bullying technology · Software engineering · Infomation systems

V. Margarita · A. A. Pramudji · B. Oktavianus · R. Dwi Information Systems Department, School of Information Systems, Bina Nusantara University, Jakarta 11480, Indonesia e-mail: [email protected] A. A. Pramudji e-mail: [email protected] B. Oktavianus e-mail: [email protected] R. Dwi e-mail: [email protected] H. L. H. S. Warnars (B) Computer Science Department, Graduate Program, Doctor of Computer Science, Bina Nusantara University, Jakarta 11480, Indonesia e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 I. J. Jacob et al. (eds.), Expert Clouds and Applications, Lecture Notes in Networks and Systems 444, https://doi.org/10.1007/978-981-19-2500-9_3

31

32

V. Margarita et al.

1 Introduction In the last few years, “bullying” has been increasingly heard in Indonesia. Bullying is expressing aggressive behavior intentionally to abuse weaker people. There are several types of bullying, such as: 1.

2.

3.

4.

5.

Physical oppression Victims of physical oppression receive various abusive physical treatments ranging from blocking the victim’s path, stumbling, pushing, hitting, grabbing, throwing things at him etc. Verbal oppression Verbal oppression is carried out with painful or teasing words, statements, nicknames, etc. that cause psychological stress. Exclusion Exclusion victims may not be physically or verbally abused, rather be hostile and ignored by their social environment. Victims become isolated and forced to be alone and find it difficult to find friends, since the bully has a strong influence to persuade others to isolate the victim. Cyber-oppression Bullying in cyberspace is known as cyberbullying. Because of its free nature, the victim may receive oppression from someone he does not know or someone with a disguised username. This oppression occurs in cyberspace, for example, through social media and chat applications. Bullying that occurs is usually in the form of insults or satire. It could also be gossip about victims that spread through social media. Sexual-oppression The oppressor will comment on, tease, try to peek, or may touch the victim sexually. Moreover, sexual oppression also includes spreading photos of victims secretly to satisfy the sexual arousal of the perpetrators or forcing victims to watch or see pornography [1].

Based on data from the KPAI or the Indonesia Child Protection Commission in 2017, 73 cases of bullying occurred in children aged 12 to 17 years. In 2018, KPAI said that the increase in bullying on social media was 112 cases [2]. Bully behavior has many adverse effects on victims, for example: a.

Having mental disorders: • • • • •

b.

Depression Inferiority Anxious Difficulty to sleep well Want to hurt their self

Changes are visible in victims: • Not enthusiastic to go to school/work

Design of a Mobile Application to Deal with Bullying

• • • • •

33

Achievement decreases Decreased appetite Run away from home Stressed while returning home Physical injury on the victim

Bully actions are troublesome and inappropriate to be experienced. Bullying also causes damage to the future, which hurts the victim and those around him. People who like to bully, should be given a deterrent effect not to repeat his actions. Because of the rampant bullying cases, this study aims to design a mobile-based application to help victims get psychological help from the psychiatrists. This research has worked along with the Indonesian Psychiatrist Association, which has helped to solve the problem of bullying in society by implementing online consultation in the application with guaranteed confidentiality. Additionally, the application assists victims in reporting to the local police if they want it to be legally processed so that the authorities may take deterrent actions on the bullying perpetrators. This application would have an impact on many people and advance the country’s growth by making minor changes to the younger generation. If bullying is continued without serious measures to stop it, perhaps the younger generation may become embarrassed and unable to make a better future. This application can positively impact victims of bullying to express their feelings.

2 Literature Review Previous research papers from both national and international journals or conferences have been reviewed. According to recent research, young people in this era, are very vulnerable to emotional problems which leads to various kinds of disorders, for example, depression and anxiety. Disorders are caused due to various reasons, such as genetic issue, caused by other disease, or environmental influences [3]. There are cases of bullying beginning from school environment; where schools are supposed to be a place to learn and shape one’s character, it can be a terrible place for some youngsters [4]. Most perpetrators do such immoral acts because of the lack of moral education. Communication within the family is essential to overcome bullying. According to [5] that study about bully-victim behavior and communication between families, it turns out that the role of the family is vital to subdue bullying. According to data that classifies young people based on bullying, 23% are being bullied via Internet, 16% are perpetrators of cyberbullying, the remaining 61% are bullied directly both in the school and in the community [6]. Bully victims become depressed and tend to be quiet and unconfident. The depression rate experienced by verbal bullying is higher than that of cyberbullying because, verbal bullying is done directly to victims and could not be avoidable unlike bullying through social media or the Internet [7]. According to research, the impact of a verbal bully is harrowing.

34

V. Margarita et al.

Victims can be traumatized, severely depressed, or stressed, and the worst effect is suicide. Mocking or spreading a rumor, can be counted as an indirect bully that causes victims to become stressed and depressed [8]. Research indicates that the urge to bully can be conceived through one’s personal character, family problems, and mental health problems. However, moral education is critical to be taught in childhood. Therefore, it requires the willingness of parents and people around, to model good things [9]. The school is also expected to pay attention to the students. School must employ professionals to observe students, who display the following characteristics such as being reticent, anxious, uncomfortable, or abandoned by the community. Sufferers are not very confident, always see the negatives inside them, and think of themselves as stupid or failure [10]. Bullying has had a severe adverse effect for a long time. People who are victims of bullying can turn into bullies due to severe stress, depression, and trauma [11]. According to the data, most direct bullying perpetrators are boys. They are more involved in verbal or physical bullying, while female perpetrators are more involved in indirect bullying and cyberbullying, such as insulting and spreading rumors that hurt victims. Cyberbullying in schools happens a lot nowadays. The perpetrators are usually not aware that their behavior is considered bullying. Victimized students usually become afraid and are embarrassed to go to school [12]. There are many anti-bullying programs being run in schools nowadays. It is expected that the anti-bully program can reduce the incidence of bullying in schools. In the program, it is taught to understand the concept of bullying, which is a stressful life event. According to research, most of the bully perpetrators at school had been victims of bullying on the streets. This makes them want to dominate the school area by bullying other students [13]. So, the role of the social environment becomes essential, especially for teenagers, who are usually looking for their identity. According to Holt and Espelage [13], a study of 784 young people was conducted about their opinions on bully cases. It was found that 61.6% of young people had never been involved in bullying, 14.3% were perpetrators, 12.5% were victims, and 11.6% were bully victims, proving that there are teenagers who have good character and do not wish to be involved in bullying [14]. Apart from being in school, bullying also occurs in the workplace, which results in victims becoming uncomfortable at work. This behavior severely impacts individuals and companies. In many cases, such bullying incidents go unnoticed [15]. Therefore, a right approach is needed to deal with this incident. Cyberbullying also occurs in a relationship. Usually, consequences include breaking up, envy, and jealousy among them. Therefore, bullying actions carried out by the perpetrators need to be discussed and considered, so that there is no more making others life miserable [16].

3 Proposed Work The proposed mobile application is made for bully victims so that they can consult a psychologist and share their stories or events. This mobile-friendly application

Design of a Mobile Application to Deal with Bullying

35

Fig. 1 Use case diagram of the anti-bully mobile application

called “Protect your Smile”, with the tagline “Be Happy Now” helps to reduce the level of depression or trauma experienced. Figure 1 shows the use case of the proposed application where the user can register, login, create a bully report, consult psychologists, create a forum, comment in forum’s thread, and read the information regarding bullying. Moreover, the class diagram of content attributes from the class of database is seen in Fig. 2, where there are seven tables of database such as Survivor, Report, Forum, Admin, Consultation, Psychiatrist, and Report Detail. The survivor table consists of eight attributes which serves the purpose of saving the user’s data. The table is connected to three tables: consultation, report, and forum. The relationship between the consultation table and survivor table is 1..* to 1 because one survivor can make one or more consultations. Similarly with the survivor table and the forum table relationship, because one survivor can make one or more forum threads and comments. A report table is used to save report data, such as descriptions and supporting evidence. The relationship between survivor and report tables is 1 to 1..* because one survivor can have one or many reports. The consultation table is used to save the consultation details between users and psychiatrists. It is related to the psychiatrist table which stores the details of the psychiatrist. A psychiatrist can handle one or more consultations.

36

V. Margarita et al.

Fig. 2 Class diagram of the anti-bully mobile application

Meanwhile, the forum table stores news, events, and thread information. It is related to the admin table because one admin can handle one or many forum posts. The admin table is used to store the admin’s data. The admin table is related to the report detail table because the admin has to check, and follow up on the submitted reports.

4 Results and Discussion The user interfaces of the mobile application are shown in Figs. 3, 4 and 5. The main menu of the application is displayed in Fig. 3a. As seen in the use case diagram (Fig. 1), the activities are elaborated as follows: (1)

Register Register is used by new users when they use this mobile application for the first time. Users have to fill in their data on this page, such as name, email, password, and password confirmation, which is shown in Fig. 3b.

Design of a Mobile Application to Deal with Bullying

Fig. 3 a Main menu, b register form

Fig. 4 a Login form, b create a bully report

37

38

V. Margarita et al.

Fig. 5 a Consultation page, b forum menu

(2)

(3)

(4)

(5)

(6)

Login The login page is used for registered users to open the mobile application. They have to fill in their data on this page, such as their registered email and password, as shown in Fig. 4a. Create a Bully Report This page is used for the users who want to report bullying that happened to themselves or others. They can add evidences as photos, as shown in Fig. 4b. Users can choose different format such as jpg, pdf, movies, mp4, mp3, gif, png etc., for uploading the files. Moreover, the admin can access this page, who has the right to view the report and contact the police if required. Consultation This page is used for the users to consult the psychologists. Fig. 5a shows that consultation is done via text message. The users can choose to display their names or appear anonymous so that they do not feel scared and embarrassed because of privacy leaks. Forum This page is used for the users to communicate with other users, as shown in Fig. 5b. Users can make threads and comment on others. Information This page displays the user’s profile, as shown in Fig. 6a, and the psychologists’

Design of a Mobile Application to Deal with Bullying

39

Fig. 6 a User profile, b psychologists’ profile

profile, as shown in Fig. 6b. The users can see psychologists’ educational backgrounds and career life. Moreover, there are various information about bullying. One of the regulations on bullying in Indonesia (as shown in Fig. 7) which emphasizes child protection is that the perpetrators of bullying can be convicted under Law Number 23 of 2002. Article 54 of Law 35/2014 says that children in and within the education unit are required to receive protection from acts disturbing their physical and mental growth.

5 Conclusion This application is intended to help survivors of bullying to report to the police, consult with psychiatrists, and share their experiences and incidents with others. In the future, there are opportunities to improve and expand this application by adding more features, such as, emergency contacts and emergency records. This work hopes to reduce intimidation because, without intimidation, the community would be a better place, the development would not be disturbed, and everyone can feel safe. In the future, use of Artificial Intelligence (AI) technology would be carried out by implementing machine learning or deep learning algorithms. Using sensors as the

40

V. Margarita et al.

Fig. 7 Regulation page of anti-bully mobile application

Internet of Things (IoT) implementation is also a part of AI technology implementation, where machines using sensors will be able to talk with other machines and hence could be executed in the upcoming research.

References 1. F. Dehue, C. Bolman, T. Vollink, Cyberbullying: youngsters’ experiences and parental perception. CyberPsychol. Behav. 11(2) (2008) 2. T.L. Faiza, Perbedaan Tingkat Depresi pada Korban Bullying Verbal dan Cyberbullying pada Remaja, University of Muhammadiyah Malang Bachelor Thesis, 2019 3. A.C. Baldry, The impact of direct and indirect bullying on the mental and physical health of Italian youngsters. Aggressive Behav. 30(5), 343–355 (2004) 4. L. Arseneault, L. Bowes, S. Shakoor, Bullying victimization in youths and mental health problems: ‘Much ado about nothing? Psychol. Med. 40(5), 717–729 (2010) 5. I. Rivers, V.P. Poteat, N. Noret, N. Ashurst, Observing bullying at school: the mental health implications of witness status. Sch. Psychol. Q. 24(4), 211–223 (2009) 6. P.R. Smokowski, K.H. Kopasz, Bullying in school: an overview of types, effects, family characteristics, and ıntervention strategies. Child. Sch. 27(2), 101–110 (2005)

Design of a Mobile Application to Deal with Bullying

41

7. D. Wolke, S.T. Lereya, Long-term effects of bullying. Arch. Dis. Child. 100(9), 879–885 (2015) 8. J. Wang, R.J. Iannotti, T.R. Nansel, School bullying among adolescents in the United States: physical, verbal, relational, and cyber. J. Adolesc. Health 45(4), 368–375 (2009) 9. P.W. Agatston, R. Kowalski, S. Limber, Students’ perspectives on cyber bullying. J. Adolesc. Health 41(6), S59–S60 (2007) 10. M.M. Ttofi, D.P. Farrington, Effectiveness of school-based programs to reduce bullying: a systematic and meta-analytic review. J. Exper. Criminol. 7(1), 27–56 (2011) 11. S.M. Swearer, S. Hymel, Understanding the psychology of bullying: moving toward a socialecological diathesis-stress model. Am. Psychol. 70(4), 344–353 (2015) 12. H. Andershed, M. Kerr, H. Stattin, Bullying in school and violence on the streets: are the same people ınvolved? J. Scand. Stud. Criminol. Crime Prev. 2(1), 31–49 (2010) 13. M.K. Holt, D.L. Espelage, Perceived Social Support among Bullies, Victims, and BullyVictims. J. Youth Adolesc. 36(8), 984–994 (2007) 14. L. Hidayati, Pembulian di Tempat Kerja dalam konteks Asia, in Research Report of Seminar Nasional dan Gelar Produk (SENASPRO) (2016), pp. 133–142 15. D.L. Hoff, S.N. Mitchell, Cyberbullying: causes, effects, and remedies. J. Educ. Adm. 47(5), 652–665 (2009) 16. D. Olweus, S.P. Limber, Bullying in school: evaluation and dissemination of the Olweus Bullying Prevention Program. Am. J. Orthopsychiatry 80(1), 124–134 (2010)

Selection of Human Resources Prospective Student Using SAW and AHP Methods Ahmad Rufai, Diana Teresia Spits Warnars, Harco Leslie Hendric Spits Warnars, and Antoine Doucet

Abstract The role of Human Resources (HR) in an organization is significant, and the role of the world of education plays an essential role in producing and educating qualified and qualified human resources. In this paper, 20 students were evaluated as learning with Simple Additive Wight (SAW) and Analytic Hierarchy Process (AHP), which applied six criteria: experience, recommendations, interviews, discipline, skills, and physical health. The SAW and AHP methods are techniques for determining the best value from several predetermined criteria so that they are suitable for use in the selection of HR candidates the company will accept. The results showed that the consistency of similarity between the SAW and AHP methods had a score of (7 + 8)/20 = 0.75 or a similarity consistency index of 75%. The similarity consistency index is evaluated using the same rank order and the opposite with a reversal. Keywords Simple additive weighting · Analytic hierarchy process · Human resource selection · Decision support system

A. Rufai · D. T. S. Warnars Information Systems Department, School of Information Systems, Bina Nusantara University, Jakarta 11480, Indonesia e-mail: [email protected] D. T. S. Warnars e-mail: [email protected] H. L. H. S. Warnars (B) Computer Science Department, Binus Graduate Program-Doctor of Computer Science, Bina Nusantara University, Jakarta 11480, Indonesia e-mail: [email protected] A. Doucet Laboratoire L3i, Université de La Rochelle, La Rochelle, France e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 I. J. Jacob et al. (eds.), Expert Clouds and Applications, Lecture Notes in Networks and Systems 444, https://doi.org/10.1007/978-981-19-2500-9_4

43

44

A. Rufai et al.

1 Introduction Looking at these problems, the role of the Human Resources (HR) division is considered to be not yet maximal in handling the problem of selecting human resources. Selecting or receiving human resources is still not done professionally but using bribery, friendship, or family relations. This is happening because there is no human resource selection process with a standard systematic method to assess the eligibility of prospective HR. Therefore, in overcoming the professional acceptance of human resources, the role of the Human Resources (HR) division is to establish cooperation with educational institutions/schools. The company (Human Resources division) surrendered fully to educational institutions/schools to select quality prospective human resource students in this collaboration. Based on the background of the problem outlined above, several problems can be formulated, namely: 1 2 3

How is the decision support system for selecting prospective HR students using the SAW and AHP methods?. Regarding the calculation process, which method is the easiest to understand and is more suitable for selecting HR candidates. From the side of the final results obtained, which method is more accurate.

It is necessary to limit the practical problem so that the problem under study is straightforward and not complex. In this case, the restrictions taken are: 1. 2.

Research focuses on selecting HR candidates and the final results of the method’s calculation. Examples of the application are carried out to evaluate the selection of twenty prospective HR students.

2 Current and Previous Similar Research Papers The Decision Support System (DSS) was first revealed in the early 1970s by Michael S. Scott Morton in the term Management Decision System. The system is a computerbased system that is shown to help decision-makers by utilizing specific data and models to solve various unstructured problems. Meanwhile, Human Resources (HR) is the most crucial asset in a company or organization. If managed correctly and adequately, employees can be potential but will be a burden if mismanaged. The SAW method is also a term that often weighs the sum method. The basic concept of the SAW method is to find the number of performance ratings for each alternative on all attributes. The SAW method requires the decision matrix normalization process (X) to a scale that can be compared with all rankings of alternatives. Analytical Hierarchy Process (AHP) is a decision support model developed by Thomas L. Saaty. In essence, AHP takes into account qualitative and quantitative matters. The concept is to change qualitative values into quantitative values to be taken more objectively.

Selection of Human Resources Prospective …

45

Combination of SAW and AHP methods have been implemented in many papers, and we are limited to current paper publications such as: 1.

2.

3.

4.

5.

6.

7.

8.

9.

SAW and AHP methods are used to weigh for school e-learning readiness evaluation. They use eight criteria to assess psychological readiness, Sociological readiness, Environmental readiness, Human resource readiness, Financial readiness, Technological skill (aptitude) readiness, equipment readiness, Content readiness [1]. Meanwhile, SAW and AHP methods were compared for the tender process of television transmission stations based on four criteria: price, quality, service, and reliability. The paper mentioned that the SAW method is easy to apply rather than the AHP method[2]. A comparative study was done where SAW-AHP methods were compared to SAW, TOPSIS, and TOPSIS-AHP to determine students’ tuition fees where the SAW-AHP method had a 74% accuracy measurement[3]. Moreover, 29 websites of the international museum were assessed using a combination of SAW and AHP methods using 14 criteria in 3 categories such as content, usability, and functionality[4]. The criteria include currency/clarity/text comprehension, completeness/richness, quality content, support of research, consistency, accessibility, structure/navigation, easy to use/simplicity, user interface/overall presentation/design, efficiency, multilingualism, multimedia, interactivity, and adaptivity. Meanwhile, SAW and AHP methods were used for food crop planting recommendations, using three criteria: rainfall, temperature, and humidity. This paper used AHP for weighting criteria and SAW for ranking each criterias[5]. SAW and AHP methods were combined for determining the location of cake shop retail business using criteria such as revenue and distance using 50 cake shops around java island and assessed using specific criteria such as a competitor, infrastructure, distance, rental price, population density, size, road condition, and culture [6]. This paper discusses the combination of SAW and AHP methods for determining natural sandy gravel using ten criteria upon 19 mining locations, categorized into natural, environmental, and esthetic factors [7]. Groundwater quality assessment et al.-Shekhan area in Mosul city, Iraq, was assessed using SAW and AHP methods. Groundwater quality assessment uses 30 wells as water resources as an alternative are assessed. These 30 well are assessed using 12 criteria drinking parameters such as Depth, Ca, Mg, CI, Na, SAR, SO4 , HCO3 , NO3 , TDS, EC, and pH [8]. A Decision Support System application of the rental car system in Bali island was developed using SAW and AHP methods as a recommendation and including Geographic Information System (GIS) for location position. The AHP was used by calculating the weights on each of the criteria and then proceeding with SAW to determine the weight rating specified in the criteria [9].

46

10.

11.

12.

13.

14.

15.

16.

17.

A. Rufai et al.

The hybrid method was developed for decision support systems using SAW and AHP methods to weight bandwidth management for faculty at Semarang university [10]. The alternative used seven faculties such as FTIK, Psychology, Laws, Engineering, Economic, THP, and POST upon five criteria: number of lab computers, number of lecturers, number of students, weekly teaching time, and number of the study program. SAW and AHP methods were combined to make a decision support system for satisfaction measurement at service-provided e-procurement to evaluate customer service quality[11]. They used some criteria such as parts, employees, system acceptance, usefulness, ease of use, etc. The quality salt industry in Indonesia was assessed by comparing SAW and AHP methods to determine the highest quality salt. The comparison showed that the SAW method is better with an accuracy of 80% rather than the AHP method with an accuracy of 76%, using three criteria to weight salt quality such as NaCL, water content, and salt color[12]. A decision-making application was created on choosing a keyword, where a keyword selection website improves search engine visibility using a combination of SAW and AHP methods. The webmaster used the system to prioritize the optimal keywords, and they used some keyword searching text and assessed using four criteria: volume, result, CPC, and competition [13]. AHP method was used to weight the parameters and the SAW method to find the final value and rank. A decision support system application for significant selection for the new student was built for Buddhi high school using SAW and AHP methods combination, where the SAW method was used to weight the performance rating score for each alternative, and the AHP method was used to carry on Pairwise Comparison Matrix (PCM) including Normalization PCM (NPCM). The process used four criteria: student middle school report score, student national exam scores, academic potential test score, and interview result in score [14]. A novel hybrid approach as a combination of SAW and AHP methods was proposed to measure and evaluate the preeminent brake friction composite formulation to find the best formulation that yields optimized tribological properties considering all the performance attributes at a single time. The formulation NF-1 with five wt% ramie fiber is the best combination of tribological properties [15]. SAW and AHP methods were combined for a decision support system for private tutor business as non-formal education called “E-private” using five criteria: education, experience, cost, discipline, and teaching. AHP method was done by doing input criteria and comparison scale, PCM activity, priority weight and find the consistency ratio where if not consistent, it will loop to new input criteria and comparison scale, while if consistent, the SAW method will be applied. SAW method will value the alternatives, create a matrix, normalize the matrix, and show the ranking result [16]. Another decision support application was built using by comparison SAW and AHP methods for selection of giving subsidized home loan which is used by

Selection of Human Resources Prospective …

18.

19.

20.

21.

47

developer company using six criteria such as type of work, income, number of dependents, down payment, Indonesian Bank (BI) checking and the completeness of documents [17]. In this paper, the SAW method has a better result than the AHP method. A simple application was developed using SAW and AHP methods to help the music studio or music group in the singer selection process and find the best selection singer, using experimental results and consistent results. AHP was used for the weighting process of criteria and SAW for participants’ scores [18]. Risk analysis as risk management as anticipate the hazard and develop the action to reduce the risk at PT. PLN Persero was assessed using a combination of SAW and AHP methods, where the SAW method was applied to measure the subjective concept of human-related uncertainty and the AHP method to determine the weight criteria [19]. In this paper, the SAW method was run in 9 steps while the AHP method was in 4 steps. Decision support system application for employee placement is created and compared using four methods: SAW, AHP, TOPSIS, and Preference Ranking Organization Method for Enrichment Of Evaluations (PROMENTHEE). Five criteria are used such as knowledge, skill, ability, physical, and attitude, and based on the experiments upon 60 datasets for employee placement then AHP has 50% accuracy, SAW has 81% accuracy, TOPSIS has 95% accuracy, and the PROMENTHEE method has 93.44% accuracy [20]. Groundwater potential (GWP) in the United Arab Emirates (UAE) and Oman is modeled by the influence of physiographic variables affecting groundwater accumulation and three different methods such as SAW, AHP, and Probabilistic Frequency Ratio (PFR). The SAW and AHP were valid for well potential zones as water resources at 98% and 92%, and spring at 63% and 86%, respectively [21].

3 Simple Additive Weighting Implementation The beginning of the process carried out in the SAW method is to classify the criteria chosen to determine the decision. These criteria are divided into two attribute categories: the cost and benefit criteria. The benefit is if the match value of each criterion, the higher the value, the better. While, the cost is if the match value of each criterion, the smaller the value, the better. SAW method has four steps such as:

1 2 3 4

List of Weighting criteria, including the range of criteria and normalization of criteria range. Scoring of Sample data using a normalized range of criteria. Normalization of performance rating score. Ranking Score.

48

A. Rufai et al.

Table 1 Weighting criteria No.

Criteria

Criteria weight (%)

Range of criteria

Normalized range of criteria 1–2

1

Experience (K 1 )

5

0, 1

2

Recommendations (K 2 )

20

0, 1, 2, > 2

1–4

3

Interview (K 3 )

10

1–5

1–5

4

Discipline (K 4 )

10

1–4

1–4

5

Skills (K 5 )

25

0–5

1–6

6

Physical health (K 6 ) 30

1–4

1–4

3.1 List of Weighting Criteria, Including the Range of Criteria and Normalization of Criteria Range Following what has been determined, the weighting criteria are six criteria with each range of criteria. Next is the detail of the range criteria. a. b. c. d. e. f.

Criteria experience has ranges 1 or 0 for with or without. Criteria recommendation has ranges 0, 1, 2, > 2 for more minor, adequate, very adequate, substantial. Criteria Interview has ranges 1, 2, 3, 4, 5 for very bad, bad, adequate, good, very good. Criteria Discipline has ranges 1, 2, 3, 4 for deficient, adequate, good, very good. Criteria Skill has ranges 0, 1, 2, 3, 4, 5 for bad, deficient, adequate, passable, good, very good. Criteria Physical health has ranges 1, 2, 3, 4 for bad, deficient, satisfied, very satisfied.

The normalized range of criteria can be seen in the last column in Table 1, where all the ranges will have similarities and start from 1 to 6.

3.2 Scoring of Sample Data Using a Normalized Range of Criteria After having the weighting criteria table, the scoring students’ data will be normalized with the last column in Tables 1 and 2 showing the result of 20 students’ scoring as in this paper using 20 students as examples.

Selection of Human Resources Prospective …

49

Table 2 Results of collecting student/student assessment data Alternative

Criteria K1

K2

K3

K4

K5

K6

Student 1

2

1

2

3

3

2

Student 2

1

1

3

2

3

2

Student 3

1

2

4

3

2

3

Student 4

1

1

3

4

3

3

Student 5

2

4

3

2

4

4

Student 6

1

2

3

2

3

3

Student 7

2

3

3

2

3

2

Student 8

1

1

2

2

2

2

Student 9

2

3

4

3

5

3

Student 10

1

1

3

3

6

3

Student 11

2

1

2

3

3

3

Student 12

1

1

3

3

6

3

Student 13

1

2

2

1

1

3

Student 14

1

1

2

2

4

2

Student 15

1

2

3

2

4

3

Student 16

2

1

4

3

3

4

Student 17

1

1

2

2

3

2

Student 18

1

1

1

1

2

2

Student 19

1

2

2

3

4

2

Student 20

2

4

5

3

5

3

3.3 Normalization of Performance Rating Score In order to find normalized performance rating, then Eq. (1) is used where there are two options for benefit and cost, but in this paper, we limit to benefit purpose only X then we use equation r ij = Maxi jX i j .  ri j =

Xi j , MaxX i j Min X i j , Xi j

if attribut j is benefit if attribut j is cos t

where r ij X ij Max X ij Min X ij i j

is the normalized performance rating score is the Weighting criteria score is the highest weighting criteria score is the lowest weighting criteria score is an alternative (student) is Criteria

(1)

50

A. Rufai et al.

Table 3 Normalized matrix Alternative

Criteria K1

K2

K3

K4

K5

K6

Vi

Ranking order

Student 1

1.00

0.25

0.40

0.75

0.50

0.50

0.49

14

Student 2

0.50

0.25

0.60

0.50

0.50

0.50

0.46

16

Student 3

0.50

0.50

0.80

0.75

0.33

0.75

0.588

8

Student 4

0.50

0.25

0.60

1.00

0.50

0.75

0.585

9

Student 5

1.00

1.00

0.60

0.50

0.67

1.00

0.827

2

Student 6

0.50

0.50

0.60

0.50

0.50

0.75

0.585

10

Student 7

1.00

0.50

0.60

0.50

0.50

0.50

0.585

11

Student 8

0.50

0.25

0.40

0.50

0.33

0.50

0.398

19

Student 9

1.00

0.75

0.80

0.75

0.83

0.75

0.788

3

Student 10

0.50

0.25

0.60

0.75

1.00

0.75

0.685

4

Student 11

1.00

0.25

0.40

0.75

0.50

0.75

0.565

12

Student 12

0.50

0.25

0.60

0.75

1.00

0.75

0.685

5

Student 13

0.50

0.50

0.40

0.25

0.17

0.75

0.457

17

Student 14

0.50

0.25

0.40

0.50

0.67

0.50

0.482

15

Student 15

0.50

0.50

0.60

0.50

0.67

0.75

0.627

7

Student 16

1.00

0.25

0.80

0.75

0.50

1.00

0.68

6

Student 17

0.50

0.25

0.40

0.50

0.50

0.50

0.44

18

Student 18

0.50

0.25

0.20

0.25

0.33

0.50

0.353

20

Student 19

0.50

0.50

0.40

0.75

0.67

0.50

0.557

13

Student 20

1.00

1.00

1.00

0.75

0.83

0.75

0.858

1

As seen in Table 2, Max Xij for Criteria K 1 , K 2 , K 4 , K 5 , and K 6 are 2, 4, 5, 4, 6, and 4. For example, for student 1, as seen in the first line in Table 3 using equation X ri j = Maxi jX i j where i = 1 and j = 1 to 6, then: r ij = r 11 = 2/2 = 1, r ij = r 12 = 1/4 = 0.25, r ij = r 13 = 2/5 = 0.4, r ij = r 14 = 3/4 = 0.75, r ij = r 15 = 3/6 = 0.5 and r ij = r 16 = 2/4 = 0.5. The first line student 1 as shown in Table 3 shows the result of implementation Eq. (1) with score 1, 0.25, 0.4, 0.75, 0.5 and 0.5. Table 3 shows the result for 20 students’ data in Table 2 which are applied with Eq. (1).

3.4 Ranking Score The final step is to rank using the Eq. (2) Vi =

n  j=i

W j ri j

(2)

Selection of Human Resources Prospective …

51

Information: Vi Wj r ij i j

ranking for each alternative (student) the weight score of each criteria normalized performance ranking value is alternative (student) is Criteria

W j is criteria weight as seen in the third column in Table 1 with composition Criteria K 1 , K 2 , K 3 , K 4 , K 5 , and K 6 with weight score 0.05, 0.2, 0.1, 0.1, 0.25, and 0.3 respectively. For example for student 1 as seen in first line in Table 3 using Eq. (2) where i = 1 and j = 1 to 6, then: V i = (0.05*1) + (0.2*0.25) + (0.1*0.4) + (0.1*0.75) + (0.25*0.50) + (0.3*0.5) = 0.05 + 0.05 + 0.04 + 0.075 + 0.125 + 0.15 = 0.49. Last column in Table 3 shows the V i as ranking score for each student and the highest score 0.858 for student 20 shows as the best student and the lowest score 0.353 for student 18.

4 Analytical Hierarchy Process Implementation The AHP method will be run with two steps such as. 1. 2.

Processing Criteria. Processing alternatives. The AHP method will process the criteria as there are six criteria in this paper such as K 1 , K 2 , K 3 , K 4 , K 5 , and K 6 , and 20 alternatives such as student 1 to student 20, as seen in Table 3. Each step will apply the Pairwise Comparison Matrix (PCM) and Normalized Pairwise Comparison Matrix (NPCM). Based on NPCM, the alternatives will be scored to get the best criteria and the best alternatives.

4.1 Processing Criteria Processing criteria are a step that process criteria as there are six criteria in this paper such as K 1 , K 2 , K 3 , K 4 , K 5 , and K 6 , as seen in Table 3. The processing criteria have three sub-steps such as: a. b. c.

Pairwise Comparison Matrix. Normalized Pairwise Comparison Matrix. Calculating the consistency.

52

4.1.1

A. Rufai et al.

Pairwise Comparison Matrix

Pairwise comparison matrix (PCM) is created based on criteria weight on the third column in Table 1 with composition Criteria K 1 , K 2 , K 3 , K 4 , K 5 , and K 6 with weight scores 0.05, 0.2, 0.1, 0.1, 0.25, and 0.3, respectively. The pairwise comparison matrix is created with Eq. (3). Pairwise Comparison matrix = row/col

(3)

In Eq. (3), row and column numbers are row and column numbers in Table 4. For example for first row is K 1 /K 1 = 0.05/0.05 = 1, K 1 /K 2 = 0.05/0.2 = 0.25, K 1 /K 3 = 0.05/0.1 = 0.5, K 1 /K 4 = 0.05/0.1 = 0.5, K 1 /K 5 = 0.05/0.25 = 0.2, K 1 /K 6 = 0.05/0.3 = 0.17, then in the first column in Table 4 will have score, 1, 0.25, 0.5, 0.5, 0.2 and 0.17. The last column in Table 4 as summarization for each column which will be used for next step to NPCM. Table 5 is a scale of relative importance (Saaty Scale) for Table 4, giving meaning for the PCM. For example, score five at row 5, and column 1 (K 5 /K 1 ) shows essential importance based on the Saaty scale, as shown in Table 5. Table 4 Pairwise comparison matrix for criteria Row/col

K1

K2

K3

K4

K5

K6

K1

1.00

0.25

0.50

0.50

0.20

0.17

K2

4.00

1.00

2.00

2.00

0.80

0.67

K3

2.00

0.50

1.00

1.00

0.40

0.33

K4

2.00

0.50

1.00

1.00

0.40

0.33

K5

5.00

1.25

2.50

2.50

1.00

0.83

K6

6.00

1.50

3.00

3.00

1.20

1.00

sum

20

5

10

10

4

3.33

Table 5 Saaty scale

Comparative scale

Definition

1

Equal Importance

3

Weak importance of one over another

5

Essential or strong importance

7

Demonstrated importance

9

Extreme importance

2, 4, 6, 8

Intermediate values between the two adjacent judgments

Reciprocal

The opposite

Selection of Human Resources Prospective …

4.1.2

53

Normalized Pairwise Comparison Matrix

Normalization of Pairwise Comparison (NPCM) will be done using Eq. (4) by dividing each score with score summarization for each column as shown in the last row in Table 4. Normalized PCM = score/sum_col

(4)

A score is a number for each score, and sum_col summarizes each column. Table 6 shows the Normalized PCM with criteria weights column using Eq. (5), an average cumulation row score average. Criteria weights can be called a priority vector or Eigenvector. Criteria weights = Avg(sum_row_score)

(5)

Criteria weights = (EV(1/n) ) ∗ sum(EV)

(6)

where EV = Eigen Value as multiplication all score in the same row. Furthermore, finding Eigenvector can be done by finding “eigenvalue” first by calculation all the scores per row in the PCM table as shown in Table 4. For example, they used the first row in Table 4, Eigen Value = (1 * 0.25 * 0.5 * 0.5 * 0.2 * 0.17) = 0.0021. Using Eq. (6), then Eigen Value power or ^ (1/n) * sum(Eigen Value) where n is the number of criteria = 6, = 0.0021^(1/6) = 0.359 as shown in the first row in column “Eigen Value” in Table 6. After all the criteria have found the Eigenvalue, the sum(Eigen Value) is 7.146, as shown in the last row in column “Eigen Value” in Table 6. Then, for example, the first row in column “Eigen Value” in Table 6, “Criteria Weights” = 0.359/7.146 = 0.05 as the same score value in the first row in column “Criteria Weights” in Table 6.

4.1.3

Calculating the Consistency

Meanwhile, Calculation of the consistency is used to check whether the calculation scores are correct, where each score in the PCM in Table 4 is multiplied with the “criteria weights” column in Table 6. After the process was done, the result was the same as shown in Table 6. The column “weighted sum value” in Table 6 was getting using Eq. (7), where each score summarizes each row. Weighted sum value = sum(row score)

(7)

  Criteria Weights = EV(1/n)

(8)

0.25

0.30

0.25

0.30

1.00

K5

K6

Sum

0.10

1.00

0.10

0.10

0.10

K3

0.20

0.05

K2

K4

0.05

0.20

K1

K2

K1

Row/col

1.00

0.30

0.25

0.10

0.10

0.20

0.05

K3

Table 6 Normalized PCM for criteria

1.00

0.30

0.25

0.10

0.10

0.20

0.05

K4

1.00

0.30

0.25

0.10

0.10

0.20

0.05

K5

1.00

0.30

0.25

0.10

0.10

0.20

0.05

K6

7.146

2.144

1.786

0.714

0.714

1.431

0.359

Eigen Value

6.00

1.80

1.50

0.60

0.60

1.20

0.30

Weighted sum value

1.00

0.30

0.25

0.10

0.10

0.20

0.05

Criteria weights/Eigenvector

λ max = 6

6

6

6

6

6

6

Ratio

54 A. Rufai et al.

Selection of Human Resources Prospective …

55

Table 7 Random consistency index (RI) n

1

2

3

4

5

6

7

8

9

10

RI

0

0

0.58

0.9

1.12

1.24

1.32

1.41

1.45

1.49

Furthermore, finding “Criteria weights” can be done with Eq. (8) upon normalized PCM Table 6 and using EV = Eigen value as multiplication all scores in the same row, for example, using the first row in Table 6 then Eigen Value = (1 * 0.05 * 0.05 * 0.05 * 0.05 * 0.05) = 0.000000016. After that, Eigen Value power or ^ (1/n) = 0.000000016^(1/6) = 0,05, where n is the number of criteria = 6. The same score value is in the first row in column “Criteria Weights” in Table 6. Ratio = Weighted sum value/Criteria weights

(9)

λ max = AVG(Ratio)

(10)

CI =

λ max −n n−1

CR = CI/RI

(11) (12)

Moreover, column ratio was got it using Eq. (9) by dividing the “Weighted sum value” with “Criteria weights.” Lambda maximum (λ max) as = the largest eigenvalue has a score of 6 using Eq. (10) as the average for all ratio scores. CI (Consistency −n = 6−6 = 05 = 0. Index) was done using Eq. (11) CI = λ max n−1 6−1 Furthermore, Consistency Ratio (CR) is a comparison between CI and RI, where CI was processed using Eq. (11) and Random Consistency Index (RI), which randomly generated can be seen in Table 7 based on the number of n as the number of criteria = 6. Since n = 6, the RI score = 1.24, as seen in Table 7. Then, CR = CI/RI = 0/1.24 = 0. If CR < = 0.1, then the consistency is acceptable, and on the other hand, if CR > 0.1, the comparison needs to be reevaluated again. The value of 0 is the lowest and can be the most consistent, so the value cannot be harmful, even if the value is negative, meaning there is an error in the calculation process. Since the finding CR = 0, then the equation is consistent.

4.2 Processing Alternatives After processing criteria, processing alternatives is a step that processes 20 alternatives, such as student 1 to student 20, as seen in Table 3. The processing alternatives have two sub-steps such as:

56

a. b.

A. Rufai et al.

Pairwise comparison matrix. Normalized pairwise Comparison matrix.

4.2.1

Pairwise Comparison Matrix

Pairwise Comparison Matrix (PCM) for alternatives is created based on 20 student assessment data in Table 2, where each student has weighing criteria, and each PCM will be scored based on each criterion where in this paper, there are six criteria such as K 1 , K 2 , K 3 , K 4 , K 5 , and K 6 . So, there will be six tables like Table 8, which shows the PCM alternative for each criterion, and Table 8 shows the PCM alternative for criteria K 1 as “experience” criteria as shown in Table 2. Attributes S 1 to S 20 in Table 8 represent student 1 to student 20. Each column in Table 8 is a comparison matrix for criteria K 1 , as shown in Table 2, where each criterion score per student in column K 1 will be pairwise with other K 1 student criteria scores. For example, the first row S 1 in Table 8, such as column S 1 /S 1 = 2/2 = 1 where S 1 = 2 is score student 1 at column K 1 in Table 2. Moreover, S 1 /S 2 = 2/1 = 2, S 1 /S 3 = 2/1 = 2, S 1 /S 4 = 2/1 = 2, S 1 /S 5 = 2/2 = 1 and etc., where S 2 = 1, S 3 = 1, S 4 = 1, S 5 = 2 as shown at each student number at column K 1 in Table 2. Meanwhile, the last row in Table 8 as summarization for each column will be used for the next step to normalize the PCM.

4.2.2

Normalized Pairwise Comparison Matrix

Normalization of Pairwise Comparison (NPCM) will be done using Eq. (4) by dividing each score with score summarization for each column as shown in the last row in Table 8. As seen in Eq. (4), Normalized PCM = score /sum_col, where the score is a number for each score and sum_col is summarization for each column in Table 8. Since there are six criteria such as K 1 , K 2 , K 3 , K 4 , K 5 , and K 6 , then the same like PCM will have six tables then there are 6 NPCM tables, and Table 9 is an example of NPCM for an alternative based on criteria K 1 as “experience” criteria as shown in Table 1. For example, the first row S 1 in Table 9, such as column S 1 /S 1 = 1/13.5 = 0.074, where 1 is a score at position S 1 , S 1 , and 13,5 is summarization in Table 8. Moreover, S 1 /S 2 = 2/27 = 0.074, where 2 is a score at position S 1 /S 2 and 27 is the summarization score column S 2 at Table 8. Meanwhile, the last row in Table 9 as summarization for each column that should have scored one and score 1 in the last row shows that the equations were done correctly. After having 6 PCM and NPCM tables based on Tables 8 and 9, respectively then from each NPCM table will process the Criteria Weights or Eigenvector using Eqs. (5) or (6) or (8), and Table 10 shows the Eigenvector for all six criteria such as K 1 , K 2 , K 3 , K 4 , K 5 , and K 6 respectively. Similar to summarization in the last row for NPCM in Table 9, where the score should be one, the last row of Table 10 should have a score of 1 to represent the success of equations implementation.

Selection of Human Resources Prospective …

57

Table 8 Pairwise comparison matrix for alternatives based on criteria K 1 Alter native

S1

S2

S3

S4

S5

S6

S7

S8

S9

S 10

S1

1

2

2

2

1

2

1

2

1

2

S2

0.5

1

1

1

0.5

1

0.5

1

0.5

1

S3

0.5

1

1

1

0.5

1

0.5

1

0.5

1

S4

0.5

1

1

1

0.5

1

0.5

1

0.5

1

S5

1

2

2

2

1

2

1

2

1

2

S6

0.5

1

1

1

0.5

1

0.5

1

0.5

1

S7

1

2

2

2

1

2

1

2

1

2

S8

0.5

1

1

1

0.5

1

0.5

1

0.5

1

S9

1

2

2

2

1

2

1

2

1

2

S 10

0.5

1

1

1

0.5

1

0.5

1

0.5

1

S 11

1

2

2

2

1

2

1

2

1

2

S 12

0.5

1

1

1

0.5

1

0.5

1

0.5

1

S 13

0.5

1

1

1

0.5

1

0.5

1

0.5

1

S 14

0.5

1

1

1

0.5

1

0.5

1

0.5

1

S 15

0.5

1

1

1

0.5

1

0.5

1

0.5

1

S 16

1

2

2

2

1

2

1

2

1

2

S 17

0.5

1

1

1

0.5

1

0.5

1

0.5

1

S 18

0.5

1

1

1

0.5

1

0.5

1

0.5

1

S 19

0.5

1

1

1

0.5

1

0.5

1

0.5

1

S 20

1

2

2

2

1

2

1

2

1

2

Sum

13.5

27.0

27.0

27.0

13.5

27.0

13.5

27.0

13.5

27.0

Alter native

S 11

S 12

S 13

S 14

S 15

S 16

S 17

S 18

S 19

S 20

S1

1

2

2

2

2

1

2

2

2

1

S2

0.5

1

1

1

1

0.5

1

1

1

0.5

S3

0.5

1

1

1

1

0.5

1

1

1

0.5

S4

0.5

1

1

1

1

0.5

1

1

1

0.5

S5

1

2

2

2

2

1

2

2

2

1

S6

0.5

1

1

1

1

0.5

1

1

1

0.5

S7

1

2

2

2

2

1

2

2

2

1

S8

0.5

1

1

1

1

0.5

1

1

1

0.5

S9

1

2

2

2

2

1

2

2

2

1

S 10

0.5

1

1

1

1

0.5

1

1

1

0.5

S 11

1

2

2

2

2

1

2

2

2

1

S 12

0.5

1

1

1

1

0.5

1

1

1

0.5

S 13

0.5

1

1

1

1

0.5

1

1

1

0.5

S 14

0.5

1

1

1

1

0.5

1

1

1

0.5 (continued)

58

A. Rufai et al.

Table 8 (continued) Alter native

S 11

S 12

S 13

S 14

S 15

S 16

S 17

S 18

S 19

S 20

S 15

0.5

1

1

1

1

0.5

1

1

1

0.5

S 16

1

2

2

2

2

1

2

2

2

1

S 17

0.5

1

1

1

1

0.5

1

1

1

0.5

S 18

0.5

1

1

1

1

0.5

1

1

1

0.5

S 19

0.5

1

1

1

1

0.5

1

1

1

0.5

S 20

1

2

2

2

2

1

2

2

2

1

Sum

13.5

27.0

27.0

27.0

27.0

13.5

27.0

27.0

27.0

13.5

Moreover, the score for each column should be related to each criterion, where column K 1 uses criteria K 1 . For example, for the position, Student 1/K 1 in Table 10 has a score of 0.07407407, and using Eq. (5) where Criteria weights = Avg (sum_row_score), then the score 0.07407407 is the average of the first row (S 1 ) in Table 9. Moreover, the Total column Eigenvector is average for each row using Eq. (5) for finding the total Eigenvector for alternatives student 1 to student 10. Finally, based on the Total column Eigenvector, then ranking order was created where the highest Total Eigenvector score is 0.0778149 for student 20, and the lowest Total Eigenvector score is 0.0283161 for student 18.

5 Calculation Result of SAW and AHP Method After carrying out the stages of the process of calculating all the methods and the results obtained, it is concluded that: • The SAW method has four steps, while the AHP method has two steps. • The SAW method uses two equations while the AHP method uses eight equations where equations Criteria Weight or Eigenvector, as seen in Eqs. (5), (6), and (8), are similar but with different variables. Next are the different stages for the SAW and AHP methods where the SAW method does: • SAW method weighs criteria, ranges of criteria, and normalization of criteria. • SAW method scoring alternatives against criteria using a normalized range of criteria as a matrix. • SAW method normalizes the performance rating score of the matrix • SAW method does ranking score Meanwhile, the AHP method does:

Selection of Human Resources Prospective …

59

Table 9 NPCM for Alternatives based on criteria K 1 Alter native

S1

S2

S3

S4

S5

S6

S7

S8

S9

S 10

S1

0.074

0.074

0.074

0.074

0.074

0.074

0.074

0.074

0.074

0.074

S2

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

S3

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

S4

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

S5

0.074

0.074

0.074

0.074

0.074

0.074

0.074

0.074

0.074

0.074

S6

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

S7

0.074

0.074

0.074

0.074

0.074

0.074

0.074

0.074

0.074

0.074

S8

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

S9

0.074

0.074

0.074

0.074

0.074

0.074

0.074

0.074

0.074

0.074

S 10

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

S 11

0.074

0.074

0.074

0.074

0.074

0.074

0.074

0.074

0.074

0.074

S 12

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

S 13

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

S 14

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

S 15

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

S 16

0.074

0.074

0.074

0.074

0.074

0.074

0.074

0.074

0.074

0.074

S 17

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

S 18

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

S 19

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

S 20

0.074

0.074

0.074

0.074

0.074

0.074

0.074

0.074

0.074

0.074

sum

1

1

1

1

1

1

1

1

1

1

Alter native

S 11

S 12

S 13

S 14

S 15

S 16

S 17

S 18

S 19

S 20

S1

0.074

0.074

0.074

0.074

0.074

0.074

0.074

0.074

0.074

0.074

S2

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

S3

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

S4

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

S5

0.074

0.074

0.074

0.074

0.074

0.074

0.074

0.074

0.074

0.074

S6

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

S7

0.074

0.074

0.074

0.074

0.074

0.074

0.074

0.074

0.074

0.074

S8

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

S9

0.074

0.074

0.074

0.074

0.074

0.074

0.074

0.074

0.074

0.074

S 10

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

S 11

0.074

0.074

0.074

0.074

0.074

0.074

0.074

0.074

0.074

0.074

S 12

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

S 13

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

S 14

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

(continued)

60

A. Rufai et al.

Table 9 (continued) Alter native

S 11

S 12

S 13

S 14

S 15

S 16

S 17

S 18

S 19

S 20

S 15

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

S 16

0.074

0.074

0.074

0.074

0.074

0.074

0.074

0.074

0.074

0.074

S 17

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

S 18

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

S 19

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

0.037

S 20

0.074

0.074

0.074

0.074

0.074

0.074

0.074

0.074

0.074

0.074

Sum

1

1

1

1

1

1

1

1

1

1

• AHP method processes criteria and alternatives differently and does the Pairwise Comparison Matrix (PCM) and Normalized Pairwise Comparison Matrix (NPCM). • PCM for criteria is a process to create a comparison matrix among the criteria scores, while PCM alternatives are a process to create a comparison matrix among the alternatives for each criterion. • NPCM is the percentage of PCM score, and each NPCM has NPCM average and NPCM total score, where the NPCM average is Criteria Weights or Eigenvector score, and the NPCM total score has a score one and score one shows as legitimate NPCM. Specifically for the criteria process, consistency is carried on using Random Consistency Index (RI) to check if the calculation scores are correct. Based on Table 11, six students have similar ranking order such as student 20, student 3, student 19, student 1, student 8, and student 18. Meanwhile, there are similarity SAW method score for student 10 and student 12 with score = 0.685, student 4, student 6 and student 7 with score = 0.585. For similarity consistency, the ranking order is revised to become ranking order 9, 10, and 11 for student 7, student 4, and student 6, respectively. Since student 4, student 6, and student 7 have similar scores and are in rank numbers 9, 10, and 11, respectively, there is another one similarity ranking order for student 4, which makes a total there are seven students with similarity ranking order. The SAW method has 7/20 = 0.35 or 35% similarity consistency. Moreover, there is also a similar AHP method score for student 10 and 12 with a score = 0.0538194. Interestingly, there is four opposite ranking order, such as ranking order 2 and 3, wherein the SAW method has ranking order student 5 and student 9 while the AHP method has opposite ranking order student 9 and student 5. The next opposite ranking order, such as ranking order 11 and 12, wherein the SAW method has ranking order student 6 and student 11 while the AHP method has opposite ranking order student 11 and student 6. Moreover, another opposite ranking order, such as ranking order 15 and 16, wherein the SAW method has ranking order student 14 and student 2 while the AHP method has opposite ranking order student 2 and student 14. Furthermore, another opposite ranking order, such as ranking order 17 and 18, wherein the SAW

0.02857143

0.02857143

0.05714286

0.02857143

0.02857143

0.02857143

0.02857143

0.05714286

0.02857143

0.05714286

0.11428571

0.07407407

0.03703704

0.03703704

0.03703704

0.07407407

0.03703704

0.07407407

0.03703704

0.07407407

0.03703704

0.07407407

0.03703704

0.03703704

0.03703704

0.03703704

0.07407407

0.03703704

0.03703704

0.03703704

0.07407407

1

Student 2

Student 3

Student 4

Student 5

Student 6

Student 7

Student 8

Student 9

Student 10

Student 11

Student 12

Student 13

Student 14

Student 15

Student 16

Student 17

Student 18

Student 19

Student 20

Sum

1

0.02857143

0.02857143

0.05714286

0.02857143

0.08571429

0.02857143

0.08571429

0.05714286

0.11428571

K2

K1

Criteria weights or Eigenvector

Student 1

Alternative

1

0.08928571

0.03571429

0.01785714

0.03571429

0.07142857

0.05357143

0.03571429

0.03571429

0.05357143

0.03571429

0.05357143

0.07142857

0.03571429

0.05357143

0.05357143

0.05357143

0.05357143

0.07142857

0.05357143

0.03571429

K3

1

0.06122449

0.06122449

0.02040816

0.04081633

0.06122449

0.04081633

0.04081633

0.02040816

0.06122449

0.06122449

0.06122449

0.06122449

0.04081633

0.04081633

0.04081633

0.04081633

0.08163265

0.06122449

0.04081633

0.06122449

K4

1

0.07246377

0.05797101

0.02898551

0.04347826

0.04347826

0.05797101

0.05797101

0.01449275

0.08695652

0.04347826

0.08695652

0.07246377

0.02898551

0.04347826

0.04347826

0.05797101

0.04347826

0.02898551

0.04347826

0.04347826

K5

Table 10 Criteria weights or Eigenvector for 20 students (alternatives) upon six criteria

1

0.05555556

0.03703704

0.03703704

0.03703704

0.07407407

0.05555556

0.03703704

0.05555556

0.05555556

0.05555556

0.05555556

0.05555556

0.03703704

0.03703704

0.05555556

0.07407407

0.05555556

0.05555556

0.03703704

0.03703704

K6

1

0.0778149

0.0476878

0.0283161

0.0371091

0.0588085

0.050349

0.0395245

0.0367251

0.0538194

0.0497697

0.0538194

0.0700768

0.0346936

0.0557819

0.0479336

0.0691321

0.0499744

0.0518957

0.0400853

0.0466833

Total Eigenvector

1

13

20

17

4

9

16

18

7

11

6

2

19

5

12

3

10

8

15

14

Ranking order

Selection of Human Resources Prospective … 61

62 Table 11 Ranking order similarity consistency of SAW and AHP methods

A. Rufai et al. Ranking order

SAW method

AHP method

Rank = 1

Student 20 = 0.858

Student 20 = 0.0778149

Rank = 2

Student 5 = 0.827

Student 9 = 0.0700768

Rank = 3

Student 9 = 0.788

Student 5 = 0.0691321

Rank = 4

Student 10 = 0.685

Student 16 = 0.0588085

Rank = 5

Student 12 = 0.685

Student 7 = 0.0557819

Rank = 6

Student 16 = 0.68

Student 10 = 0.0538194

Rank = 7

Student 15 = 0.627

Student 12 = 0.0538194

Rank = 8

Student 3 = 0.588

Student 3 = 0.0518957

Rank = 9

Student 4 = 0.585

Rank = 10

Student 6 = 0.585

Student 4 = 0.0499744

Rank = 11

Student 7 = 0.585

Student 11 = 0.0497697

Rank = 12

Student 11 = 0.565

Student 6 = 0.0479336

Rank = 13

Student 19 = 0.557

Student 19 = 0.0476878

Rank = 14

Student 1 = 0.49

Student 1 = 0.0466833

Rank = 15

Student 14 = 0.482

Student 2 = 0.0400853

Rank = 16

Student 2 = 0.46

Student 14 = 0.0395245

Rank = 17

Student 13 = 0.457

Student 17 = 0.0371091

Rank = 18

Student 17 = 0.44

Student 13 = 0.0367251

Rank = 19

Student 8 = 0.398

Student 8 = 0.0346936

Rank = 20

Student 18 = 0.353

Student 18 = 0.0283161

Student 15 = 0.050349

method has ranking order student 13 and student 17 while the AHP method has opposite ranking order student 17 and student 13. Thus, since there is four opposite ranking order where in this case we reverse the opposite of each ranking order, then it will add the number of similar ranking order with eight students such as student 5 and student 9 in ranking order 2 and 3, student 6 and student 11 in ranking order 11 and 12, student 14 and student 2 in ranking order 15 and 16, student 13 and student 17 in ranking order 17 and 18. Finally, the similarity consistency between the SAW method and the AHP method has a score (7 + 8)/20 = 0.75 or 75% similarity consistency index.

Selection of Human Resources Prospective …

63

6 Conclusions From the discussion and review of the decision support system model for selecting HR candidates, conclusions can be drawn, namely: This decision support system can provide convenience in data collection and calculation of the value of selecting prospective HR students so that producing HR candidates with good competency scores can be selected. The decision support system using the SAW and AHP methods can produce accurate decision-making to improve effectiveness and help the selection process of prospective HR students. Moreover, the experiment shows that the option between using the SAW and the AHP methods has a similarity consistency of 75% as roughly both of them have closed equally result in a score. Interestingly, both results show the same result for the first ranking order 1 for student 20 as the highest-ranking score and the last ranking order 20 for student 18 with the lowest ranking score. The current experiment using 20 students is an initial experiment, and the actual implementation will use a web-based application with Personal Home Pages (PHP) as server programming with MySQL database including using HTML, Code Style Sheet (CSS), and Javascript as client programming.

References 1. S. Andayani, B.H.M. Sumarno, H. Waryanto, Comparison of promethee-topsis method based on SAW and AHP weighting for school e-learning readiness evaluation. J. Phys: Conf. Ser. 1581, 012012 (2020) 2. P. Sutoyo, D. Nusraningrum, Comparative study decision support system AHP and SAW method in tender process TV transmission stations. Dinasti Int. J. Digital Bus. Manage. 1(5), 842–856 (2020) 3. W. Firgiawan, N. Zulkarnaim, S. Cokrowibowo, A comparative study using SAW, TOPSIS, SAW-AHP, and TOPSIS-AHP for tuition fee (UKT). IOP Conf. Ser. Mater. Sci. Eng. 875(1), 012088 (2020) 4. K. Kabassi, C. Karydis, A. Botonis, AHP, Fuzzy SAW, and Fuzzy WPM for the evaluation of cultural websites. Multimodal Tech. Interact. 4(1), 5 (2020) 5. F. Noviyanto, A. Tarmuji, H. Hardianto, Food crops planting recommendation using analytic hierarchy process (AHP) and simple additive weighting (SAW) methods. Int. J. Sci. Technol. Res. 9(2), 4743–4749 (2020) 6. D. Wiguna, Decision support system to determine the location of a cake shop retail business using the AHP method and simple additive weighting (SAW). Systematics 2(2), 79–85 (2020) 7. A.S. Ajrina, R. Sarno, H. Ginardi, A. Fajar, Mining zone determination of natural sandy gravel using fuzzy AHP and SAW, MOORA, and COPRAS methods. Int. J. Intell. Eng. Syst. 13(5), 560–571 (2020) 8. M.F. Ahmed, R.M. Faisal, GIS based modeling of GWQ assessment at Al-Shekhan area using AHP and SAW techniques. Sci. Rev. Eng. Environ. Sci. 29(2), 172–183 (2020) 9. N.N.A.P. Siwa, I.M. Putrama, G.S. Santyadiputra, Development of car rental system based on geographic information system and decision support system with AHP (analytical hierarchy process) and SAW (simple additive weighting) method. J. Phys: Conf. Ser. 1516(1), 012013 (2020)

64

A. Rufai et al.

10. W. Adhiwibowo, B.A. Pramono, S. Hadi, N. Hidayati, Bandwidth management decision support system with a hybrid (SAW and AHP) method, in Engineering, Information, and Agricultural Technology in the Global Digital Revolution, ed. by Hendrawan, W.D. Arifin (Taylor and Francis Group, London, 2020) 11. Y.S. Sari, Decision support system (DSS) for measuring satisfaction of E-procurement service provider using simple additive weighting (SAW) and analytical hierarchy process (AHP). Int. Educ. J. Sci. Eng. (IEJSE) 3(2), 4–8 (2020) 12. A. Khozaimi, Y.D. Pramudita, E.M.S. Rochman, A. Rachmad, Sales quality determination using simple additive weighting (SAW) and analytical Hirarki process (AHP) methods. Jurnal Ilmiah Kursor 10(2), 95–100 (2020) 13. A.W. Murdiyanto, Decision support system of keyword selection web site using analytical hierarchy process (AHP) and simple additive weighting (SAW). Compiler 8(1), 81–93 (2019) 14. D.A. Pertiwi, B. Daniawan, Y. Gunawan, Analysis and design of decision support system in major assignment at buddhi high school using AHP and SAW methods. Tech-E. 3(1), 13–21 (2019) 15. N. Kumar, T. Singh, J.S. Grewal, A. Patnaik, G. Fekete, A novel hybrid AHP-SAW approach for optimal selection of natural fiber reinforced non-asbestos organic brake friction composites. Mater. Res. Express 6(6), 065701 (2019) 16. N.K.Y. Suartini, I.M.A. Wirawan, D.G.H. Divayana, DSS for" E-private" using a combination of AHP and SAW methods. IJCCS (Indonesian J. Comput. Cybern. Syst.). 13(3), 251–262 (2019) 17. R.D. Surbakti, R.S. Simamora, E. Cendana, D. Sitanggang, J. Banjarnahor, M. Turnip, Application selection lending houses subsidized by the method of AHP and SAW. J. Phys. Conf. Ser. 1230(1), 012082 (2019) 18. A. Cahyapratama, R. Sarno, Application of analytic hierarchy process (AHP) and simple additive weighting (SAW) methods in singer selection process, in International Conference on Information and Communications Technology (ICOIACT) (2018), pp. 234–239 19. A.F. Apriliana, R. Sarno, Y.A. Effendi, Risk analysis of IT applications using FMEA and AHP SAW method with COBIT 5, in International Conference on Information and Communications Technology (ICOIACT) (2018), pp. 373–378 20. M.M.D. Widianta, T. Rizaldi, D.P.S. Setyohadi, H.Y. Riskiawan, Comparison of multi-criteria decision support methods (AHP, TOPSIS, SAW & PROMENTHEE) for employee placement. J. Phys: Conf. Ser. 953(1), 12116 (2018) 21. W. Abrams, E. Ghoneim, R. Shew, T. LaMaskin, K. Al-Bloushi, S. Hussein, F. El-Baz, Delineation of groundwater potential (GWP) in the northern United Arab Emirates and Oman using geospatial technologies in conjunction with simple additive weight (SAW), analytical hierarchy process (AHP), and probabilistic frequency ratio (PFR) techniques. J. Arid Environ. 157, 77–96 (2018)

Implementation of the Weighted Product Method to Specify scholarship’s Receiver Chintia Ananda, Diana Teresia Spits Warnars, and Harco Leslie Hendric Spits Warnars

Abstract Scholarship assistance for children of educational age, especially during the COVID-19 pandemic, has provided some relief for some families affected by the economic turmoil during the COVID-19 pandemic, where some parents were laid off, or their parents were reduced in salary. This scholarship at least helps students to continue their dreams of obtaining and continuing the education they are pursuing so that in the end, they can continue their dreams for a brighter future for the welfare of themselves, their families, and the nation. However, on the other hand, new problems arise with the limited number of scholarships awarded and the manipulation of student data to obtain these scholarships. Therefore, this paper discusses the implementation of decision-making applications in scholarships using the weighted product method that underlies the awarding of scholarships with four criteria: parents of students participating in the family program of hope, orphan status, and several dependents of parents, and parents’ income. Keywords Decision support systems · Weighted product (WP) · Information systems

1 Introduction In this life, humans are often faced with moments where they have to make decisions, and one of them is when they realize that education is critical where getting a proper education and following the wishes of the individual will at least increase his dignity C. Ananda · D. T. S. Warnars Information Systems Department, School of Information Systems, Bina Nusantara University, Jakarta 11480, Indonesia e-mail: [email protected] D. T. S. Warnars e-mail: [email protected] H. L. H. S. Warnars (B) Computer Science Department, Graduate Program, Doctor of Computer Science, Bina Nusantara University, Jakarta 11480, Indonesia e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 I. J. Jacob et al. (eds.), Expert Clouds and Applications, Lecture Notes in Networks and Systems 444, https://doi.org/10.1007/978-981-19-2500-9_5

65

66

C. Ananda et al.

as a human being and besides that to improve the human capital of the nation. Furthermore, the state has an interest and responsibility to facilitate its citizens to get proper education according to the tastes and abilities of each community, and besides that, considering the number of Indonesian people is too large, so that educational activities are also given to the community as private education sector to provide education as high as—height. However, as humans, we are inseparable from human destiny, which is determined by the creator and the human behavior itself so that, in the end, it impacts the education that is sought, especially in society. Moreover, children who receive education whose funds come from their family or parents cannot continue their current education at the end of the day. The problem of layoffs and reduced working hours due to working from home online during the COVID-19 pandemic also reduced income, which is usually the income partially allocated to continue the education. Based on these problems, this is where there are scholarships to help with problems in financing these educational activities. Scholarships can be defined as allocating funds that do not come from private funds but funds from the government, companies, and educational foundations [1]. In addition, scholarship funds can be provided employing exemption from tuition fees categorized as a substitute for work costs or official ties that will be carried out after the scholarship recipient carries out the education period. Scholarships are submitted to recipients entitled to receive them and are determined by several criteria stipulated by each scholarship granting organization. Scholarships are awarded according to the interests of each scholarship provider who looks more at the institution’s long-term goals as stated in the vision and mission of an organization. For example, if it is a religious organization, it will also look at religious similarities, and if it is a local government organization, it will prioritize giving scholarships to students from which the local government comes. In this paper, the implementation will focus on SMA Negeri 8 Serang City in Banten Province, which is a government-owned school, and based on semester 2020/2021 data, and it has a total of 54 teachers with 823 male students and 1,337 female students in 30 classes with the student–teacher ratio at 40. So far, SMA Negeri 8 Kota Serang has experienced difficulties because a manual process still carries out scholarship selection, so it takes a very long time. Based on the observation of the scholarship acceptance process, scholarship applicants only submit the scholarship registration requirements documents, and from these files, a meeting will be held to compare prospective students manually. Based on the meeting results and discussion at the teacher meeting, it was announced which students were entitled to receive scholarships through wall magazines. Based on the results of interviews, observations, and group discussion forums with teachers, students, and parents, it is concluded that the current scholarship selection process is very ineffective and unfair in determining scholarship recipients. There are accusations of nepotism where the scholarship recipients come from the families of teachers and employees, even though recipients of scholarships from teachers or employees are entitled, intelligent, and deserving of scholarships. Seeing the problems mentioned above, it is essential to make a system supported by the application of information technology, where the decisions carried out are based

Implementation of the Weighted Product Method …

67

on the output of a computer program. It is hoped that this decision-making system can help principals and teachers determine which students are eligible for scholarships objectively and avoid improper scholarship distribution. Also, it avoids accusations of nepotism among teachers and school employees in giving scholarships.

2 Current and Previous Similar Research Papers To look at current and past similar research, we used scholar google to search for the term “decision support system,” which was restricted to papers between 2017 and 2021 and produced 52,900 papers. In addition, we also restricted searches to terms like “decision support system” + scholarships for papers between 2018 and 2021, so there were 6580 results. Besides, we also expanded our search for papers between 2017 and 2021 by typing in the term “decision support system” + “weighted product”, resulting in 938 papers and continued using the term “weighted product” + scholarships with 313 results papers and continued using next term such as “decision support system” + “weighted product” + scholarships with 82 results papers obtained. However, after going through each search result, not all of the generated titles have the results as mentioned in the text search, so the value is not as many as the number of searches generated. In this literature review section, to know the current and five previous years, a similar topic was using the weighted product (WP) method for decision support systems. This WP method has been applied in many areas such as education, health systems, agriculture, restaurant and hotel, housing, office, and business, etc. In this paper, the literature review is limited to papers in education, housing, culinary systems, offices, business, health systems, and agriculture. Meanwhile, The Weighted Product (WP) is a multiplication weighting method for combining attribute ranks, and the rank for each attribute is incremented according to the associated attribute value, and this process is sometimes also called the normalization process. The WP algorithm has three stages: first, determining the value of the weight W, namely weighting, the second determining the value of the vector S, namely lifting, and the third determining the value of the vector V as a rating. A desktop application evaluates the effectiveness level of digital libraries. In the education system, the decision support system as a decision support tool by top management has been implemented in every field of education, such as using the WP method combined with the Alkin model for evaluating educational services in a computer college library implemented as a program. The implementation uses the Borg & Gall development model, including interviews and questionnaires as user requirement tools [2]. Moreover, WP has been implemented to determine exemplary high school students using nine criteria: average report card score, ranking, number of absences, competition participation, active extracurricular activities, extracurricular positions, discipline, morals, and accumulated point of violation [3].

68

C. Ananda et al.

Decision-making applications built with the WP method determine which students are most eligible for a scholarship using four criteria as Average Report Card, Attendance, Attitude, and Extracurricular activities [6]. Moreover, a Decision Support System was applied using the WP method to search Achieving Student [4] where one paper used five criteria such as Average class score, Discipline, Attendance, Extracurricular, and Non-Academic [5]. Another implementation was built by creating a cellular application with an android studio and Java development kit, which used the WP method to select the most suitable university in Pekanbaru, which applied three criteria: distance, accreditation value, and several lecturers [7]. In addition, in another paper, determining which students are most eligible for scholarships is also carried out using the WP method by applying 14 criteria combined with the K-Nearest Neighbor (KNN) method to calculate as a result of the final recommendation for scholarship recipients [8]. Another implementation is the MultiObjective Optimization Method based on Ratio Analysis (MOORA) to determine scholarship recipients using eight criteria, namely the value of parents’ income, number of dependents, academic grades, diploma achievement, monthly electricity bills, home status, type of house, and the value of the interview results [9]. Applications Another implementation also uses the MOORA method to determine scholarship recipients for high school students applying the same eight criteria [10]. Another implementation for determining high school student scholarship recipients uses the Simple Additive Weight (SAW) method, which applies seven criteria, namely the age range, the amount of parents’ income, the number of dependents of the parents, the number of siblings, the average score of report cards, the activities in organizations, and distance of residence [11]. Whereas in the type of paper that discusses housing and culinary, the WP method has been applied to various advantages in making decision support system applications such as: selecting the best choice of cooking products suitable for use by a housewife as part of the ingredients for serving food at the table and not. It is also for chefs and restaurant owners when choosing which cooking products are suitable for serving in their restaurant [12]. Also, a mobile application with the WP method is used to determine the culinary that best suits the pockets of culinary connoisseurs consumers in the city of Kudus, Central Java, by using five criteria, namely menu variations, affordable prices, availability of wifi, availability of chargers, and easy distance to travel [13]. The WP method is also applied to applications for the granting of industrial permits for household businesses issued by the health office in ensuring healthy and quality food for the community by using five criteria, namely the production method used, the type of food sold, the product packaging used, and has participated in preparation counseling. Healthy and clean food and the following is completing the profile file [14]. In a paper that discusses the application of the WP method in the office and business area, there are several applications combined with other methods. For example, the WP method is applied to find the best employees based on 13 criteria such as teamwork, craft, applying superior instructions, working diligently, responsible, understanding the job, achieving work, having an average working speed, multiskills, attendance, no information, often leaves the workplace and complies with

Implementation of the Weighted Product Method …

69

company regulations [15]. The selection of other best employees is applied using the WP method for the Pringsewu district revenue office, Yogyakarta province, using five criteria: employee attendance, employee behavior, employee experience, employee discipline, and employee teamwork [16]. Besides, other implementations combine the WP method with the El Chinix Traduisant La Realite (Electre) method for employee recruitment using three initial criteria, namely age range, educational background, and experience, and the second 4 criteria if employees are to be interviewed, namely psychological scores, ability values, and skills, TOEFL scores, and interview scores [17]. Besides that, a Simpeunan Savings and Loans Cooperative in the city of Tasikmalaya uses the weighted product method and the Simple MultiAttribute Rating (SMART) in determining creditworthiness by using five criteria such as the amount of savings, total income, character rating, a form of guarantee, and condition of guarantee [18]. Furthermore, the k-means grouping is combined with the WP method to map crime-prone locations that assist security forces and the public in controlling crime in the Kudus Regency by using three criteria, namely the number of crime incidents, the number of houses in an area, and the distance to the scene and the police station [19]. There are currently three papers in the health system that discuss using the WP method for their approach, one of which is to create a decision-making system to provide information to pregnant women in paying attention to nutritious foods for their babies who pay attention to the composition of the food they eat, which consists of 5 criteria such as vegetables, fruit, meat, nuts, milk, etc. [20]. After that, another paper discusses creating a system to help decision-making at the GrandMed Hospital in the city of Deli Serdang for the selection of suppliers or drugstores that will provide medicines to patients by paying attention to 4 criteria such as price worthiness, delivery performance, drug variations and delivery distance [21]. Another paper discusses an expert system that helps determine the types of diseases babies suffer that pay attention to 7 criteria for baby health conditions, namely excessive crying, fever conditions, diarrhea, skin problems, weight gain, height growth, and communication development [22]. In agriculture, two recent papers use the WP method approach, namely a paper that discusses systems to help chili farmers in determining the success of planting chilies by paying attention to 4 criteria, including the condition of the soil elevation above sea level in meters, the measure of soil pH (Power of Hydrogen), nutritional value, and Ambient Temperature in Celsius [23]. Furthermore, the following paper discusses creating a system to assist rice farmers in determining rice varieties to be planted by considering five criteria: potential success, average yield, harvest time, resistance to brown planthoppers, and resistance to bacterial leaf blight [24].

3 Result and Discussion In this paper, the discussion will be carried out in 3 stages: system modeling, Weighted Product implementation, and system implementation (Fig. 1).

70

C. Ananda et al.

Fig. 1 Use case diagram of the proposed application

3.1 System Modeling In modeling this system, the system is modeled with two diagrams in the Unified Modeling (UML) language, namely use case diagrams and class diagrams, where the use case diagram shows the process in the application being built, while the class diagram shows the database design model to be used. There are four use case diagram activities in the use case diagram, namely registration, login, entering student data, and running the results. When users want to use the system, they must enter their data such as username, name, gender, email, date of birth, and password as seen in table “User” in the class diagram in Fig. 2. In the use case activity login, the user must first enter the username and password that they have previously registered, and the system will check into the database if their entries match the data in the user table. If it does not match, the user cannot enter the system, and there will be a confirmation that the username or password does not match. In the use case entry student data activity, the user can enter the number of students to be assessed by entering the student code attributes and students names, and the data will be stored in the database as an alternative table as seen in the class diagram in Fig. 2. The entry process will include the scoring criteria for each student, which will be saved in the scoring table as seen in the class diagram in Fig. 2. In this use case activity, there are two sub-use case activities which are drawn with the symbol include, which means that the subcase must be carried out, and two sub-use case USer IdUser UserName Name Gender Email DateofBirth Password

Alternav e IdStudent Name Alternave

Scoring 1

1..* IdStudent IdCriteria Score S_score

Fig. 2 Class diagram of the proposed application

Criteria 1..* 1

IdCriteria NameCriteria Weight Percentage

Implementation of the Weighted Product Method …

71

activities include entering the criteria that will be used to determine the predictions of students who will receive the scholarship and the weight that will be applied to any given criterion. The process of entering criteria and weights will be stored in the criteria table as shown in the class diagram in Fig. 2. When entering the criteria, the criteria are entered, and the percentage attributes in the table will be generated automatically based on the average total weight. Finally, the use case activity running the result will be used to run the result based on the inputted data in previous use case activities such as student data, criteria, and weight. Figure 2 shows the application class diagram, which contains four classes into four database tables: user, alternative, scoring, and criteria, wherein, in this case, the user table is not related to other tables. An alternative table is a table that stores student data that has 1 to many relationships with the assessment table. Besides that, the assessment table has a multi-to-one relationship with the criteria table, or in other words, the alternative table has a many-to-many relationship with the criteria table.

3.2 Weighted Product Implementation The Weighted Product (WP) method has four steps such as 1. 2. 3. 4.

Determination of criteria, criteria name, weight, and range of criteria. Finding the percentage of criteria weights. Finding preference value for each alternative. Finding ranking score.

The first step is determining the criteria, name of criteria, weight, and range of criteria as shown in Table 1 with four criteria, namely the involvement of parents in the family hope program, status as an orphan, the number of dependents of the parents, and parents earning with a weighted score. 4, 4, 3, and 5 with a total weight score of 4 + 4 + 3 + 5 = 16, respectively. These four criteria and the weighting composition Table 1 Equalization of criteria weights Criteria

Criteria name

Weight

Range of criteria

Percentage of criteria weights

C1

Involvement of parents in the hope family program

4

1, 2

0.25

C2

Orphans

4

1, 2

0.25

C3

Number of dependents of parents

3

1–5

0.1875

C4

Parents earning

5

1–5

0.3125

Total

16

1

72

C. Ananda et al.

of each of these criteria were obtained from the results of interviews and discussion group forums with schools represented by teachers and students represented by parents of students. Moreover, as seen in the fourth column in Table 1, the range of criteria for each criterion is also based on a forum discussion group between teachers and parents. 1.

2.

3.

4.

The first criteria, namely “parental involvement in the family hope program,” is an activity where parents must be registered in the family hope program organized by the government as a sign of underprivileged families, and a score of 1 means not registered while a score of 2 means registered. The second criteria are the provision of an orphan status which has two criteria choice scores, namely 1 and 2 where the score 1 indicates that a student has only one parent, either mother or father, while score 2 indicates that he is a student who does not have both father and mother. The third criteria are the number of dependents of parents who have five criteria of choice scores, where the choice scores of 1, 2, 3, 4, and 5 indicate the parents have 1, 2, 3, 4, and > 4 dependents, respectively. The fourth criteria are earning of parents who have five criteria of choice scores, where the choice scores of 1, 2, 3, 4, and 5 indicate parents have income > 4 million, 3–3.99 million, 2–2.99 million, 1–1.99 million, and < 1 million earning, respectively.

The second step is to find the percentage of criteria weights (PCWj) as shown in the last column of Table 1, which uses Eq. (1), where the weight of each criterion (W j) is divided by the total weight criteria (W j) or sigma W j and in this case 4 + 4 + 3 + 5 = 16. For example, the criteria C1 as shown in the first row of Table 1 has PCWj = W j/W j = 4/16 = 0.25. The next criteria are also processed with Eq. (1) and have scores such as 0.25, 0.1875, and 0.3125, respectively, and the percentage of criteria weights has a total of 0.25 + 0.25 + 0.1875 + 0.3125 = 1. wj PCWj =  wj

(1)

where: PCWj = Percentage of criteria weights W j = Weight of each criterion W j = Total Weight Criteria (Sigma weight criteria) j = Criteria The third step is to find the preference value of each alternative (S i ) shown in the Total S i column in Table 2, which contains 21 students/alternatives (i) that have been composed with four criteria (j) in Table 1, where the assessment of each criterion for each student (Xij) was made based on the results of filling in the registration form scholarships filled by students and their parents. Meanwhile, the W j is applied as rank positive for the benefit criteria and negative value for cost criteria, and in this paper, C1 criteria as cost criteria while others as benefit criteria.

Implementation of the Weighted Product Method …

73

Table 2 Matrix of a combination of alternatives and criteria, a score of total S i and V i No.

Name

Alternative

C1

C2

C3

C4

Total S i

Vi

1

Andini Putri

A1

1

1

5

2

1.6793018

0.053637

2

Sentanu Eka Prasetyo

A2

2

1

1

2

1.0442738

0.033354

3

Iin Sumiati

A3

2

1

1

2

1.0442738

0.033354

4

Alexander HB

A4

2

1

1

3

1.1853399

0.037859

5

Mar Ah

A5

2

1

1

3

1.1853399

0.037859

6

Jumanah

A6

2

1

1

3

1.1853399

0.037859

7

Munihah

A7

2

1

3

4

1.5934795

0.050895

8

Siti Fatonah

A8

2

1

2

3

1.3498516

0.043114

9

Yanah

A9

1

2

5

2

1.9970376

0.063785

10

Lilis

A10

2

1

3

3

1.4564753

0.046519

11

Hilman

A11

1

1

4

2

1.6104903

0.051439

12

Ahmad Riyan Rifai

A12

2

1

2

4

1.4768261

0.047169

13

Astriyah

A13

1

1

5

1

1.3522496

0.043191

14

Irmawati

A14

1

1

3

2

1.5259212

0.048738

15

Aryadi Wijaya

A15

1

1

4

3

1.828044

0.058387

16

Nazwa Marisa Awalia

A16

1

1

3

3

1.7320508

0.055321

17

TB Aldiron

A17

1

1

4

3

1.828044

0.058387

18

Muhamad Fiqi

A18

2

1

4

4

1.6817928

0.053716

19

Deviani

A19

2

1

2

2

1.1892071

0.037983

20

Siti Irnawati

A20

2

1

4

4

1.6817928

0.053716

21

Sunariah

A21

2

1

4

4

1.6817928

0.053716

31.308925

1

Total

For example, in the alternative A1, a student Andini Putri, as shown in the  named w first row of Table 2, has a score of Si = nj = 1 X ij j , where S 1 = (X 11 −W 1 ) (X 12 W 2 ) (X 13 W 3 ) (X 14 W 4 ) = (1–0.25 ) (10.25 ) (50.1875 ) (20.3125 ) = 1 * 1.3522496 * 1.2418578 * 1.6793018 = 0.053637. In this case, the rank for the first criteria C1 is given negative rank (X 11 −W 1 ) since C1 criteria as cost criteria, and other criteria such as C2, C3, and C4 are given a positive rank (X 12 W 2 ) (X 13 W 3 ) (X 14 W 4 ) because they are benefit criteria. Other alternatives are also applied using Eq. (2), and each has its own Si result, as shown in the Total S i column in Table 2. Si =

n j=1

where: S i = preference value for each alternative

w

X ij j

(2)

74

C. Ananda et al.

i = alternative j = criteria n= number of criteria n j=1 = Pi (product) of criteria (j) or multiplication of criteria (j) X ij = assessment of each criterion for each alternative W j = percentage of criteria weights w X ij j = X ij to the rank of W j The fourth step as the lastw step is to find the ranking score using Eqs. (3) or (4), where the attribute nj = 1 X i j j in Eq. (3) is the attribute S i in Eq. (2), which is the    wj n attribute of the divisor divided by the total S i j = 1 X ij , which is 31.308925 as shown in Table 2. Therefore, the equation V i , as shown in Eq. (4), is a representation   si and of the value of S i in Eq. (2) which is divided by the total value of S i as shown in the lower right corner of Table 2, the total of all values V i for 21 students/alternatives is 1. n wj j=1 X ij Vi =  n (3) wj j=1 X ij Si Vi =  si

(4)

Based on Eq. (3) or (4), the highest ranking score is the 9th student named Yanah with V i score = 0.063785, and the lowest ranking score is the second and third students named Sentanu Eka Prasetyo and Iin Sumiati with similar V i score = 0.033354. This score ranking indicates that the student with the highest ranking score is the top priority for receiving the scholarship, while the student with the lowest ranking will be the last to receive the scholarship. Figure 3 is the graphic for the matrix combination of alternatives and criteria, as seen in Table 2. Figure 3 shows conformity with what is described in Table 2 before where a student named Yanah has the highest score while students named Sentanu Eka Prasetyo and Iin Sumiati have the lowest score.

3.3 System Implementation In this implementation system, the user interface menu will be displayed as seen in Figs. 4, 5, 6, 7 and 8, representing the implementation of the system modeling described using use cases and class diagrams in Figs. 1 and 2. This implementation system is carried out using personal home pages (PHP) as a web server programming and using the MySQL database to store tables modeled in Fig. 2. Figures 4, 5, 6, 7 and 8, are the user interface for the application, which can help the user to understand and interact with the application developed using personal

Implementation of the Weighted Product Method …

75

2.5 2 1.5 1 0.5

C2

C4

Sunariah

Deviani

Si Irnawa

TB Aldiron

Muhamad Fiqi

Aryadi Wijaya

Nazwa Marisa Awalia

Astriyah Si

Irmawa

Ahmad Riyan Rifai

Lilis C3

Hilman

Yanah

Munihah C1

Si Fatonah

Mar Ah

Jumanah

Iin Sumia

Alexander HB

Andini Putri

Sentanu Eka Prasetyo

0

Vi

Fig. 3 Graphic of matrix of a combination of alternatives and criteria

Fig. 4 Main menu user interface

home pages (PHP) web-based and MySQL database. Figure 1 shows the use case diagram, which contains some activities such as registration, login, entry of student data, and running the result. Especially for entry, student data use case has two subuse case activities: entry criteria and entry weight. Entry student data is the entry in the application regarding alternative as data of students which content with some activities as field column.

76

C. Ananda et al.

Fig. 5 Login menu user interface

Fig. 6 Menu entry criteria and weight

Meanwhile, Fig. 4 is the main menu that shows the application’s content, which includes all submenus as seen in the use case diagram in Fig. 1. Figure 5 is the login menu where the user should enter a username and password, and on this page, if a new user, then needs to register by entering user data such as email, name, address, gender, date of birth, username, and password. After the registration, the email confirmation will come to the user’s email address and confirm to enter the system by entering with username and password as seen in Fig. 5. Moreover, Fig. 6 shows the entry of student data as an alternative such as name and the content of criteria is input including the weight for each of the criteria. Figure 6 shows the user interface for use case activity entry student data include sub-use case activities such as entry criteria and weight, as seen in Fig. 1. Furthermore, Figs. 7

Implementation of the Weighted Product Method …

77

Fig. 7 Form selection vector S

Fig. 8 Form selection vector V

and 8 show the process of running the result use case as seen in Fig. 1, whereafter the user pushes the button, then the application will automatically apply and run the Eqs. (1), (2), and (3) on the application. Figure 7 shows the running of the S i vector using Eq. (2), while Fig. 8 shows the running V i vector using Eq. (2). Meanwhile, Fig. 8 is the print screen where the use case running the result in Fig. 1 is run, and then it will show the content result respectively based on the student data

78

C. Ananda et al.

input as data alternatives and criteria as condition filtering for finding the best result, the ranking score.

4 Conclusion Based on the results of research conducted in making the application of the decisionmaking system to determine the prospective scholarship recipients who cannot name in SMA Negeri 8 Serang City, the results of the research conclusions obtained are as follows: 1.

2.

3.

To determine the prospective scholarship recipients can no longer do it manually. The existence of this decision-making system can help the school determine that the scholarship recipients cannot accurately target all of the students of Serang City 8 High School and no longer require a long time. A ranking decision in Serang City 8 High School becomes more optimal by applying the decision-making system using the weighted product method to make it easier for schools and application users to determine the highest weighting value of several students on the list of potential scholarship recipients. Using four criteria determined using the Ministry of Education and Culture indicators with the Ministry of Education and Culture regulation of the Republic of Indonesia Number 6 of 2016. The application of web-based decision support systems in SMA Negeri 8 Kota Serang can be realized.

References 1. B. Subaeki, M. Irfan, R.S. Adipradana, C.N. Alam, M.A. Ramdhani, Decision support system design of higher education scholarship recipients with android-based. J. Phys: Conf. Ser. 1280(2), 022016 (2019) 2. D.G.H. Divayana, I.M. Ardana, I.P.W. Ariawan, Implementation of educational services evaluation application based on weighted product-Alkin as a digital library evaluation tool. J. Talent. Dev. Excellence 12(1), 1798–1812 (2020) 3. S.R. Arifin, R.H. Pratama, ımplementation of weighted product (wp) methods in decision support system for determining exemplary students. Positif: J. Sistem Dan Teknologi Informasi 6(1), 76–84 (2020) 4. D. Rahayu, S. Mukodimah, Decision support system of achieved students using weighted product method. Int. J. Inf. Syst. Comput. Sci. (IJISCS) 3(2), 72–77 (2019) 5. R.R. Mohamed, M.A. Mohamed, D. Rahayu, W. Hashim, A. Maseleno, Decision support system of achieving student using weighted product method. Int. J. Psychoso. Rehabil. 23(4) (2019) 6. R.S. Septarini, R. Taufiq, S. Al Fattah, The implementation of weighted products in the support system of scholarship acceptance decisions at the MA AL-Falahiyah AL-Asytari. Jurnal Informatika Universitas Pamulang 5(4), 438–444 (2021)

Implementation of the Weighted Product Method …

79

7. D. Nasien, M.H. Adiya, A. Mulyadi, A. Sukabul, G. Rianda, D. Yulianti, Decision support system for selecting university in Pekanbaru based on android. Int. J. Electr. Energy Power Syst. Eng. 3(2), 30–34 (2020) 8. L.A. Nasher, N. Bahtiar, Application of decision support system using the K-nearest neighbor and weighted product method for determining the recipients of low-income family scholarship (GAKIN) (case study: Poltekkes Kemenkes Semarang). J. Phys.: Conf. Ser. 1217(1), 012117 (2019) 9. A. Utami, E.L. Ruskan, Development of decision support system for selection of yayasan alumni scholarship using moora method, in Sriwijaya International Conference on Information Technology and Its Applications (SICONIAN 2019) (Atlantis Press, 2020), pp. 706–710 10. R. Mardhiyyah, R.H.P. Sejati, D. Ratnasari, A decision support system of scholarship grantee selection using Moora. Int. J. Appl. Bus. Inf. Syst. 3(1), 21–27 (2019) 11. T. Susilowati, S. Suyono, W. Andewi, Decision support system to determine scholarship recipients at SMAN 1 Bangunrejo using SAW method. Int. J. Inf. Syst. Comput. Sci. (IJISCS) 1(3), 56–66 (2017) 12. S.S. Goswami, D.K. Behera, S. Mitra, A comprehensive study of weighted product model for selecting the best product in our daily life. Braz. J. Oper. Prod. Manage. 17(2), 1–18 (2020) 13. A.K. Wardhani, E. Lutfina, Application culinary decision support system in Kudus city with weighted product method based on mobile phone. J. Comput. Sci. Eng. (JCSE) 1(1), 10–16 (2020) 14. G.M. Putra, Application of household ındustrial certificate license using weight product method, in International Conference on Social, Sciences and Information Technology, vol. 1(1), pp. 61–70 (2020) 15. S.A. Susanto, E. Prasetyo, M.M. Hidayat, P. Setiawan, Applying the weighted product method for the best selection of personal quality control in Pt. Pacific Equinox Surabaya. J. Electr. Eng. Comput. Sci. 5(1) (2020) 16. N. Aminudin, E. Sundari, K. Shankar, P. Deepalakshmi, R.I. Fauzi, A. Maseleno, Weighted product and its application to measure employee performance. Int. J. Eng. Technol. 7(2.26), 102–108 (2018) 17. M. Irfan, U. Syaripudin, C.N. Alam, M. Hamdani, Decision support system for employee recruitment using El Chinix Traduisant La Realite (Electre) and weighted product (Wp). Jurnal Online Informatika 5(1), 121–129 (2020) 18. A. Ilah-Warnilah, I.N. Hawa, Y.S. Mulyani, The analysis of determining credit worthiness using weighted product and smart methods in SPB cooperatives. Indonesian J. Inf. Syst. 2(2), 140–151 (2020) 19. Y. Rahmatika, E. Sediyono, C.E. Widodo, Implementation of K-means clustering and weighted products ın determining crime-prone locations. Kinetik: Game Technol. Inform. Sys. Comput. Netw. Comput. Electron. Control 5(3), 195–202 (2020) 20. E. Darnila, Application of decision support system for ınfant nutrition women using weighted product on health. Login: Jurnal Teknologi Komputer 14(1), 25–31 (2020) 21. D. Fanita, B. Sinaga, Supplier selection decision support system drug weighted methods product (WP). J. Comput. Netw. Archit. High Perform. Comput. 2(1), 135–139 (2020) 22. M. Muslihudin, G. Devika, P. Manickam, K. Shankar, D.P. Putra, E.P.S. Putra, A. Maseleno, Expert system in determining baby disease using web mobile-based weighted product method. Int. J. Recent Technol. Eng. 8(1), 3299–3308 (2019) 23. R. Ramadiani, B. Ramadhani, Z. Arifin, M.L. Jundillah, A. Azainil, Decision support system for determining chili land using weighted product method. Bull. Electr. Eng. Inform. 9(3), 1229–1237 (2020) 24. G.R. Cahyono, J. Riadi, I. Wardiah, Decision support system for the selection of rice varieties using weighted product method. J. Phys. Conf. Ser. 1450(1), 012059 (2020)

A Survey on E-Commerce Sentiment Analysis Astha Patel, Ankit Chauhan, and Madhuri Vaghasia

Abstract Sentimental Analysis for products and services available on various ecommerce website and applications has been an important and crucial research task in current era. With the ease of e-commerce and m-commerce every seller tries to show their service and products by promoting on various platforms. Opinion mining, as well referred as sentiment analysis, is a significant element in natural language processing. It is used to evaluate what individuals or audiences consider about the products and services currently offered on collective media channels or social media platforms or e-commerce sites. To detect sentimental polarity a better method should be chosen. We have reviewed some of the research work, those have been tested and proven as good research work for sentimental analysis, as reviews of any products are available on website or applications related to products or services. But it is difficult to find out sentiments when large number of reviews from various sources are collected. You can find different reviews from different sources. There are lakhs of products reviews and services available on various e-commerce portal. To check manually review is difficult task. An automated review system without bios is need of the time. Keywords Machine learning · E-commerce · Review · Sentimental analysis · Support vector machine · Naïve Bayes · Opinion mining · Natural languange processing

A. Patel (B) · A. Chauhan · M. Vaghasia Parul Institute of Engineering and Technology, Vadodara, India e-mail: [email protected] A. Chauhan e-mail: [email protected] M. Vaghasia e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 I. J. Jacob et al. (eds.), Expert Clouds and Applications, Lecture Notes in Networks and Systems 444, https://doi.org/10.1007/978-981-19-2500-9_6

81

82

A. Patel et al.

1 Introduction In the era of e-commerce and m-commerce, online shopping for various products as well as various services has been booming. Recent time of lockdown due to covid in various countries and areas also boosted up online shopping. Users’ comments on various platform for various products and services has been an important source of feedback. It has been increasing day by day. People can easily share their views pros and cons of products as well as for sellers and selling platforms also. With the advancement of various technological changes and use of the computer and mobile world, the end user or individuals are using social media platforms to state their opinions and thoughts about products and services in an unorganized (unstructured) approach. Opinions expressed on social media channels such as for example, Facebook can be classified to describe all other forms of positive, negative and neutral evaluations of the given text in reviews. It is challenging task to analyze sentiments from lots of reviews from various sites for so many products. Machine Learning can play an important role for the same. In this survey paper we have reviewed some of the research work that has been carried out for the Sentimental Analysis for e-commerce. The main challenges for sentiment analysis of e-commerce and mobile commerce is Dimension Mapping (mapping with other contents) and Double Meaning of Words (ex. ‘which’). The dimension mapping problem primarily refers to the mapping of opinionated text words (containing words) with the correct dimensions of the content. The sentiment of word disambiguation problem is similar to the scenario in which a sentiment word can be linked with two or more content dimensions. As reviews are in unstructured data, for research purpose structural data is required. Using NLP, it is extremely possible for computers and devices to read text or hearing speech, interpreting it, measure sentiments of the content and also determine which parts are important. NLP in sentimental analysis plays a crucial role in generating dataset and extracting informational data of textual data. Data Processing stage in sentimental analysis is performed by mainly NLP. Final step in sentimental analysis is classification. A better classification is applied to classify the review in classes like positive, negative or neutral.

2 Related Works As work related to sentimental analysis is been reviewed, various researchers have used various techniques. Each technique has its own pros and cons. We have also discussed strong features and limitations of the same work. In research [1] SVM Algorithm is applied to generate a model for sentimental analysis. In this research Amazon dataset is used for research work. POS (Part of the Speech) technique is used for this research. Along with this, negation phase identification is also used to detect words like “not good”, “not bad” etc. Tokenization, Removing Stop Words, POS Tagging and Stemming are included in Data Processing. Sentiment classification of information, also known as polarity categorization (PC),

A Survey on E-commerce Sentiment Analysis

83

Fig. 1 General System Flow [1]

is the process of discovering or detecting and naming (classifying) a particular viewpoint depending in terms of preference (positive, negative or neutral) application. During their research, they used the Natural Language Toolkit (NLTK) system as well as the Scikit-learn (scilib) Python library for SVM implementation (Fig. 1). Another study [2] employed the PySpark platform and a resilient distributed dataset (RDD) based sentiment analysis utilizing the Spark NLP tool to solve scalability and availability challenges in sentiment analysis (SA) on a conventional ecommerce platform. They likewise utilized Python’s Scrapy framework for web scraping to extract important data from e-commerce websites (Fig. 2). In another research work [3] they have applied Amazon based customer review dataset. The emphasis is mostly on discovering various aspect phrases from each review in the dataset, determining the parts of speech (POS) and using particular categorization algorithm to assess the positivity, negativity and neutrality of each review. They have applied three levels of sentimental analysis. Document level sentiments: The entire document instance must be processed at once at the document level sentiment analysis stage. While using the sentence level technique sentiment analysis, the research information must be separated into sentences. Subjectivity categorization refers to the process of dividing a document into sentences. The primary goal of

84

A. Patel et al.

Fig. 2 FLASK based SA system [2]

Table 1 Result analysis of aspect level [3]

Parameter

Naïve Bayes

SVM 83.423

Accuracy

90.423

F-Measure

0.952

0.841

Precision

0.947

0.852

Recall

0.959

0.83

this stage of analysis in their research is to determine the features of a product. By applying Naïve Bayes classifier, they have achieved 90% accuracy (Table 1). In research [4] it was studied that in their research work they have applied supervised learning method for sentimental analysis. They have extracted large scale dataset from Amazon. They have tested their work over 48,500 products. They have mostly used electronics category and musical instrument category for their research work. Figure 3 explains workflow for large scale sentimental analysis. In their research they extracted data, preprocessed the same data and applied feature extraction. After that supervised learning method is applied and classification is done. In research [5] researchers applied sentimental dictionary approach. Researchers have applied emotional resources for building sentimental dictionary. Text emotion analysis is done with machine learning algorithms (Fig. 4). As another research [6] researchers manually assess product reviews by gathering data in the form of an excel file. Feature extraction matrix (FEM) is applied in this research. Product recommendations and a list of products based on the greatest value of FEM for searched features are generated depending on the user searched feature. In research, the Support Vector Machine as well as Naive Bayes approach have produced good and competent outcomes [7]. According to them, the Part of Speech (POS) principal boosts sentiment potency. As a result, integrating POS with SVM is useful for sentimental analysis in e-commerce.

A Survey on E-commerce Sentiment Analysis

85

Fig. 3 Workflow for supervised learning [4]

In another research [8] SLCABG is applied. The study presented in this paper is based on the sentiment lexicon and merges Convolutional Neural Network (CNN) with attention-based Bidirectional Gated Recurrent Unit (BGRU) techniques (BiGRU). This work is tested only on Chinese language. Hybrid Recommendations is one of the important modules in research [9]. To improve the accuracy, they used a supervised learning technique (Support Vector Machine). A cluster-based approach is applied in their work (Fig. 5). Deep learning modified neural network (DLMNN) [10] is a strategy in which the system uses Neural Network to portray the results as negative, positive and neutral ratings. This system has also provided significant result improvement on particular dataset. After reviewing various research work, findings are summarized as following (Table 2).

86

Fig. 4 Dictionary approach for SA [5] Fig. 5 Cluster based SA system [9]

A. Patel et al.

A Survey on E-commerce Sentiment Analysis

87

Table 2 Summary of literature survey Dataset

method used

Findings

Limitation

May Real-time sentiment analysis -2019 on E-commerce application [1]

Amazon Review Data from Amazon.com

SVM

Advantages Two levels of categorization: review level and sentence level Drawback: No weightage given to features

Tough in managing complex structures of sentences and different languages. No weightage given to features. No advanced deep learning approach used

Sentiment analysis for e-commerce products using natural language processing [2]

Amazon

NLP Flask

NLP enables better sense from review Web scraping is dependent, may fail to get data every time

Data fetching might result in failure. Sentiments generated can be bias

Aspect-level 2018 sentiment analysis on E-commerce data [3]

Amazon

SVM

Aspect level classification applied Limited dataset tested

Different languages issues

Sentiment analysis on large scale amazon products reviews [4]

May 2018

Amazon

Naïve Bayesian (MNB) and support vector machine (SVM)

Big Data is tested and got better result Feature weighting is not applied

Data collection problem as not enough data is provided by e-commerce site publicly. Can’t scrape enough Data to consider it as real-life public reviews over different products

Sentiment analysis of E-commerce text reviews based on sentiment dictionary” [5]

Jun 2020

Amazon

TF/IDF

TF—IDF may be used for extracting exact keywords Other classifier with TF/IDF may improve result

Part of speech tagging turns out to be biased and affects the output results

Title of paper

Year

May 2021

(continued)

88

A. Patel et al.

Table 2 (continued) Title of paper

Year

Dataset

method used

Findings

Limitation

An experimental Mar analysis on 2021 E-commerce reviews, with sentiment classification using opinion mining on web [6]

E-commerce websites

Feature extraction matrix

Better system for low quantity data Manually data collection is not proper method for large scale products

Data collection of mobile products only and selecting top reviews only

Sentiment analysis on product reviews [7]

Online API

SentiWordNet

POS + SVM got better result Bigger dataset should be tested

No multi language reviews are taken into consideration Challenges in finding the sentiments for sarcastic reviews

2020 Sentiment analysis for e-commerce product reviews in Chinese based on sentiment lexicon and deep learning [8]

dangdang.com

CNN

Sentiment lexicon and Deep learning applied Only Chinese language tested

Only categories the sentiments into positive and negative not suitable for high preferences for sentiment refinement

“Sentiment analysis in E-commerce using Recommendation System” [9]

2021

Amazon

SVM

Clustering approach applied Weighted features can improve the result

Product features wise sentiments are not carried out

Sentiment analysis of online product reviews using DLMNN and future prediction of online product using IANFIS” [10]

2020

E-commerce

Deep learning modified neural network

Deep learning modified neural network is better than traditional classification Hyper parameter should be tuned

Cannot understand the complete context of the entire piece while keyword processing to get correct sentiments

2019

3 Conclusion After reviewing some of the research work done in recent time for sentimental analysis in E-Commerce, we have founded some important findings. One of the common

A Survey on E-commerce Sentiment Analysis

89

issues is found like single source of data that can misguide the system generated sentiments. As entire review is analyzed, no particular features of the product are extracted in most of cases. Feature of the products e.g., Battery of Laptop, Processor, Speed etc. can be separated for sentimental analysis. Multilanguage review extraction and translation for enriching the dataset can also be applied. Region wise sentimental analysis can also be helpful for end user. As a future work better preprocessing methods, NLP methods for enriching the dataset and feature selection process can be improved before classification. Also, fake review detection can help the entire system to perform better for end user.

References 1. J. Jabbar, et al., Real-time sentiment analysis on E-commerce application, in 2019 IEEE 16th International Conference on Networking, Sensing and Control (ICNSC) (IEEE, 2019) 2. B.K. Jha, G.G. Sivasankari, K.R. Venugopal, Sentiment analysis for E-commerce products using natural language processing. Ann. Romanian Soc. Cell Biol. 166–175 (2021) 3. S. Vanaja, M. Belwal, Aspect-level sentiment analysis on e-commerce data, in 2018 International Conference on Inventive Research in Computing Applications (ICIRCA) (IEEE, 2018) 4. T.U. Haque, N.N. Saber, F.M. Shah, Sentiment analysis on large scale Amazon product reviews, in 2018 IEEE international conference on innovative research and development (ICIRD) (IEEE, 2018) 5. Y. Zhang, et al., Sentiment analysis of E-commerce Text reviews based on sentiment dictionary, in 2020 IEEE International Conference on Artificial Intelligence and Computer Applications (ICAICA) (IEEE, 2020) 6. S.S. Latha, An experimental analysis on E-commerce reviews, with sentiment classification using opinion mining on web. Int. J. Eng. Appl. Sci. Technol. 5(11), 2143–2455 (2021) 7. X. Fang, J. Zhan, Sentiment analysis using product review data. J. Big Data 2(1), 1–14 (2015) 8. L. Yang et al., Sentiment analysis for E-commerce product reviews in Chinese based on sentiment lexicon and deep learning. IEEE Access 8, 23522–23530 (2020) 9. R. Krithiga, M. Anbumalar, D. Swathi, Sentimental analysis in E-commerce using recommendation system. Int. Sci. J. Contemp. Res. Eng. Sci. Manag. 6(1), 53–58 (2021) 10. P. Sasikala, L. Mary Immaculate Sheela, Sentiment analysis of online product reviews using DLMNN and future prediction of online product using IANFIS. J. Big Data 7, 1–20 (2020) 11. Kaggle, Consumers reviews of amazon products. Consumer Reviews of Amazon Products|Kaggle

Secured Cloud Computing for Medical Database Monitoring Using Machine Learning Techniques M. Balamurugan, M. Kumaresan, V. Haripriya, S. Annamalai, and J. Bhuvana

Abstract A growing number of people are calling on the health-care industry to adopt new technologies that are becoming accessible on the market in order to improve the overall quality of their services. Telecommunications systems are integrated with computers, connectivity, mobility, data storage, and information analytics to make a complete information infrastructure system. It is the order of the day to use technology that is based on the Internet of Things (IoT). Given the limited availability of human resources and infrastructure, it is becoming more vital to monitor chronic patients on an ongoing basis as their diseases deteriorate and become more severe. A cloud-based architecture that is capable of dealing with all of the issues stated above may be able to provide effective solutions for the health-care industry. With the purpose of building software that would mix cloud computing and mobile technologies for health-care monitoring systems, we have assigned ourselves the task of designing software. Using a method devised by Higuchi, it is possible to extract stable fractal values from electrocardiogram (ECG) data, something that has never been attempted previously by any other researcher working on the development of a computer-aided diagnosis system for arrhythmia. As a result of the results, it is feasible to infer that the support vector machine has attained the best classification accuracy attainable for fractal features. When compared to the other two classifiers, M. Balamurugan Department of Computer Science and Engineering, School of Engineering and Technology, Christ (Deemed to Be University), Bangalore, India e-mail: [email protected] M. Kumaresan School of Computer Science and Engineering, Jain (Deemed to Be) University, Bangalore, India e-mail: [email protected] V. Haripriya · J. Bhuvana (B) School of Computer Science and IT, Jain (Deemed to Be) University, Bangalore, India e-mail: [email protected] V. Haripriya e-mail: [email protected] S. Annamalai School of Computing Science and Engineering, Galgotias University, Greater Noida, India e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 I. J. Jacob et al. (eds.), Expert Clouds and Applications, Lecture Notes in Networks and Systems 444, https://doi.org/10.1007/978-981-19-2500-9_7

91

92

M. Balamurugan et al.

the feed forward neural network model and the feedback neural network model, the support vector machine excels them both. Furthermore, it should be noted that the sensitivity of both the feed forward neural network and the support vector machine yields results that are equivalent in quality (92.08% and 90.36%, respectively). Keywords Neural network · Cloud database · Electrocardiogram · Fractal features

1 Introduction One of the world’s most remarkable networked systems, the Internet enables devices to communicate with one another across the globe by utilizing an established set of standard protocols and connecting a diverse range of different networks, including academic institutions, commercial and government organizations, among other things. Early on, the Internet was largely constituted of static websites and electronic mail communication, which was the norm. In today’s world, there are many different Internet execution modes visible everywhere we look, and they are integrated into many different aspects of our lives, providing an endless array of services and applications while attempting to meet the needs of every user, regardless of where they are or what time of day they are doing so. Consumers now have more access to and mobility with Internet technologies than they have ever had before, as shown by the fact that all of their devices include Internet technologies in some way or another. The availability of smart devices, which allow us to maintain a constant connection with other areas of the world, is considered a vital feature of modern society’s everyday life. As a consequence, the number of connected devices continues to increase at an alarming pace year after year. In order to do this, it will be required to establish an autonomous device communication system. The Internet of Things (IoT) is one of the most promising choices accessible right now, according to industry experts (IoT). When it comes to real-world goods, the Internet of Things (IoT) is an information network that allows for the retrieval of information about them as well as their direct connection with one another. It is a platform that is meant to store and analyze data from the Internet of Things (IoT), often known as the Internet of Things Cloud (IoT Cloud) (IoT). Devices, sensors, websites, applications, users, and business partners all generate massive volumes of data, which we can’t keep up with. The platform is meant to take in this information and initiate operations in order to respond in real time. IoT cloud systems are service frameworks that are dynamic and accessible on demand, as indicated by the Internet of Things. Currently, the global market is flooded with Internet of Things cloud platforms that are tailored to meet the needs of a wide range of user and application groups, such as businesses, governments, farmers, and health-care providers, as well as communication and transportation companies and manufacturers, among others.

Secured Cloud Computing for Medical Database …

93

Different cloud computing methods are being used in the fields of bio-medical and signal processing, with the future health-care sector utilizing current cloud architecture and the Internet of Things as findings from this study. As a result of the study’s findings, the future health-care sector will use current cloud architecture and the Internet of Things (IoT). This objective focuses on the design and management of existing biomedical signal processing systems in conjunction with IoT and cloud systems, as well as the formation of a network with the least amount of latency, in order to reduce the complexity of waiting time and to provide fast access to information with minimal delay. Many researchers have experimented with a variety of methodologies, but they have not fully used the resources that are accessible to them. It is our goal to design a system model that is efficient and effective in terms of data collection and analysis as well as transmission and utilization in a highly efficient and effective manner. • When using a cloud-based system, the data allocation process should be more efficient, and the effective utilization of the present network should be enhanced as a consequence. In order to keep time complexity to a minimum, a calm method must be followed for data transmission and contact with the end system (the end user). In order to minimize overall delay in the transmission of information, as well as the quantity of redundant information, the available network bandwidth must be used completely with no idle time. • The available network bandwidth must be utilized completely with no idle time. It is important to develop software that integrates cloud computing with mobile technologies in order to implement a health-care monitoring system. The second goal is to develop a rudimentary prototype model for identifying the ECG signal (ECG signal detection). A feature extraction neural network model as well as a support vector machine model are being created in order to categorize the Arrhythmia. 4. To evaluate the performance of the suggested method when it is implemented on a cloud-based computing platform, when doing ECG. When doing ECG signal analysis, the fundamental goal of the feature extraction and classification approach is to distinguish between five distinct forms of arrhythmias: atrial fibrillation, ventricular fibrillation, ventricular fibrillation, and ventricular fibrillation. Classifications such as Class N, Class S, Class V, Class F, and Class Q are just a few examples.

2 Literature Survey An architecture for cloud computing is a technical term that refers to the design of a computer system that makes use of cloud computing technologies. In a cloud computing system, the front end and the back end are two separate components that work together to create the cloud computing system. Their connections to one another are made possible through a network, which is most typically the Internet. As opposed to the back end, which is located on-premises, the front end is located in the cloud and may be accessed by the client (user) using a web browser. Cloud computing services, such as multiple computers, servers, and data storage facilities, are included inside the back end. The front end is comprised of the client’s computer and the

94

M. Balamurugan et al.

application that will be used to access the cloud. Managing traffic and responding to client requests fall within the purview of the central server, which is responsible for monitoring and managing traffic. It complies with a set of standards known as protocols and makes use of specialized software known as middleware to do this. Middleware is software that allows computers linked to a network to communicate with one another and exchange information. When doing their study, Lin et al. delved into great detail into how Organizations Biology perceives the living organism as an organic network. The field of systems biology is distinct from reductionist approaches in that it is more concerned with the interactions between molecules located at multiple omics levels. A variation on this concept has been used to the research of network biomarkers and network medications, which combine clinical data with knowledge of network sciences and, as a result, support the study of sickness in the age of biomedical informatics. The prior is concerned with the identification of precise signals for the purpose of sickness detection and diagnosis in humans. As previously stated by Das et al. [11], electronic health is a rising star that represents the collaboration of medical research and statistical technology, a ray of hope for a prosperous future in health and prosperity, as well as a straightforward option to rely on when in need of medical assistance and assistance. However, the question arises as to whether or not e-health is a truly dependable replacement to traditional healthcare. In this essay, we will evaluate the ethics with which e-health is done, whether or not the code of principles is being carefully obeyed, and, if not, whether or not proper e-health is entirely unpleasant and ineffectual in its implementation. By cooperating with health-care providers and information-technology providers, it is vital to encourage the motivated thinkers who have taken the initiative to deliver a better and faster solution to all health-related concerns. As Chatman pointed out, the most significant aspects of cloud computing offerings are those that are transforming the landscape of health-care information technology in the United States of America. Specifically, it addresses the relevance of cloud computing in addition to the role that it plays in the aforementioned industry. There are a multitude of reasons why health-care institutions, public-sector organizations, and a range of other well-being facilities are adopting the aforementioned contributions, one of which is the speed with which they may be given. 2014 was the year in which Knut Haufe and colleagues tackled the issue of cloud computing is quickly becoming one of the most popular areas in information systems research, coming in second only to database management systems in terms of popularity among researchers. Particularly in the case of health-care companies, it is necessary to assess and handle the specific risks connected with cloud computing in their evidence security management system, taking into consideration the nature of the information being processed. This article presents an overview of the most important security procedures in the context of cloud expansion in the health-care industry, which will be examined in more depth in the next paper. The most important information security processes for health-care organizations utilizing cloud computing will be identified and branded in accordance with the general information security management processes derived from the ISO 27000 family of standards, taking into

Secured Cloud Computing for Medical Database …

95

consideration the primary risks associated with cloud computing as well as the nature of the data being handled. The methods that have been defined will aid a health-care organization that is using cloud computing to focus on the most vital tasks and to create and operate them at a level of development that is suitable given the limited resources that are now available. Kwon and his associates—According to a paper published by the Institute of Electrical and Electronics Engineers, prognostics and systems health management (PHM) is a supporting discipline that employs sensors to assess the health of systems, detect anomalous behaviour, and forecast the remaining useful routine over the life of the benefit. Having been made possible by the Internet of Things (IoT), predictive asset management (PHM) may now be applied to any kind of resource across any industry sector, resulting in a paradigm shift that is opening up significant new economic prospects for enterprises of all sizes. A summary of their thoughts on PHM, as well as the opportunities offered by the Internet of Things, is presented in this presentation. Engineering change and consumer goods are being developed inside a framework being built. Consequently, with the rising use of IoT-based PHM, there has also been a surge in the number of challenges in correlating. There are a variety of topics covered, including correct analytical methodologies, security, Internet of Things platforms, sensor energy collection, Internet of Things business models, and verification procedures, among others. Ghanavati and his associates—It is suggested that the Internet of Things (IoT), which makes use of connected sensors (such as wireless body area networks (WBAN), may give chances for real-time monitoring of patient health status and management of patients and cure in a publication by Springer that is available online. Consequently, the Internet of Things (IoT) will play a crucial role in the development of the next-generation health-care organization. In recent years, remote monitoring of patient health conditions through the Internet of Things has gained popularity. However, monitoring patients outside of hospital settings requires enhancing the Internet of Things’ capabilities by providing additional resources for health data storage and processing, which are currently lacking. A continuous patient health status monitoring framework that connects WBAN through smartphones to cloud computing is presented in this research. The Internet of Things-based serviceoriented background for the framework is provided in this paper. Experimentation has shown that the proposed architecture outperforms baseline WBANs in a sub-static way, as measured by sensor lifetime, existing cost, and energy usage. According to the findings, the SQA methodology delivers promising results in recognizing ECG signals with unsatisfactory quality and surpasses existing techniques based on morphological features and mechanism learning approaches in detecting unacceptable quality. Finally, they have accomplished the transmission of ECG signals of acceptable quality, which has the potential to dramatically extend the battery life of IoT-enabled devices in the future. In order to increase the accuracy and reliability of unproven diagnostic systems, as well as other applications, the quality-aware Internet of Things model has a high chance of calculating clinical acceptability of electrocardiogram data.

96

M. Balamurugan et al.

3 Materials and Methodology It is proposed in this research that three possible enhancements for creating intelligent architectural designs be explored further. Because of these developments, the sensor’s framework will be strengthened, allowing the client to analyze the information acquired by the sensor. A basic current process is regulated at their control stations, with variables such as the level, the weight, and the temperature being monitored. In order for the measuring process to be successful, the physical property that is being measured must be translated into an electrical quantity such as current or voltage. To be more specific, this is necessary for the transfer of such characteristics to control stations for the purpose of monitoring those located in remote areas. Various industries will gain from the monitoring, the analysis, and the control that are carried out, as well as the technology that is employed to carry out these operations. With the main purpose of identifying RR peaks in ECG data, the fundamental objective of this approach is to utilize the ECG signals as input. The service-oriented sensor network architecture (SOSA) is used to enable heterogeneous sensor systems with the communication of the server in the network. The services for sensor developers are provided by the SOSA. The sensor services will be distributed into the eb-xml registry, which will make use of the service description language in order to facilitate their usage. With the use of client sensor apps, the essential arrangements in such services may be discovered and also triggered. It is this kind of network configuration that is founded on a little segment known as an integration controller, which takes a role in linking two technologies, namely, SOSA and the cloud. A detailed description of this approach, which is flavoured and adapted from the client’s PC to the condition of remote computing, follows. As part of this kind of architecture, each Integration Controller will be responsible for relaying the information received from the different sensor systems to the general public using the cloud infrastructure outlined above. Figure 1 depicts the process by which the layered engineering of the system in question was done. In order to analyze this sensed data and convert it into an XML format, extension software will be built and stored on a web server that will be accessible from anywhere in the globe over the Internet. In addition, this engineering allows clients of the sensor to take an active role in the process in an easy way and searches for a large amount of sensor data in a variety of systems at the same time. In order to ensure that the sensor data is captured throughout the sensor’s lifespan, this data will be kept in the system’s back-end storage system. Sensor data is expected to continue to move along the lines of a loosely regulated system to a highly managed cloud, which will complete the data management system that has been created for this sensor data. A cloud-based sensor system will be created by using vast quantities of information and inquiry to produce the sensors that will be necessary for the cloud, as well as improvements in information technology, which will give a superior presentation for such cloud-based sensors. Sensor information will be stored in the sensor cloud, which will then be made available to different sensor customer applications

Secured Cloud Computing for Medical Database …

Get UPD Data

97

Refresh Data Every 60 Seconds

Counters

Timers Data Classification

Statistics Sets

Guages

Back-End Processing

Back-End Processing 1

Back-End Processing 2

V

Statistical Data

Back-End Processing N

Fig. 1 Architecture of Big data system

in order to better address the needs of the customers. They must sustain massive information conditions while also giving some important information and a fast pace of sensor data transfer, all while maintaining a large amount of information. Because the sensors proposed in this framework will provide information on a constant basis, vast quantities of information that has been saved for factual report arrangements will be leaked as a consequence of the continuous flow of information. Businesses will be able to identify and estimate such large volumes of sensor information if they conduct a precognitive assessment of this sensor information from a future-situated perspective depending on what is coming down. This massive sensor information in the Sensor Cloud possesses many important capabilities, including the ability to capture the cloud, manage and control the information in accordance with industry requirements (in this case, paper mills), and the combination of performance information that will deliver the right sensor at the ideal location and time. When it comes to building a structure of capacity, this social model is based on a social database, which uses a table of information as its source for information collection and is produced from scratch. In this scenario, it is vital to differentiate between each of the tables as well as to portray them in an official manner in line with the social model. The “essential key,” as previously stated, will be used to distinguish every line on the table from the lines of the other table by means of the creation of an “outside

98

M. Balamurugan et al.

key,” which can be either a segment or a gathering of the table to that of the essential key of the other table. It will be interesting to see how this is accomplished. Using the T-SQL language, this method will process the piece and store it after checking its linguistic structure with the T-SQL that will be received in the next step. The Idea of a Web-based Health-Care Service: What Is It? In order to distribute and discover Web Services on the Web 4, it is necessary to use reusable components that are capable of being disseminated and found on the Internet. The use of an open-sort standard (such as XML, SOAP, and so on) of the Web correspondence convention, as opposed to a closed-sort standard, is accomplished through the use of a web correspondence convention that is close by having information that will provide administrations to a variety of applications. Its capabilities will be restricted to replying to any calls sent by the client to its server, if any such calls are received. The applications of the different phases, which will create an opportunity for them to interact with one another, will be the key cause in this scenario. When this application is utilized, it will examine the information sources and then request that the web service associated with it be registered with the application. The design of the user interface for this project is accomplished through the use of the C# programming language; the assembly of this application may be accomplished via the use of the .NET framework. When it comes to dialect forms and language structures, the XML format is a kind of language structure that can be extended by the client and is distinct from the PC programme in terms of flexibility. Instead of being emitted from the SGML, this is emitted from it (Standard Generalized Markup Language). If the XML is concerned with effective ways of communicating in an organized manner, it may be possible to replace HTML, which is concerned with the appearance of records in programmes. However, if the XML is concerned with effective ways of communicating in an organized manner, it may be possible to replace HTML. In addition to the HTTP, FTP, SMTP, and TCP protocols, the Cleanser uses the Simple Object Access Protocol (SOAP). SOAP is a standard that is equivalent to the HTTP, FTP, SMTP, and TCP protocols, among others. Because of the LINQ approach for interfacing the dialects of a different programme that can resist organized programming, this T-SQL dialect can be made as easily understandable as it is in C#, in which an important dialect will be distinct in generating related information, as well as in other programming languages. Pre-processing: Among other things, pre-processing of raw ECG data is necessary in order to remove unwanted components such as muscle noise and 60 Hz impedance. The typical meander and T-wave interference are also removed during pre-processing. The standardization and filtering operations will be carried out during the pre-processing stage of the procedure. During this procedure, the signal’s amplitude is normalized, and the signal is then sent through a band pass filter with noise rejection. The best pass band for maximizing this QRS energy will have a frequency range of around 5–15 Hz. Businesses will be able to identify and estimate such large volumes of sensor information if they conduct a precognitive assessment of this sensor information from a future-situated perspective depending on what is coming down. This massive sensor

Secured Cloud Computing for Medical Database …

99

information in the Sensor Cloud possesses many important capabilities, including the ability to capture the cloud, manage, and control the information in accordance with industry requirements (in this case, paper mills), and the combination of performance information that will deliver the right sensor at the ideal location and time. When it comes to building a structure of capacity, this social model is based on a social database, which uses a table of information as its source for information collection and is produced from scratch. In this scenario, it is vital to differentiate between each of the tables as well as to portray them in an official manner in line with the social model. The “essential key,” as previously stated, will be used to distinguish every line on the table from the lines of the other table by means of the creation of an “outside key,” which can be either a segment or a gathering of the table to that of the essential key of the other table. It will be interesting to see how this is accomplished. Using the T-SQL language, this method will process the piece and store it after checking its linguistic structure with the T-SQL that will be received in the next step. The Idea of a Web-based Health-Care Service: What Is It? In order to distribute and discover Web Services on the Web 4, it is necessary to use reusable components that are capable of being disseminated and found on the Internet. This is accomplished by the use of an open-sort standard (such as XML, SOAP, and so on) of the Web correspondence convention, which is close by to having information that will supply administrations to a range of applications.

4 Implementation A linear predictor that self-changes in reaction to approaching statistics of incoming signals will be used in this proposed JQDC method, rather than a fixed indicator, with the goal of producing an indication that self-changes in response to approaching statistics of incoming signals. The utilization of a tapped delay line structural design is used to accomplish this predictor’s results. When updating the weight of the predictors, the LMS system will be utilized, and the predictors will be as follows: In a Linear Predictor, the following is the order of the predictors: Due to the fact that there would be a link between the proposed JQDC and its performance, it is necessary to undertake an inquiry into correlations between the order of linear predictors and their performance. This was achieved effectively by the use of ECG signals collected from the MIT and the BIH databases, respectively, as described above. Following predictions, the CR will improve as the predictors are placed higher and higher up in the order of occurrence. The performance of QRS detection based on the SE and +P, on the other hand, reveals a different pattern of behaviour when compared to other methods. Performance increases with each iteration of the sequence; but, by the fourth iteration, there is an instantaneous error in the signal component, as a consequence of which the detection accuracy has been decreased to an unsatisfactory level of accuracy. It is also possible that there are few orders because the accuracy of prediction is weak, and this may result in low frequency baseline fluctuations and

100

M. Balamurugan et al.

asymmetric P/T wave components, and the erroneous output will have an effect on the accuracy of the QRS detection algorithms. The following are the startup and step size parameters: A considerable number of cycles are required by adaptive systems before they reach their optimum point. The optimal point is dictated by the characteristics of the incoming signal, and in order to speed up this process, the adaptation must be initiated early. The SSLMS predictor is used in combination with previously derived data to provide a more accurate prediction as seen in Fig. 2. The detected data will be captured and analyzed with the help of the WSNEDU2110CB Wireless Sensor Network Educational Kit, which operates at 2.4 GHz Data

Data

dependency

dependency

batch

batch

Product transformation step batch batch

Data transformation step

Data

Data

dependency

dependency

shift Data dependency

shift

Fig. 2 Cloud with medical data

*

Secured Cloud Computing for Medical Database …

101

and contains the Data Acquisition boards with sensors as well as the PC Interface Boards (MIB520). According to QusayH. Mahmoud (2004), the JAX-RPC 1.1 API on the J2EE 1.4 platform provides complete support (QusayH. Mahmoud, 2004) for services that are available via the J2EE 1.4 platform. Written in such a way that the XML namespace is specified, configuration files are compiled in order to generate the SSDL, which contains some inputs such as the address of the server for the client reference and a mapping file that contains a port number and the endpoint location of service for the server reference. The war files are generated in this fashion by the deployment tool from the services that have been built and are then deployed in a server of this kind, as shown in Fig. 2. Sensor system registry (also known as sensor system registry): The sensor system registry is used to facilitate the aggregation of services inside applications simpler by allowing information to be sent between the apps and the sensor system. In order to make them accessible for distribution, the eb-xml registry is used. In order for the sensor services to be registered in the repository, it was necessary to have SSDL files for the sensor services that had service bindings. A breakdown of the services available is shown in Fig. 3. Sensor as a cloud-based service: What exactly is it? The Integration Controller will now upload the sensed data to the Cloud server, where it will be stored for later retrieval. On this illustration, you can see an example of XML code for sensed data that is being disseminated in the cloud. The sensor data is acquired from the sensor cloud via the use of the Hive query command tool, which is supplied with the Hadoop installation of the programme.

Data Center Network Diagram

Data Center WWW Server Communicate Database Client Server

File Server

Financ e

CommunicateCommunicate Server Server Database

Record

Custo m

Database DDN

Router

Switch Others

Manager Manager

Data Center

Router

Switch Finance

Internet Router

Monitor Fireware Internet

Record Financ e Communicate Server

Client Center

Fig. 3 Input register

Clients

Customs

Switch Others

Client Center

102

M. Balamurugan et al.

Fig. 4 File collected in cloud

The sensor data is stored in the sensor cloud until it is needed. A query command has been executed on the hive in order to collect and also analyze sensor data from the device that is being watched. This sensor file contains information that is being analyzed via the usage of the MapReduce algorithm. In order to build connections between a user control centre and a Web Service, it has been found that there are two separate methods to do so: in one direction for data and in another for the Panorama Map. Control Panel for the End-User (Fig. 4): For the purpose of storing the vast quantity of data produced by the front-end sensors as a consequence of the employment of end sensors, a virtual machine will be constructed that will store the humidity, temperatures, and pH value of the virtual machine, among other things. The Web Service will be used to get the contents of the virtual machine at a later time. The display window for this operation will be similar to the display window for the Web Service, and it will extract the quality and amount of data from the database, after which the cloud will calculate the results, which will be shown in the Web Service’s display window. First, you must make a list of all of the elements you wish to include in your Panorama Map. For example, you may add: Through the usage of this software, the start instruction is communicated to the cloud side, which in turn notifies the cloud side of the user’s location and computes the image result, which is then transmitted back to the user control centre. Before the map can be used, it is required to specify seven different parameters on it. In addition to the width and height of the user control centre display window, the current map of the location’s X and Y coordinates, the displacement’s X and Y coordinates, and the newly shown new map hierarchy are all presented. The map may be accessed and manipulated by selecting the “Operate” button on the user control centre’s main display window. Any changes made to this map by the user, such as zooming in or out, will result in the control centre displaying a new map that restores the zoom level at when the user initially logged in. Images

Secured Cloud Computing for Medical Database …

103

of farmland and the sensor data that relates to them may be shown alongside one another if the map’s display is enlarged to its maximum size and the map’s display is expanded to its maximum size. When the sensor is activated, the user gets access to the information and data that has been gathered by the device. UI is an abbreviation for User Interface, and it is comprised of the following elements: The design of this data curve may be split into three categories: the original design, the static design, and the dynamic design. The original design is the most basic kind of design. The original design is the most basic of the three sorts of designs. It is also the most common. The most basic kind of design is the original design, which is the most common type of design. As shown in Fig. 5, the QRS-complex, which involves multiple waves in succession, results in the formation of a wave combination. Figure 5: QRS-complex in action (Fig. 6). On the x-axis, you can see a comparison of the input ECG signal, and on the y-axis, you can see how the amplitude of the signal is measured. 6. The decomposed

Fig. 5 ECG trace

Fig. 6 Input signal

104

M. Balamurugan et al.

signal is referred to as the decomposition signal since the data used in the simulation is compressed. When looking at Fig. 7, it can be seen that the reconstruction of the data is carried out numerous times, with the error for the prediction reducing with each level of reconstruction. This graph is shown by an x-axis showing the passage of time and a y-axis reflecting the amplitude of the signal that is represented by the passage of time. QRS peaks are recognized at the end of each level of prediction, and the total number of peaks is calculated at the beginning of the experiment. Figure 8 depicts the calculation of the total peak value based on the ECG signal, with the peak value highlighted to illustrate that this is the initial step in the detection of the QRS. On the right, you can see the high and low peak values of the electrocardiogram signals, as well as the associated values, which are used to distinguish between distinct peak values of the signals (as shown on the left in Fig. 9). Figure 10 illustrates the final signal discovered, known as the QRS, which has various peak values such as P, Q, R, S, and T, which are separated by the use of different colours to help in understanding. Figure 10: The last signal identified, known as the QRS. This pattern illustrates that the fractal feature of the ECG arrhythmia is the most acceptable characteristic for use with the classification technique. In order to get the lowest possible classification accuracy for classification problems, the feedback neural network has been employed to do this. Nevertheless, when it comes to discriminating between the five different forms of arrhythmia, the feedback neural network system obtains an accuracy of around 90 percent using this method. Alternatively, the observed discrepancies in classification efficiency between the fractal features are mostly due to the continual change in morphological elements of the arrhythmia, as has been previously established. According to Table 1, an overall comparison of the suggested work with several existing algorithms such as Naive Bayes, Decision Tree, Bayes Net, and Fuzzy Petri Net is presented for the sake of simplicity. The measures’ accuracy, sensitivity, and specificity are all evaluated in comparison. When compared to other existing algorithms, the overall result reveals that the proposed technique outperforms them in all areas, as seen in Table 1 (Fig. 11). In accordance with current legislation, this proposal this figure, which compares the accuracy of several current and proposed techniques, indicates unequivocally that the accuracy of the proposed Feed Forward NN and SVM approaches outperforms the accuracy of other existing approaches.

5 Conclusion A fundamental analysis of QRS signal detection is presented in this chapter. Additional talks of specific challenges related with wireless sensor networks and their integration into cloud infrastructure are also presented in this chapter. When compared

Secured Cloud Computing for Medical Database …

Fig. 7 Reconstruction data

105

106

M. Balamurugan et al.

Fig. 8 Selected overall peak

to intracommunication, wireless intercommunication provides more flexibility while also lowering the total cost of ownership. Additional to this, as a consequence of technical improvements, the use of BAN is becoming more widespread. A dataset from the Massachusetts Institute of Technology and the Boston University School of Medicine, which included 48 h of data from 47 patients, was utilized in this study. The arrhythmia data was collected using two-channel ambulatory ECG recordings, which were then used to do the analysis on the information. Higuchi et al. created the Higuchi fractal approach for extracting the fractal dimension characteristic from an ECG arrhythmia, which may be used to diagnose heart failure. Three different classifier models were employed to assess the performance of a computer-aided diagnostic system. The results were compared. A study conducted by the University of California at Berkeley found that SVM outperforms other classification algorithms in terms of classification accuracy, sensitivity, and specificity.

Secured Cloud Computing for Medical Database …

Fig. 9 Generated classes

Fig. 10 Train phase

107

108 Table 1 Performance metrics

M. Balamurugan et al. Classifier

Accuracy

Sensitivity

Specificity

Naive Bayes

83.8

85.0

83.4

Decision tree

93.2

87.5

89.5

Bayes Net

90.8

85.0

82.4

Fuzzy Petri net

85.4

92.5

83.4

Associative Petri Net (APN)

93.5

85.0

85.9

Feed forward NN

94.34

92.08

90.24

Feedback NN

92.23

90

88.65

SVM

95.65

92.36

91.35

Fig. 11 Comparison graph for accuracy

References 1. U.R. Acharya, S.L. Oh, Y. Hagiwara, J.H. Tan, M. Adam, A. Gertych, R. San Tan, A deep convolutional neural network model to classify heartbeats. Comput. Biol. Med. 89, 389–396 (2017) 2. Actigraph, Available at: http://www.mtiactigraph.com. Accessed: July 2005. 3. S. Ahmed, N. Javaid, M. Akbar, A. Iqbal, Z.A. Khan, U. Qasim, LAEEBA: link aware and energy efficient scheme for body area networks, in Proceedings—International Conference on Advanced Information Networking and Applications, AINA, pp. 435–440 (2014) 4. A. Al-Fuqaha, M. Guizani, M. Mohammadi, M. Aledhari, M. Ayyash, Internet of Things: a survey on enabling. IEEE Commun. Surv. Tutorials 17(4), 2347–2376 (2015) 5. M. Ali, H.S.M. Bilal, M.A. Razzaq, J. Khan, S. Lee, M. Idris, M. Aazam, T. Choi, S.C. Han, B.H. Kang, IoTFLiP: IoT-based flipped learning platform for medical education. Digital Commun. Networks 3(3), 188–194 (2017) 6. S. Amendola, R. Lodato, S. Manzari, C. Occhiuzzi, G. Marrocco, RFID technology for IoTbased personal healthcare in smart spaces. IEEE Internet Things J. 1(2), 144–152 (2014) 7. M.R. Arefin, K. Tavakolian, R. Fazel-Rezai, QRS complex detection in ECG signal for wearable devices, in Engineering in Medicine and Biology Society (EMBC), 2015 37th Annual International Conference of the IEEE, pp. 5940–3 (2015)

Secured Cloud Computing for Medical Database …

109

8. L. Atzori, A. Iera, G. Morabito, The internet of things: a survey. Comput. Netw. 17(2), 243–259 (2010) 9. Available:http://www.texasheart.org/HIC/Topics/Cond/arrhycat.cfm. [Accessed: 04-Jul-2017] 10. D. Curtis, E. Shih, J. Waterman, J. Guttag, J. Bailey, T. Stair, R.A. Greenes, L. Ohno-Machado, Physiological signal monitoring in the waiting areas of an emergency room, in Proceedings of the ICST 3rd International Conference on Body Area Networks, p. 5 (2008) 11. D. Das, P. Maji, G. Dey, N. Dey, ‘Ethical E-Health: A Possibility of the Future or a Distant Dream?’, Int. J. E-Health Med. Commun. 5(3), 17–28 (2014) 12. J. Gotman, P. Gloor, N. Schaul, Comparison of traditional reading of the EEG and automatic recognition of interictal epileptic activity. J. Electroencephalogr. Clin. Neurophysiol. 44, 48–60 (1978) 13. M.T. Hagan, M. Menhaj, Training feedforward networks with Marquardt algorithm. IEEE Trans. Neural Netw. 5, 983–993 (1994) 14. Y.M. Hart, Management of newly diagnosed epilepsy. In S. Shorvon, D. R.Fish, E. Percucca & W. E. Dodson (Eds.), The treatment of epilepsy (2nd edn.). Balckwell Science, Oxford (2004), pp. 161–173 15. F.J. Harris, On the use of windows for harmonic analysis with the discrete Fourier transform, in Proceedings of Instrumentation, Electrical and Electronic Engineers (1978) 16. A.L. Loomis, E.N. Harvey, G.A.I. Hobart, Cerebral states during sleep. As studied by human brain potentials. J. Exp. Psychol. 21, 127–44 (1937) 17. M.T. Hamood, S. Boussakta, Fast Walsh–Hadamard–Fourier transform algorithm. IEEE Trans. Signal Process. 59(11), 5627–5631 (2011) 18. A.I. Baba, C. Câtoi, Bucharest: Tumor Cell Morphology, Comparative Oncology (The Publishing House of the Romanian Academy, 2007). http://www.ncbi.nlm.nih.gov/books/NBK 9553. Access Date: 22-03-2013

Real-Time Big Data Analytics for Improving Sales in the Retail Industry via the Use of Internet of Things Beacons V. Arulkumar, S. Sridhar, G. Kalpana, and K. S. Guruprakash

Abstract Various discoveries achieved by applying Apache Spark in medical administrations foundations are acceptable for a considerable data plan. A part of the educational or research-based social protection organizations are either trying new things with big data or using it in front-line research expeditions. In the medical industry, there is an almost unlimited amount of data that is being created. The Electronic Medical Record (EMR) alone gathers a vast quantity of information. The goal of using new trends and technologies such as the IoT, big data and others to analyze real-time beacon-based sensor data is to help these shopping mall-based retail shops or any other physical retail store compete with online shopping in terms of customized sales promotion, customer relations, different types of analysis such as predictive, diagnostic, and preventive using customer-sales data, and other aspects. We tested a number of beacon-based sensor systems for identifying neighboring mobile phones. The approach that has been proposed Apache Spark Streaming was utilized for various studies, with sample Amazon sales data being used as input. Keywords Ant colony optimization · Mobile sink · Wireless sensor networks · Efficiency · Internet of Things

V. Arulkumar (B) School of Computer Science and Engineering, Vellore Institute of Technology, Vellore, Tamilnadu, India e-mail: [email protected] S. Sridhar Department of Computer Science and Engineering, Saveetha School of Engineering, Saveetha Institute of Medical and Technical Sciences, Chennai, India G. Kalpana Rajalakshmi Institute of Technology, Chennai, India K. S. Guruprakash Department of Computer Science and Engineering, K. Ramakrishnan College of Engineering, Trichy, India © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 I. J. Jacob et al. (eds.), Expert Clouds and Applications, Lecture Notes in Networks and Systems 444, https://doi.org/10.1007/978-981-19-2500-9_8

111

112

V. Arulkumar et al.

1 Introduction The phrase “Internet of Things,” which is also known by its abbreviation, IoT, is composed of two words, the first of which is “Web,” and the second of which is “Things.” The first word is the main word, and the second word is “Things.” There are billions of customers all over the globe served by the Internet, which is a global arrangement of linked computer networks that use the standard Internet convention suite (TCP/IP) to communicate with one another. It is a system of systems that consists of millions of private, open, academic, commercial, and government systems, ranging in scale from local to global, all of which are linked together by an extensive cluster of electronic, remote, and optical systems management advances. Through the Internet, more than 100 countries are now linked together in the exchange of information, news, and emotions with one another. News and suppositions are disseminated through the Internet. While traveling to the Things, keep in mind that any question or person who can be recognized by our current reality may be encountered. Regularly occurring items incorporate not just the electronic gadgets that we experience and use daily, but also “things” that we don’t ordinarily consider to be electronic by any stretch of the imagination, for example, nourishment, clothing, and furniture; materials and parts; stock and concentrated things; points of interest, landmarks, show-stoppers, and so forth, in addition to the variety of trade, culture, and modernity. These systems generate enormous amounts of data, which is then sent to personal computers for analysis. When items are capable of both sensing and transmitting information, they are transformed into tools for comprehending and responding to the complexities of nature on a timely basis [1]. In the digital age, enormous amounts of information have been readily available to executives nearby. When we talk about massive information, we’re talking about datasets that are not just large in size, but also high in variety and pace, which makes them difficult to manage using traditional tools and processes. Because of the rapid creation of such information, procedures should be developed and provided with the aim of handling and concentrating esteem and learning from these datasets in mind. Furthermore, leaders must be able to accumulate important pieces of knowledge from such fluctuating and rapidly changing information, which may range from dayto-day interactions to client relationships and informal community information [2]. This kind of esteem may be bestowed via the use of massive information investigation, which is the use of cutting-edge investigation techniques to massive amounts of information. This article intends to deconstruct a few of the many investigation methods and apparatuses that may be used in conjunction with enormous data, as well as the opportunities provided by the use of enormous information investigation in a variety of distinct choice sectors [3]. After the massive information hoarding, follows the methodical planning and preparation of information. As previously said, there are four fundamental requirements for massive information preparation. The ability to quickly stack information is the most important need. Because plate and system activity interferes with the execution of inquiries during information stacking, it is critical to reducing the amount of time spent stacking information. The second

Real-Time Big Data Analytics for Improving …

113

requirement is the ability to prepare questions quickly. Many inquiries are response time essential, to meet the demands of heavy workloads and continuous solicitations as a result of the end goal of meeting these requirements. When a result, the information position structure must be capable of maintaining high inquiry handling speeds even as the number of inquiries increases at an alarming rate. Furthermore, the third need for large-scale information management is the very effective use of storage space. A limited circle of space necessitated by the rapid development in client activities necessitating adaptable capacity limits and calculating power, and information stockpiling necessitates that issues on how to store the information so that space use is increased be addressed during the planning process. Finally, the fourth need is robust workload designs that are capable of adapting to very dynamic workloads. Considering that enormous information sets are investigated by various applications and clients, for various purposes, and in various ways over time, the hidden framework must be extremely adaptable to sudden changes in information preparation and not be restricted to a specific workload design or set of applications and clients. Retailers must alter their approach to the market to succeed. As a consequence, the customer advertises, and the need is met by the client. Products of superior quality at a reasonable price, as well as welcoming after-sale services, are the three most important areas in which the company must concentrate to an extraordinary degree. It is necessary to provide additional administrations to buyers to misfortune them and build up a reputation of dependability, which would ensure consistent sales in the years ahead. The very essence of retail has changed over time. Today, retailing entails going into strip malls, getting on the Internet, and being mobile in your approach [4]. In all of these instances, small merchants miss out on a significant opportunity somewhere. The next-door shop is consistently the most important in terms of compassion for all reasons and seasons, and they must make use of modern technologies. Both online shops and brick-and-mortar stores must exist, but none can do so at the expense of the other. The study made a significant contribution by identifying neighboring mobile phones with the use of sensor devices that were already accessible. Obtaining information on the specified mobile phones is the goal of this operation. Analyze the history data of the mobile phone users who have been identified. To deliver the personalized sale promotion based on the studied data in real-time, the following steps must be taken: To examine the sales history of each category to predict the most challenging project [5].

2 Literature Survey It is well-known that much research has been conducted and published in the past on the deployment of mobile sinks for data collection, with the most recent publication being in 2012. In computer science, an NP-hard problem [6] is a problem in which organizing nodes into optimal clusters is very difficult to solve effectively. It is important to think about optimizing the size of the cluster since it is an effective

114

V. Arulkumar et al.

method of lowering the overall size of the cluster overall size. To maintain control over the size of a cluster, it is essential to keep the distances between member nodes and CHs as short as possible [7], as well as the amount of energy used by the whole network. To find the optimum cluster size, PSO reduces the distance between member nodes and cluster heads (CHs) in the directive. PSO is used in their suggested method, which they go into great depth to describe. Based on the findings of Bifet [8], GA has been used in wireless sensor networks to build the most energy-efficient clusters possible. According to the authors of [9], it is recommended that a self-clustering method for diverse WSNs be used in combination with a GA to prolong the lifetime of the network. To accomplish a range of objectives, several methods of modifying cluster size are used. They think that a routing tree that maximizes network lifetime while keeping the direction-finding path between each device and sink to an absolute minimum should be used to accomplish this [10]. It is shown in [11] that an energy consumption cost model is used in conjunction with a routing tree to evaluate the queries that are sent out. The study in [12] is aimed at determining the most effective route for each moveable sink as well as calculating the break length of a piece mobile sink along the trajectory to maximize the network’s longevity. For example, in [13], the use of a single mobile descendant to prolong the life of a network was studied in more depth to determine its effectiveness. In a recent study published in [14], researchers discovered a technique for extending the life of mobile base stations while simultaneously managing network demand at the same time. Chen et al. [15], Cohen et al. 16] proposes, as an example, that the mobile base station be run in the way of the greediest algorithm, according to the authors. The use of joint flexibility and steering in the case of a stationary base station with a limited range of capabilities is proposed [17–19]. The final analysis reveals that [20–22] control the route via the use of the mobile sink routing protocol, which is implemented in several different ways. Mobile sinks, on the other hand, are only useful in a very restricted number of situations in the real world of practice. Because a WSN is often located in highly populated regions, mobile sinks can only serve a small proportion of the total population. Because of several variables, including road conditions and mobile equipment, these autonomous vehicles are only capable of traveling at a certain pace. Aside from that, since mobile tools have a limited quantity of accessible energy, the maximum distance they can go on a trip is also restricted as a result of this. To maximize network lifespan while still maintaining speed, we must consider these limitations when developing a routing protocol. The mathematical representation of an intensive care unit’s collection of n randomly distributed, similar device nodes is the purposeless graph n = (VMS, E), where V represents the collection of n randomly distributed, similar device nodes in the intensive care unit and MS represents the mean temperature of the monitoring zone. When there are k mobile sinks, there are also k connections between sensors and sinks. When there are k connections between sensors and sinks, there are also k connections between sensors and sinks. A sensor system is regarded to be connected when the broadcast variations of two sensors u and v coincide; otherwise, the system is said to be unconnected. Remote monitoring using mobile sinks is unique in that they can both receive sensed information from device nodes and transmit the sensory

Real-Time Big Data Analytics for Improving …

115

Fig. 1 Overall flow diagram of the proposed work

data they have gathered to an offsite monitoring center, making them a one-of-a-kind kind of sensor. Each sensor node has a unique identity and a limited initial energy capacity Q, while each mobile sink has a lively capacity and access to an infinite supply of energy stored at the depot. Imagine that there are k mobile sinks in the system at any one moment, to make things simpler to comprehend and remember. The road map in the monitoring area is represented by the equation R = (Vr, Wr), where each vertex in Vr represents a road junction set and each piece advantage in Wr indicates a road segment (see Fig. 1). To accommodate a maximum of k mobile sinks, each of which has a liveliness capacity of eM, map R must be constrained in terms of its capacity. Approximately one-third of the total energy used by each mobile sink is consumed by travel and data transmission, as well as the receipt of the information provided. For the sojourn sites along the route, the units of et and ec will be used to represent the sink energy consumption for unit-length travel; for the beginning point, the units of both et and ec will be used to denote the sink energy consumption for a unit-length journey. It is estimated that when it comes to energy consumption rates, the wireless connection between sensor nodes accounts for a large proportion of the energy produced by the sensor nodes. The rest of the energy use, which is comprised of sensing and processing, is insignificant.

3 Materials and Methodology IoT Beacons: “Beacons are tiny wireless sensor devices that constantly broadcast a basic radio signal,” according to the National Science Foundation. The majority of the time, the signal is picked up by neighboring smartphones that are equipped with Bluetooth Low Energy (BLE) technology or WIFI signals. Apple Inc. trademarked the term “iBeacon” to identify its product. It is a new technology that Apple has

116

V. Arulkumar et al.

included into its location framework in iOS 7 and later operating systems, and it is described below. It enables Apple phones running iOS7 (or later versions) to continuously search the surroundings for Bluetooth Low Energy devices such as beacons, allowing them to be more productive.

3.1 Apache Spark for Real-Time Data Analytics: With its high performance, Apache Spark is a strong open-source processing engine that can handle both real-time stream processing and batch processing. It is 100 times quicker than Map Reduce in terms of processing speed. APIs are available in three programming languages: Java, Python, and Scala. To achieve generality, Spark SQL is utilized in conjunction with Spark SQL. This allows for streaming and sophisticated analyses. As part of our investigation, we will make use of the Spark ecosystem for both real-time data analytics and batch processing. Flume: This program is used for the gathering and compilation of real-time event data from many sources. Additionally, this tool is more dependable, scalable, controllable, and configurable, in addition to providing excellent performance. The data collected from beacons will be used in our study, and we will utilize this application to access it in real-time. MLlib: MLlib is a spark subproject that provides primitives for machine learning applications. It is based on Apache Spark, a fast and versatile engine for largescale data processing that was developed by the Apache Software Foundation. It is the Spark machine learning library that is scalable and contains standard machine learning techniques and utilities such as classification, regression, and clustering among other things. To detect signals in the form of WIFI or Bluetooth utilizing Bluetooth low energy from any nearby mobile phone, we will utilize sensor devices (Beacons) and link these devices with IoTs. Following that, we will use flume/Kafka to collect sensor data from the Internet of Things devices and convert it into JSON data format. Now that we have this JSON data, we will examine it in real-time using Spark Streaming and in the past utilizing the Apache Spark ecosystem, respectively. Our goal is to extract useful information from JSON data, display this information on a dashboard, and then link it to the Internet through the Internet of Things. Use this information to create personalized sales promotions in real-time and to increase total sales in the physical retail store, allowing them to compete with Internet shopping (Fig. 2). When it comes to gathering and transmitting recognition information, this system just uses standard, off-the-shelf get-to-point technology. In this way, in addition to achieving high identification rates, it is possible to achieve cheap equipment and setup costs. No matter how you look at it, the vast range and sparse character of our cleverly collected WiFi broadcasts demonstrate a significant limitation issue. We propose a direction estimation technique based on Viterbi’s calculation that takes into account the second-by-second location of a moving device as information and, in addition, creates the unmistakably spatio-temporal path taken. In addition, they

Real-Time Big Data Analytics for Improving …

117

Fig. 2 Flow chart of proposed work

demonstrate a few methods that cause passing devices to transmit more messages, increasing location rates, and improving the quality of the usage flag for improved accuracy. Taking everything into consideration, we can conclude that an unmodified cell phone that makes use of WiFi screens is both down to earth, efficient, and accurate. Based on estimates from a few real-world arrangements, we have discovered that a large number of mobile phones may be tracked by using the methods described here. The precision with which inactive WiFi following is performed is highly dependent on the thickness and geometry of the arrangement. However, when comparing our results to GPS ground truth, we were able to achieve a mean error of fewer than 70 m by using displays that were more than 400 m apart. Because of the cheap gear cost, high precision, and broad reach of our suggested framework, we believe it may be suitable for a wide-scale organization in metropolitan areas, assuming that the proposed framework is implemented. Providing near continuous, high-scope estimates of surface street activity stream would certainly be beneficial to suburbanites and organizers in such zones, as would be the case in other zones. This technique allows us to identify, track, and locate mobile phones by detecting

118

V. Arulkumar et al.

the amazing distant marks of WiFi signals that they emit. For example, if we need to know how many people came into a certain store, we may use our technological advancement to find out. Additionally, it allows you to monitor these consumers without identifying them: How long did they stay in the building? What exactly are typical activity designs? We can make use of the distant marks that mobile phones broadcast from time to time via WiFi to our advantage. This means that there is no need for an application or client participation. The following are the stages that this technology can take: Smartphones should be identified and detected as follows: There are many uses for this. All we need to know is the total number of people who pass the test. All things considered, we should be able to implement this technological locator at this location. The finder will then begin compiling a list of mobile phone numbers and IDs. Not every visitor will have access to a mobile phone or a computer with WiFi enabled. Regardless, based on our previous experience, we are aware of the significant disparity between the total number of consumers and the number of mobile phone IDs collected. This means that we can check people with more than 95% accuracy thanks to this technology. Tracking the location of smartphones: Guest identification and tracking are required in more advanced applications where visitors must be distinguished and numbered. For example, we need information on how often a customer visits your retail area.

3.2 Proposed Ideal Beacon Selection In my solution section, we’ve demonstrated the arrangement, which includes the mobile applications that end-users and business owners require, as well as the API that designers can use to create proximity-based mobile applications for these two types of customers. This section describes in detail how each component of the layout should be carried out (devices, programming dialects, and so on). Because two examples were created to test the concept, we will also discuss how they may be used. The code is facilitated on Github, which is a platform that facilitates administration for activities that make use of the code. The following section outlines the critical considerations that went into the development of our Smart Places strategy. Following that, we condense the gadgets that were used in the development of this arrangement. Following that, it is clarified which innovation was used to increase the visibility of labels following our definition of a Smart Place. At that point, we demonstrate how APIs can be used by designers in their work. Following that, we’ll go over how the Smart Retail Shops came to be. Smart Locations: A Smart Place is marked with labels that allow for location-based management. The Smart Retail Shop is an example of Smart Places in action. They were developed following our definition of Smart Place. Customers can gain access to these proximity-based administrations through the use of a portable application, which is also demonstrated in Sect. 3. These administrations, on the other hand, maybe considered portable apps in and of themselves. Alternatively, we might provide an SDK for each step that would allow designers to include proximity-based designers into their apps. If we take this strategy,

Real-Time Big Data Analytics for Improving …

119

we would end up in a situation where the customers would have to install a separate application for each Smart Place that they wanted to use. We must first examine the many methods that might be used to understand why an application is required to get to any Smart Place. Smart Places may be local portable apps or web applications that operate in the background of embedded web software, depending on the situation. A local portable application is a program that is developed using the local instruments that are now available on the stage. This kind of application just continues to function at the stage for which it was designed. The term “web application” refers to an application that operates inside a web-based software. It may continue to operate at any stage when a web-based application is available. Customers would be required to offer a single application for everyone if Smart Places were developed as local apps instead of web-based applications. While walking, they would be unable to discover new nearby-based administrations because of this strategy. To get over this obstacle, we decided that Smart Places would be available as web-based apps. Any online application may thus be configured to react to the presence of labels in a Smart Place. Aside from that, it results in a situation where the clients, as it were, need a single application to get access to any Smart Place, because everyone is a web application that will operate inside an installed web program in the client flexible application. This is done by using the Web View, which is a gadget provided by the Android SDK that allows websites to be embedded within any user interface in an Android program. Smart Places may be accessed via the client’s mobile application. Brilliant places are web-based apps that operate in the background in a Web View inside the client’s multi-functional application. In this Web View, for example, the Smart Restaurant and Savvy Museum examples continue to operate as normal. Clients will not be required to introduce two separate apps in their environments. They will be able to go to any of these examples by using our flexible application. Because we chose web apps as the method by which we want to strengthen Smart Places in some way, we expect them to be able to recognize labels nearby. Where can a web application that runs within a local portable application that is placed inside another web program detect labels in a Smart Place? The labeling guidelines (also known as BLE guides) are handled locally by our method. The essential information is passed to a server-side web application operating inside the inserted web programming.

3.3 Analysis and Results A whole new era is emerging in the retail industry, one that is being propelled forward by the Internet of Things. The Internet of Things, along with massive data analysis, is altering both the route that consumers take and the way that merchants interact with them to provide better service. The Internet of Things is a unique benefit for the retail industry, as it provides merchants with the apparatuses and bits of information they need to transform their organizations. Retailers may significantly improve, computerize, and refine business processes by implementing a successful Internet of Things strategy. They can also reduce operational costs, integrate channels, and,

120

V. Arulkumar et al.

Fig. 3 Proposed beacon architecture

most importantly, better understand and lock in with customers. The fact is that, according to projections, the Internet of Objects will include 30 billion connected self-sufficient things by 2020, and that merchants may find themselves overwhelmed by the possibilities that this technology provides—particularly in the retail sector. It is essential to first develop persuasive business cases with clearly defined objectives, followed by the development of execution plans that make use of your current resources and structure (Fig. 3). To make use of guides, it is necessary to have the proper equipment and programming. Signals need the use of equipment and programming that can transmit and receive reference point signals through Bluetooth Low Energy (BLE). BLE capabilities have been included in iPhone equipment since the release of the iPhone 4 s, and Apple has integrated the ability to communicate with BLE equipment into iOS 5 and the ability to send messages using the iBeacon standard into iOS 7 since that time. 2. Beginning with the release of the Android 4.3 operating system, gadgets running the Android operating system will be able to communicate via Bluetooth Low Energy (BLE) technology. 3. As of the right moment, 30% of mobile phones in the United States are equipped to connect with reference points through Bluetooth Low Energy. This percentage is expected to rise to 80% in the next year and a half as more traditional telephones are replaced with more modern mobile phones, according to projections (see Fig. 3). The use of the iBeacon convention may help to expedite the deployment of BLE-enabled flexible apps. IBeacon is a Bluetooth Low Energy (BLE) standard that was developed by Apple to assist expedite the development of indoor area capabilities in a variety of applications. This standard

Real-Time Big Data Analytics for Improving …

121

enables mobile phones and other equipment that is BLE-enabled to communicate in a language that is indistinguishable from the language spoken by portable apps that utilize BLE. Having this standard in place assists designers in expediting the development of portable apps that make use of reference points, allowing for the release of new functionality to occur more quickly. In addition, the iBeacon standard is being used to take advantage of some of the space capabilities that are built into the iPhones themselves. We have utilized Python to do this, and we have produced the JSON in a nonstandard format, which implies that Spark’s internal reader will not be able to read it. To evaluate, we have done the following: It is necessary for our research that we have two pieces of equipment. Both the audit text, which we must interpret and the general rating, which tells us what score the client really assigned to the item, are important. If we are receiving review text with as little as 53 words and potentially as much as 500 + words, the model should be blocked in order to determine which terms are important rather than how often they appear. In this Sentiment Analysis, we will just use the review text and general data, but it is possible that a more grounded model may be developed by including other components in the analysis process. For starters, since Spark is designed for repeated computations, it facilitates the development of efficient use of large-scale machine learning calculations, which are by their very nature iterative in nature. Low-level upgrades in Spark often result in MLlib’s execution picking up where it left off, with no immediate changes to the library itself. A second benefit of Spark’s vibrant open-source community has been the rapid development and adoption of MLlib, with more than 140 people pledging their support for the project. Third, MLlib is one of a small number of anomalous state libraries built on top of the Spark framework. MLlib’s spark is a component of Spark’s diverse biological community, and it is only partially responsible for the spark. MLlib is an API for pipeline improvement that provides engineers with a broad range of devices to simplify the development of machine learning pipelines on a practical level. Cutting-edge datasets are growing in quantity and unpredictability at an alarming rate, and there is an increasing need for solutions to address this avalanche of information using quantifiable methods. For large-scale information processing, a few “next-generation” information stream motors that summarize MapReduce have been developed, and the question of how to make machine learning effective on top of these motors is one that is very intriguing. Apache Spark, in particular, has grown to become a widely used open-source motor that is widely distributed. Begin is a fault resistant and widely used group registration framework that provides APIs in Java, Scala (for Scala), Python (for Python), and R (for R), as well as an advanced motor that supports up to 204 execution charts in total. Aside from that, Spark is very productive while doing iterative computations, making it an excellent choice for the development of large-scale machine learning applications. In this work, we demonstrate MLlib, Spark’s widely distributed machine learning library, which is also the largest of its kind. The library is designed for large-scale environments that take use of information parallelism or model parallelism to store and

122

V. Arulkumar et al.

operate on information or models on a large scale. Grouping, relapsing, communityoriented separation, bunching, and dimensionality reduction are some of the basic learning computations that may be performed quickly and efficiently in regular learning situations using the MLlib library of functions. It also provides a variety of fundamental insights, direct variable-based arithmetic, and streamlined primitives, among other things. A Scala-based straight polynomial math library that uses local (C++ -based) straight polynomial math libraries on each node, MLlib includes APIs for the Java, Scala, and Python programming languages and is distributed as part of the Spark extension under the Apache 2.0 license. The close integration of MLlib with Spark provides a number of benefits. The fact that Spark is designed for iterative computation, for starters, enables faster progress in the development and production executions of large-scale machine learning calculations, which are often iterative by nature. Spark improvements at the lowest level often result in execution pick-ups in MLlib, with no modifications to the library itself being made right away. Second, Spark’s thriving open-source community has spurred rapid development and adoption of MLlib, with more than 140 people pledging their support for the project thus far. MLlib is one of a small number of anomalous state libraries built on top of the Spark framework. In part because of Spark’s diverse biological population, and in part because of MLlib’s spark, this is a significant feature (Fig. 4; Table 1). The top seven businesses in terms of maximum valuations, Amazon’s top ten highest-volume sellers include: Our second objective is to identify the ten organizations with the highest number of transactions. We will use the company Amazon as an example in our investigation. As a result of this query, we are provided with the ten most significant volumes of “Amazon,” as well as the month and year in which they occurred. In this case, the question is as follows: SELECT a date and a volume FROM the nyse ORDER BY VOLUME WHERE symbol = ‘AMAZON’ WHERE volume = DESC IMITATIONS 10 (Table 2). The results of the aforementioned research reveal the organizations that have reaped the advantages of each and every business. A significant number of investors

Fig. 4 Top 7 companies by maximum values

Real-Time Big Data Analytics for Improving … Table 1 Input sectors with preferred values

Table 2 Volumes traded for Amazon

Company symbol

123 Max_value

BFX

232,356,236

CNM

2306 458999

NML

123,456,789

BHN

112,223,555

MML

1002 000321

SLL

6,566,666

HAL

3,232,356

Month/Year

Volume

AUG-09 478

236,336,173

AUG-09 376

167,235,666

FEB-09 372

321,654,159

MAR-09 293

121,233,808

NOV-08 291

151,515,565

MAR-09 274

337,333,212

AUG-09 478

173,121,233

AUG-09 376

16,723,566

FEB-09 372

1,596,355

MAR-09

255, 867, 900

DEC-09

247, 893, 200

MAR-09

256, 110, 600

MAR-09

255, 867, 900

are interested in putting their money into a company that is doing excellently in the value advertising market. The information should be useful for monetary experts and clients who operate in the money markets and who want to break down all of an organization’s previous records in order to guide their consumers through business initiatives. Our second finding for WFC shows that the organization moved quickly in 2008 and 2009 when comparing the years 2000 to 2014 (Fig. 5). These two years have seen the highest amount of trades in the market. In this study, we showed the probability that Big Data technologies such as Hadoop and Hive would be used by the financial services sector. This method should be particularly beneficial in the area of Business Insight, where the company keeps track of its previous performance as well as other relevant information. Our future study will include the discovery of more money-related information sets in order to find more useful relationships or examples (Fig. 6).

124

V. Arulkumar et al.

Fig. 5 Comparison with iBeacon Technology

Fig. 6 Comparison with iBeacon Technology in certain years

4 Conclusion Likewise iBeacon Technology is more than simply a means of closing a transaction. For example, a design shop may increase consumer confidence in-store by displaying the most current outfits or big-name styles that have been featured in leading design online journals and magazine articles, among other things. This may include suggestions on how to “complete the outfit” by advising on which shoes go well with which trousers, which pack, which beat … all of which are available in-store to purchase. Another example might be a DIY shop, where you can provide clients with setup advice as well as occasional chores for the house or garden, all of which they can do with the help of goods available in the store. These are all examples of messages that are designed to increase sales and make a business more secure in its customers’ loyalty. Increase Customer Loyalty by: Customers may get additional reasons for visiting a shop or moving region by using the resources and technologies mentioned in this study, which can enhance a brand’s fidelity program to an even greater extent. If a customer finds that an item that he or she has tried on

Real-Time Big Data Analytics for Improving …

125

does not suit their requirements, a guide-enabled app may suggest other products and also indicate where they are located in the shop. Customers may also request assistance from a shop employee in an instant by just pressing the appropriate button on their device. Finally, most shops send out weekly notifications about hot discounts or offers that provide additional benefits for customers who have loyalty card accounts. Customers, in any event, tend not to be concerned while they are engrossed in their buying experience in-store, with the exception of a few instances. By using beacons and linking them to the Internet of Things, businesses may notify customers about upcoming arrangements as soon as they walk through the door.

References 1. L. Hand, Business strategy: developing IoT use cases for retail. Published by International Data Corporation (IDC) 3.6. http://www.idc.com/getdoc.jsp?containerId=RI250069 2. L. Dunbrack, S. Ellis, L.H. Kimberly, K.V. Turner, IoT and Digital Transformation: A Tale of Four Industries. Published by International Data Corporation IDC #US41040016 (2016) 3. A. Saha, A study on “The impact of online shopping upon retail trade business”. IOSR J. Bus. Manag. (IOSR-JBM) 74–78 (2015) e-ISSN: 2278-487X, p-ISSN: 2319-7668 4. A. Cuzzocrea, I. Song, K.C. Davis, Analytics over large-scale multidimensional data: the big data revolution!, in Proceedings of the ACM International Workshop on Data Warehousing and OLAP (2011), pp. 101–104 5. E.A. Kosmatos, N.D. Tselikas, A.C. Boucouvalas, Integrating RFIDs and Smart Objects into a Unified Internet of Things Architecture. Adv. Internet Things Sci. Res. 1, 5–12 (2011). https:// doi.org/10.4236/ait.2011.11002 6. M. Zaharia, M. Chowdhury, M.J. Franklin, S. Shenker, I. Stoica, HotCloud 2010, June 2010. MateiZaharia’s Publications 7. U. Han, J. Ahn, Dynamic Load Balancing Method for Apache Flume Log Processing. Adv. Sci. Technol. Lett. 79 (IST 2014), 83–86 2014. https://doi.org/10.14257/astl.2014.79.16 8. A. Bifet, Architectures for massive data management Apache Kafka, Samza, Storm (University Paris Saclay, telecom Paris Tech. 2015) 9. A. Zaslavsky, C. Perera, D. Georgakopoulos, Sensing as a service and Big Data, in IEEE International Conference on Computational Intelligence and Computing Research, pp. 1–6 (2013) 10. F. Chen, P. Deng, J. Wan, D. Zhang, A.V. Vasilakos, X. Rong, Data mining for the internet of things: literature review and challenges. Int. J. Distrib. Sensor Netw. 2015, Article ID 431047, 14 p (2015) 11. S. Madakam, R. Ramaswamy, S. Tripathi, Internet of Things (IoT): a literature review. J. Comput. Commun. 3, 164–173 (2015) 12. S.M. Barakat, Internet of Things: ecosystem and applications. J. Curr. Res. Sci. (2016). ISSN 2322-5009 13. L. Yao, Q.Z. Sheng, Schahram, Web-based management of the Internet of Things. Published by the IEEE Computer Society (IEEE, 2015). 1089-7801/15/$31.00 © 2015 14. D. Evans, The Internet of Things: how the next evolution of the internet is changing everything. White paper published by Cisco Internet Business Solutions Group (2011) 15. C. Chen, B. Das, D.J. Cook, Energy Prediction Based on Resident’s Activity. Washington, DC, USA (2010). Copyright 2010 ACM 978-1-4503-0224-1. 16. J. Cohen, B. Dolan, M. Dunlap, J.M. Hellerstein, MAD Skills: New Analysis Practices for Big Data. ACM VLDB ‘09, 24–28 Aug 2009, Lyon, France. Copyright 2009 VLDB Endowment (ACM, 2009)

126

V. Arulkumar et al.

17. H. Zhou, B. Liu, P. Dong, The technology system framework of the Internet of Things and its application research in agriculture, in 5th Computer and Computing, ed. by D. Li, Y. Chen (2011) 18. W. Liang, J. Luo, Network lifetime maximization in sensor networks with multiple mobile sinks, in Proceedings of the 36th Conference on Local Computer Networks (LCN ‘11), Oct 2011 (IEEE, Bonn, 2011), pp. 350–357 19. M. Moody, Analysis of promising Beacon technology for consumers. Elon J. Undergraduate Res. Commun. 6(1) (2015) 20. F. Zafari, I. Papapanagiotou, K. Christidis, Micro-location for Internet of Things equipped smart buildings. IEEE Internet Things J. 3(1) (2015) 21. V.H. Bhide, A survey on the smart homes using internet of things (IoT). Int. J. Adv. Res. Comput. Sci. Manag. Stud. ISSN: 232 7782 1 (Online) (2014) 22. H.S. Bhosale, D.P. Gadekar, A review paper on big data and Hadoop. Int. J. Sci. Res. Publ. 4(10) (2014). ISSN 2250-3153

Healthcare Application System with Cyber-Security Using Machine Learning Techniques C. Selvan, C. Jenifer Grace Giftlin, M. Aruna, and S. Sridhar

Abstract Compared to last year Internet of Things Intelligent (IoT), this year’s IoT brings a significant increase in intelligence, or “things,” into the Internet of Things (IoT). Regarding the significance of subjective pain inside such an active communication network, we are not yet beyond the reach of artificial intelligence, but we are close. Pain, as well as organs of emotions and ideas (cell phones and tablets), appear alongside home appliances and mobile devices (such as smartphones and tablets). In addition, several of these devices are accessible in markets all over the globe. When it comes to current Internet problems, the source of the pain is the access to Internet Connectivity that they provide. In order to reap the advantages of research capacity solutions, artificial intelligence methods use intelligence. Globally, healthcare services are among the most significant uses that the Internet of Things (IoT) has made possible. In order for patients to monitor their health in real time, advanced sensors may be worn on their bodies or implanted into their organs. Afterwards, the information may be analysed, grouped, and prioritised if necessary. When physicians work with algorithms, they may make adjustments to their treatment plans while simultaneously ensuring that patients get cost-effective health care. Keywords IoT · Health care · Artificial intelligence · Remote patient monitoring · Machine learning C. Selvan (B) Department of Computer Science & Engineering, New Horizon College of Engineering, Bangalore, Karnataka, India e-mail: [email protected] C. Jenifer Grace Giftlin Department of Information Technology, Sri Krishna College of Engineering and Technology, Coimbatore, TN, India M. Aruna Department of Computer Science and Engineering, Faculty of Engineering and Technology, SRM Institute of Science and Technology, SRM Nagar, Kattankulathur, Kanchipuram, Chennai, TN, India S. Sridhar Computer Science and Engineering, Saveetha School of Engineering, Saveetha Institute of Medical and Technical Sciences, Chennai, India © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 I. J. Jacob et al. (eds.), Expert Clouds and Applications, Lecture Notes in Networks and Systems 444, https://doi.org/10.1007/978-981-19-2500-9_9

127

128

C. Selvan et al.

1 Introduction The world’s population is growing at an exponential pace in cities, and the greater the number of people, the greater the amount of strain on resources. Even while there are medical resources and facilities in cities, and they are increasing in number every day, the level of adequacy has not yet been reached, placing an excessive burden on the government’s resources. Cities’ health-care systems have matured, bringing with them the appropriate answers to lightning-related problems in the process. This research represents a new step in e-health, since a large number of sensors are being utilised to develop a new multidimensional monitoring method for the action of a wide range of diseases. It conducts its research using the Raspberry Pi ARM11 (BCM2837), and it is continually developing a comprehensive process of attention for the treatment of illnesses such as cardiac convulsion diabetes, and fever [1]. The antenna continuously captures the gesture along the route, as well as they are then linked to the required statement via physical characteristics, through wireless data acquired starting the network being saved, process, with evaluated in offered health records after they have been linked to the required communication. In the absence of doctors, current data records and decision support systems may be used to conduct a decent diagnostic for first aid advice, and if physicians are not accessible, the devices can predict health problems that are occurring now. According to the database system, we can not only predict the arrival of medicines and medical equipment, but we can also monitor the effect of contemporary technology on the quality of life and health of everyone who makes a difference in the world. Each step taken towards a more accurate disease prediction is done in order to reduce the overall cost of healthcare. Patients will benefit from this article since it offers them with a financial model for technical services as well as philosophical ideas. The real-world medical sector [2] poses a significant obstacle to IoT integration. The lives of Lorem’s pupils have many safety-changing consequences, and by reducing medical costs, it accomplishes that goal while also significantly improving the accuracy of illness prediction in general. A technological tune model, as well as profitable concerns used designed for uncomplaining reassure and open space issues, is presented in this article when implementing IoT in the real world of healthcare [3]. The following is a brief summary of the document’s main goal: It is possible to acquire medical information about a patient via the Internet of Things, and it is also possible towards observing the data that has been collected about the patient. The identification and definition of disease or condition will be accomplished via data mining, starting with specific data that will allow for more effective decisionmaking. • Making healthcare choices based on the Internet of Things at any time and from any place. A number of industries, including manufacturing, transportation, and government, have benefitted from the Learning (ML)/Deep Learning (DL) system3 [3]. Since a

Healthcare Application System …

129

few years, DL has surpassed the state in terms of population. A few of the fields where art is flourishing are computer visualisation, text analysis, and word processing, to name a few. Algorithms like as ML/DL are often used. For example, social media has developed into a technical wonder that we use in our everyday lives (for example, Facebook). The ML/DL calculating method is having an increasingly positive impact on health-care delivery as well. Large-scale technologies have long posed a barrier to the treatment of disorders [4]. Body organ identification in medical imaging, longrange categorization of pneumonia, detection of lung cancer, treatment of imaging reconstructions, and sections of brain tumours are just a few of the domains where machine learning and deep learning have made major strides. To mention a few examples, smart software looks for nearby patients, and machine learning provides medical test results (Fig. 1). Clinician-assisted analysis have risen to the top of the list of possible application areas for ML/DL models, and a number of models have already been developed to fill this need. Human doctors are being phased out in fields like as clinical pathology, radiation therapy, eye disorders, and skin issues, with DL models taking their place. Many research have been published on DL models, which on average outperform human physicians [5, 6] in various areas, according to the findings. Additionally, technology and machine learning/deep learning may assist with result assessment and the development of intellectual solutions that are based on human intelligence. Additional benefits, such as activation, are available in addition. Peripheral medical services are important in the modernization of health-care technology in rural and low-income areas, and they contribute significantly to this effort [7, 8]. Fig. 1 Machine learning clinical overflow

130

C. Selvan et al.

2 Related Work Multiple researchers have suggested different models for IoT in healthcare, as well as techniques for predicting various types of diseases utilising a variety of data sources and methods. It is the purpose of this section to highlight the work that has been done in the same area. Make use of a smart chair to implement an ECG BCG system in the sitting area that will consistently detect physiological issues. By supplying classics, bio signal and a monitoring organization could be used in the exact way that they be intended to be used: without modification. Several instances of Internet of effects (IoT) applications during healthcare will be discussed in this article. Almutairi and his associates A mobile health system that gathers coincident data from mobile devices and enable patients to store it on Internet-enabled network servers with restricted admission to certain consumers was proposed by the author of the paper. This information has been generated and is available for use. Berger et al. [9] are experts in their field. Sensors for smart homes were utilised in the construction of the structure. Hockey and its prototypes are also being evaluated as part of a network that will monitor and track the movements of patients in the future. Their primary aim is to verify the behaviour patterns of their system and to be able to manage and discuss the same thing while at work. Chiuchisan and colleagues created a framework to avoid this from happening. As an intelligent intensive care system threatens them, the patients’ families and doctors voice their concerns about the situation. Incompatibility with one’s health or physical movements, as well as with the surroundings of one’s house is defined as Dwivedi et al. [10] provides a stable foundation for confidence. Clinical data must be provided via the Internet in order to be included in the EPR (Electronic Patient Internet) system. It is suggested to employ a multi-level power information system. The word break is a universal enter arrangement. Biometric technology includes exchanges smart cards, as well as biometric skill to name a few applications. Take, for example, the Gupta family. Gupta et al. [11]: A model for model measurement and recording was suggested, and it was implemented. The usage of an ECG as well as the patient’s other important symptoms are taken into consideration. A raspberry pie may be served at hospitals and other places to patients and their families, such as Gupta, to improve their overall well-being. Intel offers a Galileo-based strategy as a starting point. It is shown by a graph that depicts the evolution of different data and load types. Physicians may and do make use of electronic medical records databases. Efforts are being made to minimise patients’ birth pain, and their health indicators are being monitored on a regular basis [12]. Lopez and his associates proposed an IoT-based framework since he and his team were unable to study and seek out IoT solutions that could be helpful to them and their community. For the most recent Internet of Things test, two use cases were

Healthcare Application System …

131

developed, with the first concentrating on the technologies that could be utilised and the second on the applications that could not be used. Magarelli and Rao planned a new process for assessing the strictness of the patient’s disease which was documented in the patient’s medical record. It was via the use of a statistical technique based on illness probability threshold mining that they reached this result. Moreover, to get to know us their primary goal is to make improvements to the 0 algorithm. It is required to cut the weight of hyperlinks on websites. Sahoo et al. [13] examined the health-care system as well as a massive quantity of patient data from a variety of reports. They acquired a greater knowledge of health attitudes, which they may use to predict the patient’s or prospective health condition in the future. They take use of large information stored in the cloud. There are a variety of approaches that may be used to utilise the same analytics platform. Taigi et al. [14] investigated the health and operation of the Internet of Things. He couldn’t physically recognise her as real and genuine since she didn’t seem real and genuine to him. The use of cloud computing has been proposed as a solution. With permission, we may transmit medical data and patient size in a secure manner. To establish a bond between the sufferer and his or her family, hospitals, doctors, labs, and a variety of other organisations are all engaged in this process. In other words, the patient is financially liable for the clinic for a period of one month. Wang, as well as other online applications tailored to specific medical equipment, have lately been developed with the use of the Internet of Things. Data retrieval is dependent on the quality of the connection. UDA-IoT design methods are described in detail below. The use of information in medical applications has been shown [15]. P2P frameworks and Internet of Things therapy technologies are used to keep patients engaged with the monitoring system and to keep them inside the monitoring system. Real Web Communication (WebRTC) by Kahiki et al. [16] is tested in a variety of conditions, and Al Syndrome gives test results for each. Priority should be located on reliable information routine turn on the Bluetooth control with the help of the sphygmomanometer. SBP was a machine function that kept track of things like systolic blood strain, diastolic blood strain, and cold symptoms. Because of this, the data from the software may be transmitted more easily. Mobile devices and data are archived, disputes are recorded, and public comments are made, among other things. In real-world applications, the slowly approach to end-to-end, real-time Internet architecture poses a confront because of the large number of moving parts. Especially devices are connected towards the seat and supervise the difficulties that user encounter while wearing Bluetooth headset on the way and absent of spectacle resulting in a device that is specifically designed for the visually impaired. The inclusion of another blind sensor [17] provides the user with a comprehensive communication system that ensures precision and readiness during the glove’s extended life. In addition to previous attempts, there is a limited amount of data connectivity across various cloud environments, making it difficult to assess and analyse the data. Given this restriction, we offer a potential solution in this article that takes into consideration.

132

C. Selvan et al.

3 Materials and Methods A patient’s electrocardiogram (ECG), temperature, electromyography, and muscle activity in breathing, perspiration, and blood sugar, as well as infection such as arrhythmias, passion nerve along among muscular disorders, blood stress plumpness and diabetes, may all be monitored with this device. Today, sensors can be easily applied to the skin, and many parts of the body have seen significant improvement, so use caution while using them. A variety of physiological data, including various physiological parameters, is collected via the use of sensors implanted in the bodies of patients [18]. The data is then sent using pre-purchased data and communications software running on a small handheld device. In order to avoid getting in the way of the patient’s movements, sensors should be small and light in weight. It is recommended that these sensors be powered by small, low-energy batteries. This ensures that the sensors may be used forever without the need for shipping or recharging. Transmitting patient data from the health centre’s accurate and secure location should be possible with the right transmission components. Transmission may be accomplished via the usage of Bluetooth. You may also get the information if you go to the online health centre and request it via that channel. Activating the devices linked to the system via the hub, which may be accomplished using a smartphone, is possible [19] (Fig. 2). Fig. 2 Data collection transmission

Healthcare Application System …

133

3.1 Proposed Methodology We are admitted to the government hospital for this task because there are never enough physicians to manage sugar, blood pressure. The pulse and the transfer of heat to the body are both monitored. The work is intended to be used in the administration of a wide range of pharmaceutical products. Blood pressure, tension, with mind rate is all indicators of different types of fever that may be detected via monitoring. The sensors used to measure these limitations are divided into four categories. The four components are as follows: a blood glucose sensor, a high blood strain sensor, a heart rate antenna, and a temperature exchanger. One module [20] is responsible for connecting all of the sensors. The ARM11 gene is responsible for determining whether or not the patient is recognised by the recipient’s physical features and information processing abilities. When the patient accesses his or her profile after receiving the microprocessor, it starts to communicate with him or her via the speaker and to teach him or her how to use medical equipment and to show him or her a video demo. The employee is in charge of a wide range of responsibilities. Toto, the pressure sensor, continuously checks the amount of the fluid in the body and transmits the information to the computer. Heart and healing heart temperatures are maintained at a high level and made available to the customer in a similar manner to the management of blood sugar levels. All expenses are calculated based on data collected by sensors. The MCP3008 is an analog-to-digital converter that may be used in a variety of applications. Discrete numbers are generated by converting the sensor’s calculated data into discrete numbers, which are then sent to the channel makers. For the pump, the image sensor and door have previously been setup to work together. This section may be connected to programmes that facilitate the exchange of information between a doctor and a patient. (See also the Equipment Section.) They are, in fact, true. Developing a high-quality tool. It implements the Internet of Things concept by assigning a unique IP address to a patient database page. According to their doctor’s instructions, the patient may establish an account with an IP address in order to monitor their reading levels. A Wi-Fi unit is worn in the last step of the process, which involves sending readings to the patient’s location. International publication of special editions of mathematics and applications continues to be a priority (Fig. 3). In this case, the camera is linked to the equipment that is responsible for registering the patient’s face or adding facial information to the statement. When it occurs, a comparison between the stored face and an existing face is performed. If a patient’s face information is accessible, the picture opens a window with the patient’s information [21]. It may be used to keep track of a patient’s information. If no information is given, a column will display for them to enter their information, as if they were a brand-new machine. A patient is diagnosed and the employee starts to interact with them as well as other members of the team. Monitoring is used to keep track of and engage with the illness at the location.

134

C. Selvan et al.

Fig. 3 Proposed technology

Employees are responsible for keeping all cervical patients’ medical records up to date. In addition, the WIFI module is operational. Keeping track of all patient information and DMH readings is important for physicians since they may transmit them to medical and Ethernet networks at any time. Certain criteria have been established to enable physicians to diagnose medical data and to hand out gifts to their patients. Next, for future reference, enable doctors to enter their information and create a user ID and password for themselves. Using their login ID and password, physicians may get access to this website, which acts as an additional degree of security [22]. Your doctor may choose to give you a prescription orally or in writing, depending on your preference. Via the use of a microphone, the doctor receives audio input, which is sent to the patient through a speaker. Patients may also get medicines and health cards by contacting the pharmacy via phone, email, or text. They pay close attention to the doctor’s directions or glance at the prescription on the computer screen while they are driving. Without depending on nurses or other professionals, patients may recover their health while also ensuring that they are enrolled properly [9]. In order to determine the lowest theoretical signal intensity required by a receiver for a particular data rate, an established formula must be used. That is the case −154 dBm + 10 log 10(bit rate)

(4.1)

If the data rate is 1 Mbps, then a receiver must have a sensitivity of −94 dBm in order to have a reasonable chance of receiving good data. For example, if the

Healthcare Application System …

135

received signal is in reality at −84 dBm, the fade margin is 10 dB, which means that the received signal may be faded by a factor of up to 10 dB if necessary. Once the data rate has been determined, it is possible to compute the minimum allowable receiver power. The connection between RSSI and distance is shown by the arrow. Fm n P0 Pr F

Fade Margin Path-Loss Exponent, ranges from 2.7 to 4.3 Signal power (dBm) at zero distance Signal power (dBm) at distance Signal frequency in MHz—2412~2483.5 MHz

pi = p0 + 10nlog(ri /r 0)

(4.2)

where The received power in decibels (dBm) at a reference distance is denoted by p0. The route loss exponent, denoted by the symbol r0n, varies from 2 to 4 depending on the transmission medium. The second mode is very essential since it is only during this phase that the user’s authentication is confirmed and they are either granted or refused access to the e-health system. This procedure is carried out by the authentication server. The authentication server validates the authentication by receiving a packet from any one of the IPs listed in the authentication request. When a packet is received by the authentication server, it checks to see whether the specific IP address is on the registered list and also appears on the authentication list. If it does, the packet is rejected. Then it extracts the RSS value, which is compared to the authentication list, and if the value falls within the min and max threshold values, the authentication is successful, and the data is transmitted to the cloud server or received from the cloud server (Fig. 4).

4 Implementation The Internet of Things (IoT) is a computer progression in which each physical object is equipped with a sensor, microcontroller, and transmitter that allow them to communicate with one another. It is produced by putting together a protocol stack. Maintain open lines of contact with one another as well as with clients and clients’ customers. In Internet-based healthcare, a variety of distributed devices collect, analyse, and transmit medical data to the cloud, enabling a massive amount of data to be gathered, stored, and analysed in a number of new ways and by causing contextual disturbances. The Innovative Information Finding Model (IIFM) offers continuous and comprehensive access to medical information through any Internet connection. The

136

C. Selvan et al.

Fig. 4 SetUP phase flowchart

restricted battery life of all Internet-connected gadgets ensures that overall energy usage is maintained to a minimal. The use of the ZigBee mesh protocol to bring the Internet health system into hospitals is discussed in this article. During the implementation of the health-care system, it will be possible to monitor the pathological parameters of hospitalised patients on a regular basis. Equipment that has been well tested and proven, for example, improves overall maintenance quality while simultaneously lowering overall maintenance costs and actively participating in data collection and analysis [23]. The Raspberry Pi is programmed surrounded by Python and transmits health figures to a web server through an Ethernet connection. The name of the patient as well as his or her health condition may be discovered online. The programme makes use of a number of different components.

Healthcare Application System …

137

4.1 Camera Specifications These documents detail the product’s specs in order to verify that it is designed to meet the needs of customers. The AH5020B23-S1-2Z1 is a USB class camera with a video capability developed for laptop photography. The CMOS sensor, lens, holder, support, PCB, image processing channels, and interface, as well as access to digital video devices, are all included. It will be a reliable gadget integrated into a laptop for transferring video data through USB connection. Not only does the AH5020B23-S12Z1 provide UXGA resolution (1600 × 1200) for still image creation applications, but it also provides a video stream for the end user to watch and record a movie through USB 2.0 interface. In YUY2 mode, it can handle VGA resolution (640 × 480) at 30 frames per second. AH5020B23-S1 2Z1 creates AE, AWB, and AGC for CMOS sensors that allow auto image management. It also has a typical UVC UI for picture quality control (UI).

4.2 Comparison with Existing Algorithm The proposed Enhanced Secure Patient Monitoring Algorithm to Session Expired Identity verification in Wireless Personal Area Network is tried to compare to Priority Based Secure Cluster-based Patient Monitoring Algorithm (PBSCHCMA) (Sethupathi et al. 2015) [24]. A Secure Priority Based Treatment Monitoring Algorithm (SPBHCMA) (Sethupathi et al. 2015), and Light Weight Security Architecture (LSA) (Sethupathi et al. 2015) bench 1 show the results of measuring the delay time with various key lengths (Table 1). Because of improved routing and decreased queuing time, the proposed ESHSEA algorithm performs well in measuring the delay when compared to current methods. With a key length of 50, a delay of 0.64 s is produced, which is smaller than the LSA, SPBHCMA, and PBSCHCMA. ESHMA has a shorter delay period than the other algorithms. ESHMA has a delay time of 4.19 s even with a key length of 250, but LSA, SPBHCMA, and PBSCHCMA have delays of 4.8 s, 4.32 s, and 4.2 s, respectively (Fig. 5; Table 2). Table1 Key length versus delay time Key length 50

Delay (s) LSA

SPBHCMA

PBSCHCMA

ESCHMA

ESHMA

0.9

0.7

0.65

0.63

0.64

100

2

1.78

1.56

1.49

1.52

150

4

3.69

3.27

3.27

3.23

200

4.2

4.06

3.82

3.82

3.78

250

4.8

4.32

4.2

4.2

4.19

138

C. Selvan et al.

Fig. 5 Key length versus delay time

Table 2 Key length versus throughput Key length

Throughput (bits) LSA

SPBHCMA

PBSCHCMA

ESCHM A

ESHM A

50

504,402

542,425

584,526

586,424

586,420

100

424,892

444,088

486,744

492,754

492,744

150

408,446

429,829

442,426

452,209

452,098

200

457,984

468,795

484,687

484,687

487,445

250

267,892

277,688

284,567

284,567

284,646

Throughput: Table 2 shows the throughput numbers achieved when modelling the ESHMA. When compared to existing techniques, the proposed ESHMA algorithm outperforms them in terms of throughput measurement, as shown in Fig. 5. Comparing the results of the ESHMA’s throughput measurement to those obtained from the LSA, PBSCHCMA, and SPBHCMA, the results are promising. The red line in Fig. 6 shows that ESHMA has a higher throughput value than the other techniques that were taken into consideration. Figure 6 shows that, when compared to current methods, the proposed ESHMA algorithm works better in measuring the PDR since transmission congestion is lower. PDR is 95.14 for LSA, 96.28 for SPBHCMA, 98.02 for PBSCHCMA, and 98.43 for the ESHMA algorithm with key lengths of 50. ESHMA produces comparable better outcomes with key lengths of 100, 150, 200, and 250.

Healthcare Application System …

139

Fig. 6 Key length versus packet delivery ratio

5 Conclusion Using the cloud-based information system, the suggested solution aspires to provide patients with better-connected economic health services, allowing specialists and doctors to build on this knowledge and reach more fast and favourable conclusions for their patients. The final model contains all of the characteristics that a doctor is looking for in a patient at any given time. Therefore, the applicable expert would take action against the healthcare victim in the clinic as a result of the connected economic assistance to ill nations, resulting in reduced hospital lines and direct consultation with physicians, reducing contextual dependence, and permitting full use of the website. a. The primary goal of the approach is to provide patients with a high-quality financial life that is connected to their services via the network’s cloud of information, which comprises professionals from many fields. Doctors may be able to utilise this information to provide their patients with a quick and cost-effective resolution. Several characteristics of the final product enable the doctor to test the patient from any place and at any time. This would result in an economic advantage designed used in favour of sick people who want to cross the margin into hospital and have through consultations among doctors in order to reduce the consistency of their families’ health-care costs. This proposed method would be cost-effective in terms of ensuring proper health management in public hospitals, which is the goal. Through the process of replication, more components of the artificial intelligence system may be enhanced. Both patients and physicians are involved. The majority of people’s medical histories include parameters and appropriate results, and data mining is used to search for templates all of the time, as well as for systemic disease connections, among other things. Example: if the patient’s health measurements change in the same manner as those of previous patients in the database, it is possible to predict the result. If there is a trend like this, we will be able to recognise it immediately, and

140

C. Selvan et al.

physicians will have an easier time identifying it. Medical experts have discovered a way out to this problem.

References 1. H.T. Sullivan, S. Sahasrabudhe, Envisioning inclusive futures: technology-based assistive sensory and action substitution. Futur. J. 87, 140–148 (2017) 2. Y. Yin, Y. Zeng, X. Chen, Y. Fan, The Internet of Things in healthcare: an overview. J. Ind. Inf. Integr. 1, 3–13 (2016) 3. H.N. Saha, S. Auddy, S. Pal, Health monitoring using Internet of Things (IoT). IEEE J. 69–73 (2017) 4. S.F. Khan, Health care monitoring system in Internet of Things (loT) by using RFID, in IEEE International Conference on Industrial Technology and Management (2017), pp. 198–204 5. M. Hassanalieragh, A. Page, T. Soyata, G. Sharma, Health monitoring and management using Internet-of-Things (IoT) sensing with cloud-based processing: opportunities and challenges (2015) 6. M.S.D. Gupta, V. Patchava, V. Menezes, Healthcare based on iot using raspberry pi, in 2015 International Conference on Green Computing and Internet of Things (ICGCIoT), Oct 2015, pp. 796–799 7. P. Gupta, D. Agrawal, J. Chhabra, P.K. Dhir, Iot based smart healthcare kit, in 2016 International Conference on Computational Techniques in Information and Communication Technologies (ICCTICT), Mar 2016, pp. 237–242 8. N.V. Lopes, F. Pinto, P. Furtado, J. Silva, Iot architecture proposal for disabled people, in 2014 IEEE 10th International Conference on Wireless and Mobile Computing, Networking and Communications (WiMob), Oct 2014, pp. 152–158 9. R. Nagavelli, C.V. Guru Rao, Degree of disease possibility (ddp): a mining based statistical measuring approach for disease prediction in health care data mining, in International Conference on Recent Advances and Innovations in Engineering (ICRAIE-2014), May 2014, pp. 1–6 10. P.K. Sahoo, S.K. Mohapatra, S.L. Wu, Analyzing healthcare big data with prediction for future health condition. IEEE Access 4, 9786–9799 (2016). ISSN 2169-3536 11. B. Krishnan, S.S. Sai, S.B. Mohanthy, Real time internet application with distributed flow environment for medical IoT, in International Conference on Green Computing and Internet of Things, Noida (2015), pp. 832–837 12. V. Arulkumar, C. Puspha Latha, D. Jr Dasig, Concept of implementing big data in smart city: applications, services, data security in accordance with Internet of Things and AI. Int. J. Recent Technol. Eng. 8(3) (2019) 13. D. Azariadi, V. Tsoutsouras, S. Xydis, D. Soudris, ECG signal analysis and arrhythmia detection on IoT wearable medical devices, in 5th International Conference on Modern Circuits and Systems Technologies, Thessaloniki (2016), pp. 1–4 14. A. Mohan, Cyber security for personal medical devices Internet of Things, in IEEE International Conference on Distributed Computing in Sensor Systems, Marina Del Rey, CA (2014), pp. 372– 374 15. L.Y. Yeh, P.Y. Chiang, Y.L. Tsai, J.L. Huang, Cloudbased fine-grained health information access control framework for lightweight IoT devices with dynamic auditing and attribute revocation. IEEE Trans. Cloud Comput. 99, 1–13 (2015). IoT-based health monitoring system for active and assisted living 19 16. V. Arulkumar, An intelligent technique for uniquely recognising face and finger image using learning vector quantisation (LVQ)-based template key generation. Int. J. Biomed. Eng. Technol. 26(3/4), 237–49 (2018)

Healthcare Application System …

141

17. P. Porambage, A. Braeken, A. Gurtov, M. Ylianttila, S. Spinsante, Secure end-to-end communication for constrained devices in IoT-enabled ambient assisted living systems, in IEEE 2nd World Forum on Internet of Things, Milan (2015), pp. 711–714 18. K. Yelamarthi, B.P. DeJong, K. Laubhan, A kinect-based vibrotactile feedback system to assist the visually impaired (2017) 19. X.-W. Chen, X. Lin, Big data deep learning: challenges and perspectives. IEEE access 2, 514–525 (2014) 20. M. Siekkinen, M. Hiienkari, J. Nurminen, J. Nieminen, How low energy is bluetooth low energy? Comparative measurements with zigbee/802.15.4, in Wireless Communications and Networking Conference Workshops (WCNCW) (IEEE, 2012), Apr 2012, pp. 232–237 21. N. Bui, M. Zorzi, Health care applications: a solution based on the internet of things, in Proceedings of the 4th Int. Symposium on Applied Sciences in Biomedical and Communication Technologies, ser. ISABEL’11 (ACM, New York, 2011), pp. 131:1–131:5 22. K. Laubhan, M. Trent, B. Root, A. Abdelgawad, K. Yelamarthi, A wearable portable electronic travel aid for the blind, in IEEE International Conference on Electrical, Electronics, and Optimization Techniques (2016) 23. M. Li, S. Yu, Y. Zheng, K. Ren, W. Lou, Scalable and secure sharing of personal health records in cloud computing using attribute based encryption. IEEE Trans. Parallel Distrib. Syst. 24(1), 131–143 (2013). C. Bishop, Pattern Recognition and Machine Learning. Springer, New York (2006) 24. V. Arulkumar, C. Selvan, V. Vimal Kumar, Big data analytics in healthcare industry. An analysis of healthcare applications in machine learning with big data analytics. IGI Global Big Data Analyt. Sustain. Comput. 8(3) (2019)

Energy Efficient Data Accumulation Scheme Based on ABC Algorithm with Mobile Sink for IWSN S. Senthil Kumar, C. Naveeth Babu, B. Arthi, M. Aruna, and G. Charlyn Pushpa Latha

Abstract The current trend in Wireless Sensor Network (WSN) is based on multihop networking which is used to transmit data through various networks. The usage of multihop forwarding in large-scale WSNs cause an energy hole problem, which results in a considerable amount of transmission overhead. In this paper, a multiple portable sink-based information gathering method that combines energy balanced clustering as well as Artificial Bee Colony-based data gathering is proposed in order to address these concerns. The remaining energy of the node is used to determine which node will serve as the cluster’s centre of gravity. According to the findings of this research, mobile sink balancing may be approached from three different perspectives: data gathering expansion, mobile route distance reduction, and network reliability optimization. This study is conducted with the use of a significant and intense WSN that enables a specific level of data delay to be tolerated in order to be successful. The paper proposes the optimization technique which is known as Artificial Bee Colony optimization technique that can accept the reduction losses in data communication, improve network lifetime, save the energy of the system, maintain the reliability of the system, and increase the network efficiency. Keywords Ant colony optimization · Mobile sink · Wireless sensor networks · Efficiency S. Senthil Kumar (B) Department of Computer Science, Sri Ramakrishna Mission Vidyalaya College of Arts and Science (Autonomous), Coimbatore, India e-mail: [email protected] C. Naveeth Babu Department of Computer Science, Kristu Jayanti College, (Autonomous), Bangaluru, Karnataka, India B. Arthi · M. Aruna Department of Computer Science and Engineering, Faculty of Engineering and Technology, SRM Institute of Science and Technology, SRM Nagar, Kattankulathur, Kanchipuram, Chennai, Tamil Nadu, India G. Charlyn Pushpa Latha Department of IT, Saveetha School of Engineering, Saveetha Institute of Medical and Technical Sciences, Chennai, India © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 I. J. Jacob et al. (eds.), Expert Clouds and Applications, Lecture Notes in Networks and Systems 444, https://doi.org/10.1007/978-981-19-2500-9_10

143

144

S. Senthil Kumar et al.

1 Introduction The IWSN is one of the most popular WSN applications, and it makes use of several sensor nodes. Temperature sensors, sound sensors, vibration sensors, pressure sensors, and other types of sensor nodes are used to monitor and track data. In case anything gets out of control, the sensor nodes alert the control system by sending data to the system. At this point, the most significant parameter is the passage of time. Communication latency must be eliminated in an industrial setting since it is critical to the operation of the business [1]. As a result of their low energy capacity, the sensors’ energy consumption must be kept to a bare minimum in order to save as much energy as possible throughout the operation. The greater the efficiency with which energy is conserved, the longer the network’s lifespan. To reduce the amount of energy used by data transmission, there are a number of methods described in the literature. The use of a sink node is one of the most frequently used methods to improve communication quality in networks [2]. The sink node may be either stationary or movable. In accordance with Lu et al. and Gungor et al., the immobile sink nodes stay in a particular position at all times, and as a result, the nodes near the sink node drain more energy due to the data flow [3]. This disadvantage is addressed by the mobile sink, which is a device that travels around the network and collects data. Energy consumption of the nodes across the network is reduced because of this concept. In order to take use of the advantages of mobile sinks, this study utilizes a mobile sink node for data collection [4]. The mobile sink, on the other hand, travels the network to gather data from every single node. This increases the latency and the amount of time it takes for data to reach the sink node. In order to cope with this problem, this work makes use of the idea of clustering, which means that all of the nodes are grouped together into a single cluster, which is controlled by a node known as the cluster head node, which is in charge of multiple participant nodes. It is possible for all the cluster’s participant nodes to communicate with the cluster head node, and for the cluster head node to communicate with the mobile sink. This concept improves manageability while also reducing communication latency to a bare minimum. Because of this, energy consumption is decreased, and the network’s life expectancy has been extended. The primary goal of this study is to reduce communication latency while also reducing energy usage in order to extend the network’s lifetime. Two critical stages of this study are cluster formation and data collection by a mobile sink node, both of which are discussed in detail below. By reaching out to all the clusters, the mobile sink node accumulates data. The Artificial Bee Colony (ABC) [5] method is used to determine the optimal route for data collection by mobile sink nodes. The work’s performance is then evaluated in terms of communication throughput, latency, energy usage, time consumption, and network lifespan, among other metrics. Currently existing methods for harmonizing the energy ingesting of devices are too complicated to be used in practice due to the many boundaries forced on WSNs for dissimilar purposes, rendering them unfeasible for majority of the

Energy Efficient Data Accumulation Scheme …

145

applications. In a broad variety of actual applications, the performance of existing algorithms is significantly hampered by these limitations. For example, a sensor network with many mobile sinks is being built for the purpose of monitoring a distant region, and there is a local connected road map for all of the mobile sinks in the monitored zone, there would be a problem in scheduling these mobiles descends to assemble abundant data from devices feasibly to extend the system’s generation, and this problem must be resolved. In this paper, it is suggested to build a new constrained optimization problem for wireless sensor networks, which may be used to solve the energy management optimization problem of many mobile sinks that have limited energy availability. Following that, an effective approach is proposed for dealing with the problem, mentioned as the Artificial Bee Colony grounded portable sink undertaking algorithm. Using this method, not only is the assignment between the mobile sinks and sensor nodes are balanced, but also the energy ingesting between the sensor nodes are stable. It is proposed in this article that a mobile wireless sensor network data group technique, based on the artificial bee colony procedure, may be utilized to collect data from a mobile wireless sensor network in an efficient and reliable manner. To summarize the most significant contributions made by this essay, the following are the bullet points: This paper discusses the data collection technique for mobile sinks, cluster head assortment glitches, and mobile sink route optimization. In order to recover the efficacy of network data gathering, the route optimization of the movable sink may be phrased as a straight pathway discovery problem in order to maximize data collection efficiency. The artificial bee colony technique may then be used to search for structures of the optimal resolution, as well as the straight route of the mobile sink, in order to find the best solution. In their natural habitat, honeybees engage in a range of complex behaviours, including mating, breeding, and foraging. It has been feasible to simulate the behaviour of honeybee-based optimization techniques after this, which has been done for several different approaches. A technique for route optimization based on bee colony optimization is employed in the proposed study because of the complicated behaviour. The following organization is discussed in detail in the upcoming part of this article: Sect. 1 illustrates the introduction. Section 2 gives a survey of important literature in the area of research. Section 3 offers a description of the system model and its components. Section 4 presents a mobility-based energy efficiency algorithm, which is based on the fact that people are always moving. It is shown in Sects. 5 and 6 how the numerical results and conclusions are obtained, and how they are reached.

2 Literature Survey Many studies have been performed and published in the past on the deployment of mobile sinks for data gathering. An NP-hard issue [6] is a problem in which arranging

146

S. Senthil Kumar et al.

nodes into optimum clusters is very difficult to solve successfully. It is worthwhile to consider optimizing the size of the cluster since it is an efficient way of reducing its overall size. In order to control the size of a cluster, it is necessary to minimize the distances among member nodes and Cluster Heads (CHs) [7], as well as the amount of energy used by the network as a whole. PSO decreases the distance between member nodes and CHs in directive to determine the optimal cluster size. Based on the results of [8], GA has been utilized to construct the most energy-efficient clusters feasible in wireless sensor networks. It was suggested by the writers of [9] that a self-clustering technique for varied WSNs be employed in conjunction with a GA to extend the lifespan of the network. Other ways of altering cluster size are used in order to achieve a variety of goals. The researchers believe that a routing tree that maximizes network lifespan while possessing the direction-finding route between each device and sink to the bare minimum should do this [10]. An energy consumption cost model presented in [11], with an evaluation of the queries carried out via a routing tree. Aims of the research in [12] include finding the most efficient path for each movable sink and computing the break duration of a piece mobile sink along the trajectory in order to maximize the network’s lifespan. For example, in [13], the usage of a solitary mobile descend to extend the life of a network was investigated in more detail. The findings of a recent study published in [14] indicate a method for prolonging the life of mobile base stations while also controlling network demand at the same time. [15, 16] recommends that the mobile base station be operated in the manner of the greediest algorithm, as an example. Joint flexibility and steering are suggested [17–19] in the situation of a stationary base station with a restricted range of capabilities. Ultimately, the route is controlled by [20–22] via the use of the mobile sink routing protocol once all has been said and done. Mobile sinks, on the other hand, have a very limited range of practical applications in practice. The fact that a WSN is often situated in densely populated areas means that mobile sinks can only serve a tiny part of the population. Because of a variety of factors, including road conditions and mobile tools, these autonomous cars can only go at a particular speed. Aside from that, since mobile tools have a limited amount of available energy, the greatest distance they can go on a journey is also limited. These limitations on mobile sinks, are taken into account while developing a routing protocol in order to maximize network lifespan while still maintaining speed. An intensive care unit’s collection of n randomly distributed similar device nodes, which is represented mathematically by the purposeless graph n = (VMS, E), where V represents the collection of n randomly distributed similar device nodes in the intensive care unit and MS represents a mean temperature of the monitoring zone. In MS, there are k mobile sinks, and in E, there are k connections between sensors and sinks, with k being the number of connections between sensors and sinks [23]. It is considered to be connected when the broadcast variations of two sensors u and v coincide; otherwise, it is said to be unlinked. When it comes to remote monitoring, mobile sinks are one-of-a-kind sensor since they can both receive sensed information from device nodes and send the sensory data they have collected to an offsite monitoring centre. If each sensor node has a unique identity and a limited beginning

Energy Efficient Data Accumulation Scheme …

147

energy capacity Q, it is assumed that each mobile sink has a lively capacity and an unlimited supply of energy at the depot. To make things easier to understand, it is assumed that there are k mobile sinks in the system at any one point in time. R = (V r , W r ), where each vertex in V r represents a road junction set and each piece advantage in W r represents a road segment, used to represent the road map in the monitoring region (Fig. 1). Map R can only include a maximum of k mobile sinks, each of which has a liveliness capacity of eM, and as a result, it is limited in its capacity. Travel and data transmission, as well as the reception of the information supplied, account for about one-third of the total energy used by each mobile sink. ‘et’ and ‘ec’ are used to indicate the sink energy consumption for a unit-length journey at the sojourn sites along the route, while ‘et’ and ‘ec’ are used to denote the sink energy consumption for a unit-length trip at the starting point [24]. The wireless connection between sensor nodes consumes the vast majority of the energy generated by the sensor nodes when it comes to energy consumption rates. The remainder of the energy use, which comprises sensing and processing, is negligible.

Control Room

Mobile Sink Node

CH

CH

CH

28 14

20

11

2

19

9

1

10

4 13

5

3

27 22

12

16 17

25 24

18

7 8 6

21

15

26

23

Node Clusters

Fig. 1 Overall flow diagram of the proposed data accumulation scheme

148

S. Senthil Kumar et al.

3 Materials and Methodology This section describes the proposed data accumulation system, which is based on the stages of clustering and ideal route selection, in more detail. The clustering phase aims to bring the nodes together under the control of a cluster head node, which is in charge of managing the cluster. The task selection method, is used to carry out the clustering process. Figure 1 depicts an overview of the planned work’s overall flow diagram. It can be readily seen in the above-presented picture that the nodes are initially grouped, and that the cluster head nodes may communicate with the mobile sink node by using the optimal route selected by the ABC algorithm to do so. Periodically, the data is sent from the mobile sink to the control room. The computational complexity and time consumption for data transmission are reduced by this concept.

3.1 ABC Algorithm ABC is a metaheuristic algorithm that is designed to imitate the natural behaviour of honeybees. It was developed by Karaboga and is based on his research. The food supply, the amount of working and unemployed bees, and the number of employed bees are all essential factors in the ABC algorithm. The primary goal of this algorithm is to locate the most suitable food source. Food source: A food source may be designated by taking many factors into consideration, including the distance among the food and the hive, the quality of the food, and the ease with which the energy can be absorbed. A food source is determined to be the best based on the criteria listed above. Bees on the job: The bees on the job are responsible for spreading critical information about the food source. The majority of the information exchanged is the location of the food supply and the distance between the food source and the hive. When a swarm of bees is taken into consideration, the hired bees are responsible for half of the swarm, with the other half being taken care of by spectator bees. Jobless bees may be divided into two types: scout bees and spectator bees. Scout bees are the most common kind of unemployed bee. The scout bees are on the lookout for a new food source in the vicinity of the beehive. The observer bees remain in the hive and use the information gleaned from the employed bees to identify the source of food for the colony. The location of the food supply is often found to be the optimal answer to the analysis issue, and the quantity of honey present in the food determines the excellence of the meal produced. The fitness function of the algorithm is determined by the quality of the meal. Using the theme of employed bees, the algorithm determines that employed bees are concerned with the food source. The following is a representation of the standard pseudocode for the ABC algorithm in Table 1.

Energy Efficient Data Accumulation Scheme … Table 1 ABC algorithm

149

Standard Pseudocode of ABC Algorithm 1: Produce initial population i p = 1 to MC 2: Calculate the fitness function of the population 3: Fix counter = 1 4: Do //Employed bees phase 5: Search for the food source; 6: Calculate the fitness function; 7: Employ greedy selection process; 8: Compute the probability for the food source; // Onlooker bees phase 9: Select food source based on the probability values; 10: Generate new food source; 11: Calculate the fitness function; 12: Apply greedy selection process; //Scout bees phase 13: If food source drops out then swap it with new food source; 14: Save the best food source; 15: Counter + = 1; 16: While counter = MC;

Scout bees are responsible for conducting the first search for food sources. As soon as this phase is completed, the observer and the hired bees begin to exploit the food supply. The food supply becomes depleted after continuous exploitation, and the hired bees are relegated to the role of scout bee. The number of employed bees is always the same as the number of food sources, since each employed bee is linked with a single food source. Generally speaking, the most fundamental ABC method is divided into three phases: initialization; employed; onlooker and scout bees phase. Every step is repeated until the maximum number of iterations has been reached in a certain period of time. Initially, it is necessary to determine the total number of solutions available as well as the control parameters. The employed bees look for new food sources of higher quality in the vicinity of the old food source in which they were previously engaged. In the next step, the new food source is evaluated for its fitness, and the results are then compared to the previous food source with the aid of greedy selection. The information regarding the food source that has been gathered is disseminated among the observer bees who are present in the beehive. As part of their decision-making process, the spectator bees use a probabilistic method to choose food sources based on the information provided by the employed bees during the second phase. This phase is followed by the calculation of the fitness function of the food source that is situated close to the food source that was selected in the

150

S. Senthil Kumar et al.

previous stage. The greedy selection compares and contrasts the old and new food sources available. Finally, when it is not feasible to improve the answers within a certain number of iterations, the hired bees are promoted to the position of scout bees. The solutions that have been identified by the bees are discarded. At this moment, the scout bees begin their hunt for a new food source, and the bad options are removed from consideration. According to Karaboga and colleagues, all three stages are repeated until the stopping point is reached.

3.2 Proposed Ideal Path Selection by ABC Algorithm The mobile sink node has to handle multiple cluster head nodes for accumulating the data. Let the cluster head nodes be represented by ϕ = {CH1 , CH2 , CH3 , . . . , CHn }. The traversing path of the sink node must be detected, so as to reach all the CH nodes. The choice of route is finalized by considering two parameters, which are distance between the mobile sink and the CH node, and the energy cost. With these parameters, the fitness function is built as follows. f v(i ) =

i (ER(PT)) Dis(i, ms)

(1)

In the above equation, i is the cluster head, ER(PT) is the energy required to transmit a message and Dis(i, ms) is the distance between the cluster head and the mobile sink node. Based on this fitness function, the proposed ABC based path selection algorithm is presented as follows. The probability of the food source is computed by the following equation. P(FS) = LN(OS) + α(LN(OS) − LN(NS))

(2)

In the above equation, LN(OS) and LN(NS) are the locations of old and new cluster heads, α is a parameter that range between 1 and −1. f vi probi =  f s n=1

f vn

(3)

In the above equation, f vi is the fitness value of the ith cluster head node and n is the total number of cluster head nodes (Table 2). Using the algorithm, the ideal path is selected with the help of energy gain and distance between the mobile sink and cluster head. This idea conserves energy and reduces the communication delay. The performance of the proposed work is evaluated in the following section.

Energy Efficient Data Accumulation Scheme … Table 2 Proposed ideal path selection algorithm

151

Proposed Ideal Path Selection Algorithm Input: Clustered Nodes; Output: Ideal Path Selection; Begin 1: Produce initial population i p = 1 to MC 2: Calculate the fitness function of the population by Eq. (5.1) 3: Fix counter = 1 4: Do //Employed bees phase 5: Search for the food source; 6: Calculate the fitness function by Eq. (5.1); 7: Employ greedy selection process; 8: Compute the probability for the food source by Eq. (5.3); // Onlooker bees phase 9: Select food source based on the probability values; 10: Generate new food source; 11: Calculate the fitness function; 12: Apply greedy selection process; //Scout bees phase 13: If food source drops out then swap it with new food source; 14: Save the best food source; 15: Counter + = 1; 16: While counter = MC;

3.3 Analysis and Results The performance of the work is evaluated by implementing the work in NS2, on a standalone computer. The parameters chosen to carry out the simulation are presented in Table 3. Table 3 Simulation parameters

Parameter

Settings

Network area

200 × 200

Initial energy of nodes

2J

Communication radius

50 m

Energy for data transmission

50 nJ/bit

Speed of mobile sink

2 m/s

152

S. Senthil Kumar et al.

The performance of the proposed work is analyzed in terms of throughput, latency, energy consumption, time consumption, and network lifetime. The results attained by the proposed work are compared with the existing approaches such as DDRP and SGDD. The attained experimental results are presented as follows. From Figs. 2, 3, 4 and 5, the planned work’s throughput and average delay, its energy consumption and network lifespan can be determined. The experimental findings demonstrate that the suggested approach results in increased throughput and network lifespan while reducing latency and energy consumption to an acceptable level. The entire amount of data that is sent in a particular period of time is referred to as throughput. Regardless of the data transfer method used, the throughput must be as high as possible. The latency of data transmission should be kept to a minimum for the data to be delivered on time. Due to the understanding of industrial submissions, it is essential to guarantee that data transmissions have the shortest possible latency. This study achieves minimum latency via the application of two concepts: the use of Fig. 2 Throughput analysis

Fig. 3 Average latency analysis

Energy Efficient Data Accumulation Scheme …

153

Fig. 4 Energy consumption analysis

Fig. 5 Network lifetime analysis

sink nodes and the selection of the optimal route for sink nodes to go to and from the cluster head node. After that, the energy consumption of the planned task is evaluated in relation to the simulation duration. Energy consumption of the nodes increases with time, as shown in Fig. 4. When the suggested method is compared with others that have been used before, it uses the least amount of energy, due to the optimal route selection that takes into account both energy gain and distance metrics. As the energy usage decreases, it becomes apparent that the network’s lifespan will be extended. The lifespan of a network is assessed in terms of the number of nodes that are still operational in the network. There are about 180 living nodes in the proposed work at the conclusion of the 500th second, while the competing methods had 124 and 163 alive nodes, respectively. It is calculated and reported in Table 4 the time taken for the final node to die.

154 Table 4 Network lifetime analysis with respect to time

S. Senthil Kumar et al. Sensor count

DDRP (s)

SGDD (s)

Proposed (s)

100

2134

2304

2473

200

2581

2864

3021

300

2867

3215

3543

400

3142

4046

4873

500

3314

4628

6043

600

4017

5537

6943

700

4324

6242

8261

800

6219

7634

10,262

The proposed work has a total duration of 10,262 s, while the competing works DDRP and SGDD have a duration of 6219 s and 7634 s respectively. It is calculated and shown in the table how long the network will last in its entirety. According to the experimental findings, it has been shown that the suggested work has a longer lifespan as a consequence of the use of mobile sink nodes and optimal route selection ideas.

4 Conclusion This work describes an energy-efficient data-accumulation strategy for IWSN that makes use of a mobile sink to collect information. Every step of the process is based on clustering and optimal route selection. The nodes are grouped with the assistance of the task selection algorithm, and the ABC algorithm is used to determine the optimal route for the mobile sink to take in order to reach the cluster head. The work’s overall performance is slow in terms of amount, regular delay, energy consumption, and network lifespan.

References 1. G. Xing, M. Li, T. Wang et al., Efficient rendezvous algorithms for mobility-enabled wireless sensor networks. IEEE Trans. Mob. Comput. 11(1), 47–60 (2012). https://doi.org/10.1109/ TMC.2011.66 2. Y. Yue, J. Li, H. Fan, et al., Optimization-based artificial bee colony algorithm for data collection in large-scale mobile wireless sensor networks. J Sens. 2016, Article ID 7057490, 12 (2016). [Web of Science ®], [Google Scholar] 3. L. Malathi, R.K. Gnanamurthy, K. Channdrasekaran, Energy efficient data collection through hybrid unequal clustering for wireless sensor networks. Comput. Electr. Eng. 48, 358–370 (2015). https://doi.org/10.1016/j.compeleceng.2015.06.019 4. S.J. Tang, J. Yuan, X.Y. Li, et al., DAWN: energy efficient data aggregation in WSN with mobile sinks, in Proceedings of the IEEE 18th International Workshop on Quality of Service (IWQoS ‘10) (IEEE, Beijing, China, 2010). pp. 1–9

Energy Efficient Data Accumulation Scheme …

155

5. M. Ma, Y. Yang, Sencar: an energy-efficient data gathering mechanism for large-scale multihop sensor networks. IEEE Trans. Parallel Distrib. Syst. 18(10), 1476–1488 (2007). https://doi.org/ 10.1109/TPDS.2007.1070 6. S. Basagni, A. Carosi, E. Melachrinoudis et al., Controlled sink mobility for prolonging wireless sensor networks lifetime. Wirel Netw. 14(6), 831–858 (2008). https://doi.org/10.1007/s11276007-0017-x 7. S. Basagni, A. Carosis, E. Melachrinoudis, et al., A new MILP formulation and distributed protocols for wireless sensor networks lifetime maximization, in Proceedings of the IEEE International Conference on Communications (ICC ‘06) (IEEE, Istanbul, Turkey, 2006), pp. 3517–3524 8. H.T. Nguyen, L. Van Nguyen, H.X. Le, Efficient approach for maximizing lifespan in wireless sensor networks by using mobile sinks. ETRI J. 39(3), 353–363 (2017). https://doi.org/10. 4218/etrij.17.0116.0629 9. J. Luo, J.-P. Hubaux, Joint sink mobility and routing to maximize the lifetime of wireless sensor networks: the case of constrained mobility. IEEE/ACM Trans Networking. 18(3), 871–884 (2010). https://doi.org/10.1109/TNET.2009.2033472 10. J. Luo, J. Panchard, M. Piórkowski, et al., Routing towards a mobile sink for improving lifetime in sensor networks, in Proceedings of Distributed Computing in Sensor Systems: 2nd IEEE International Conference, DCOSS 2006, vol. 4026 (San Francisco, CA, USA, 2006) 11. B. Bhushan, G. Sahoo, E2 SR2: an acknowledgement-based mobile sink routing protocol with rechargeable sensors for wireless sensor networks. Wirel. Netw. 25(5), 1–25 (2019). https:// doi.org/10.1007/s11276-019-01988-7 12. R. Mitra, S. Sharma, Proactive data routing using controlled mobility of a mobile sink in wireless sensor networks. Comput. Electr. Eng. 70, 21–36 (2018). https://doi.org/10.1016/j. compeleceng.2018.06.001 13. J. Wang, J. Cao, R.S. Sherratt et al., An improved ant colony optimization-based approach with mobile sink for wireless sensor networks. J. Supercomput. 74(12), 6633–6645 (2018). https:// doi.org/10.1007/s11227-017-2115-6 14. R. Akl, U. Sawant, Grid-based coordinated routing in wireless sensor networks, in 4th IEEE Consumer Communications and Networking Conference (CCNC ‘07) (2007), pp. 860–864 15. S. Sharma, On energy efficient routing protocols for wireless sensor networks. Ph.D. thesis. National Institute of Technology Rourkela; 2016 16. V. Arulkumar, C. Selvan, V. Vimal Kumar, Big data analytics in healthcare industry. An analysis of healthcare applications in machine learning with big data analytics. IGI Glob. Big Data Anal. Sustain. Comput. 8(3) (2019) 17. V. Arulkumar, C. Puspha Latha, D. Dasig, Jr, Concept of implementing big data in smart city: applications, services, data security in accordance with Internet of Things and AI. Int. J. Recent Technol. Eng. 8(3) 18. V. Arulkumar, M.A. Lakshmi, B.H. Rao, Super resolution and demosaicing based self learning adaptive dictionary image denoising framework, in 2021 5th International Conference on Intelligent Computing and Control Systems (ICICCS) (2021), pp. 1891–1897. https://doi.org/ 10.1109/ICICCS51141.2021.9432182 19. V. Arulkumar, An intelligent face detection by corner detection using special morphological masking system and fast algorithm, in 2021 2nd International Conference on Smart Electronics and Communication (ICOSEC) (2021), pp 1556–1561

IoT Based Automatic Medicine Reminder Ramya Srikanteswara , C. J. Rahul, Guru Sainath, M. H. Jaswanth, and Varun N. Sharma

Abstract Monitoring healthcare is a major issue, that requires attention. In underdeveloped countries, the number of nurses for patients is relatively low, and the accessibility of 24-h medical supervision is also ambiguous, resulting in the incidence of easily avoidable deaths as well as urgent situations, causing a disturbance in the health sector. The medical dispensers that are currently available are expensive, and devices that combine a reminder and a dispenser are hard to come by. The major goal of the medicine reminder is to automatically transmit an alarm and dispense medicine to the correct individual at the stated time from a single machine. An automatic medicine distributor is developed for persons who take medicine without expert direction. It is used by a single patient or by a group of patients. It discharges the individual of the error-prone job of injecting the incorrect medicine at the incorrect time. The main goal line is to retain the device simple usage and affordable. Working software is trustworthy and steady. The older age population will benefit greatly from the device because it can substitute expensive medicinal treatment and the money spent on a personal nurse. Keywords Internet of Things · Medical update frameworks · GSM module · Arduino

1 Introduction Medicines were not [1–6] required in the years past, but now, in our day-to-day lives, most individuals are required to take their prescribed medicine at the specified time since diseases are on the rise. As a result, the majority of people come into contact with these diseases sooner or later. Under these fast or slow-spreading diseases, some diseases are not permanent while most of the others are permanent life terrifying diseases [7]. Human life expectancy is reduced as a result of various disorders. To have a better life they have to take the prescribed medicines at the prescribed time R. Srikanteswara (B) · C. J. Rahul · G. Sainath · M. H. Jaswanth · V. N. Sharma Nitte Meenakshi Institute of Technology, Bengaluru, India e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 I. J. Jacob et al. (eds.), Expert Clouds and Applications, Lecture Notes in Networks and Systems 444, https://doi.org/10.1007/978-981-19-2500-9_11

157

158

R. Srikanteswara et al.

Fig. 1 Medicine reminder pictorial representation

[8]. They are advised by the Doctor to take anticipated medicines in desired doses. Therefore patients have problems like overlooking to take medicine at the correct time. The doctor may change or update the prescription, patients would have to recollect the new list of medicine [9]. People who take prescription medications, whether for a simple virus or something more serious, only take half of the required doses, according to studies [10] (Fig. 1). In addition, one-third of kidney transplant recipients fail to take anti-rejection meds, and 41% of heart attack patients fail to take blood pressure medications. This non-compliance is predicted to result in 125,000 deaths and at least 10% of hospitalizations each year, costing healthcare systems billions of dollars [11]. Medication administration necessitates meticulous attention to detail and concentration. It may be difficult for some people to achieve this on their own, hence a medicine dispensing model would be useful. Many errors occur during the administration of drugs, according to studies [12]. This problem of drug errors can also be caused by medical personnel, putting patients’ health and even lives in jeopardy. Hence we have proposed a project called Arduino-Uno-based Smart medicine box that leverages a real-time clock to decrease difficulties such as taking medicines at the wrong time, taking the wrong dose, and unknowingly taking expired medications, all of which result in unnecessary health complications. In addition to the existing medicine reminder, the newly added feature in our project is that our system is capable of sensing whether the patient has taken the medicine or not, therefore won’t forget to take the tablet at the prescribed time [13] and also this smart pillbox assembles the response from the caretaker or the patient and sends the purchase order to the pharmacy [14]. The user can set the distribution range of the pills and the number of pills at each interval using the application and onboard keys.

IoT Based Automatic Medicine Reminder

159

2 Background The current framework is based on the Android Executing framework, which will prompt the client to take the prescribed medicine on time by displaying notice and ringing an alarm. Android is a Linux-based operating system designed specifically for contact smartphones, such as PDAs, tablets, and PCs, and developed by Google in collaboration with the Open Handset Association. Android was created from the beginning to allow designers to create adaptable applications that take advantage of each of the handset’s features and bring them to the table. The framework is built on the Android working framework because Android is the most popular platform in the business. Android also has an application development framework, which includes administrations for creating Graphical User Interface applications, data to get, and other forms of benefits, as well as an Application Programming Interface for application development. The structure is designed to make segment recycling and merging more efficient. A mandatory XML display record is used in the development of Android apps. When scheduling a time, the show record features are added to the application. This file gives fundamental information to an Android phase for dealing with an application’s existence round. Displays of the application’s parts about other structural and setup properties are examples of this type of detail coupled with a show record. Exercises, Services, Broadcast Receivers, and Content Providers are just a few of the categories that can be found with Segments. After logging in, the patient/user will be able to view a list of all the appointed specialists, including their names, contact information, phone numbers, medical clinic location, availability of specialists as needed, and any other information that the Specialist registers at the time of login to the framework. They can get a drop-down image of the infection and can easily go through the doctors’ review. It also displays the date and time of the next doctor’s appointment. This encourages patients/users to understand the doctor’s diagnosis. The management assists them correctly in following the framework to make it useful and valuable. Prescription updates aid in the reduction of drug administration errors and mismeasurements. The Android interface allows for simple changes in medication intervals and also provides a notification to the caregiver if the medication is not provided on time. Setting an alarm and getting noticed are the two aspects of the update framework. If the patient does not take the pill, the alert sounds louder, and the color of the LED strip changes. After a specified amount of time notifications will be sent to the patient on the Android app to dispense and administer the pill, turn off the alarm [15] With rapid implementations in automotive, home automation, smart cities, and other areas where everything connects to the cloud and makes one’s life easier, the Internet of Things is attracting the interest of a lot of consumers and the enterprise electronics market. Developers have access to a number of power-efficient and low-cost sensors for use in a variety of applications [16, 17]. There are a few existing systems that use IoT concepts and collect data on the type and timing of pill consumption by patients

160

R. Srikanteswara et al.

and store it in cloud storage, where it is analyzed through various applications. In addition, numerous separate pill dispensers are linked together to form a network of pill dispensers that will be monitored via the Internet of Things. The position of each medicine dispenser may be examined in real time via a web interface or an Android interface, which will provide useful information on the treatments received by the patients and what medications they are receiving. Hospitals and other organizations can easily keep track of their patients using these technologies. The network of pill dispensers provides vital data to hospitals, allowing them to deliver the best possible service to their patients [1].

2.1 Cloud Computing and IoT Cloud computing is based on resource sharing, which is essential for IoT platforms. Cloud computing is not only about sharing resources, but also about maximizing them. It is also location independent; customers can access cloud services over an Internet connection from any location and with any device. When we talk about the Internet of Things platform, it should be accessible from anywhere at any time. Another important feature is the virtualization of physical devices; virtualization allows users to effortlessly share gadgets. The multitenancy characteristic of cloud computing allows multiple users to share resources across space and time. Furthermore, Cloud provides elasticity and scalability of resources and applications, as well as easy access and availability of services and resources. As a result, the confluence of Cloud and IoT has the potential to provide enormous opportunities for both technologies [18] (Figs. 2 and 3). Fig. 2 Cloud and IoT

IoT Based Automatic Medicine Reminder

161

Fig. 3 Existing system

3 Proposed System The primary purpose is to assist the patient in taking prescribed drugs and to prevent missed doses due to neglect or poor care. When the pills are not retrieved from the tray, the medicine reminder system illuminates an LCD, sounds an alarm, and sends a notification to an Android application [1]. The patient must take the specified dose from the medicine reminder box on time, otherwise, our system will continue to generate the sound, which will be considerably louder than before, until the patient takes the medicine out of the box. This kind of altering feature increases the life years of the person [2]. This can be of massive contribution to the elderly people and also to the people who are detected with chronic diseases, which requires them to take medicines at regular intervals [2]. The medicine reminder architecture, which consists of a medicine box with a set of different columns for each. Each person can use it regularly for their medicine. The control structure of the medicine box has LEDs for notifying the patient in the form of alarms for their proper medicine taking. There is a ringer in the architecture which notifies the patient by giving an alarm by ringing sound. The alarm will ring for a particular specified time, inside that time just the person needs to press the button by taking the tablet, normally the alarm will be notified in form of messages to track the patient by GSM that patient has not consumed the tablet at the prescribed time by the doctor [19]. The ringer and LED’s are giving the reminder at the particular time set by the family. So the medicine reminder system will regularly analyze the patient’s health by using the IoT (internet of things) and it can also record the patient’s daily dosage level of the tablets. Hence the medicine has its own timing which is compared to the real clock. If the mentioned information matches, the buzzer will go off, or else the buzzer will give the alarm sound, which reminds the patients to take the medicine. Data can be recorded the patient’s health and person daily prescribed medicines to consume [20] (Figs. 4 and 5).

162

Fig. 4 Proposed system

Fig. 5 Proposed block diagram

R. Srikanteswara et al.

IoT Based Automatic Medicine Reminder

163

4 System Reqirements • • • • • • • •

ESP32 CP2102 WiFi BIuet00th DeveI0pment B0ard ICD (IIQUID CRYSTAI DISPIAY) I2C PROTOCOL Arduin0 B0ard GSM(gI0baI framew0rk f0r p0rtabIe) RTC(Real time clock) Switch Buzzer

5 Hardware Requirements • • • • • • • • •

ARDUINO APR SPEAKER LCD’S IR SPEAKER DC MOTOR H-BRIDGE NODE MCU BUZZER

6 Hardware Used 6.1 Arduino Board Arduino board is an open-supply digital platform used for manufacturing singleboard microcontrollers and microcontroller kits for constructing virtual devices. It includes each programmable circuit sheet and a software program or IDE (integrated development environment) that runs on the PC, this board is used to write down and add PC code to the physical board The Arduino platform has come to be well-known with humans truly beginning with electronics, and for modern situations. Unlike the maximum preceding programmable circuit boards, now the Arduino no longer needs any of the separate hardware (referred to as a software program programmer) in order to stack or load new code onto the board—you may truly make use of a USB cable. Furthermore, the Arduino IDE makes use of a simplified model of C++, which makes it smooth to research the code At last, it gives a well-known shape component that breaks out the abilities of the microcontroller into a greater available package [21] . Arduino specification (Fig. 6; Table 1):

164

R. Srikanteswara et al.

Fig 6 Pin specification Table 1 Arduino specifications

Micro controIIer

ATmega 328 P

Operating voIatge Input voItage

7–12 v

Input voItage Iimit

6–20 v

DigitaI I/0 Pins

6

AnaIogue input Pins

6

DC current per I/0 pins

20 mA

DC current for 3.3v Pin

50 mA

FIash memory

Of which 0.5 KB is used

SRAM

2 KB

EEPROM

1 KB

CIock speed

16 MHz

Length

68.6 mm

Width

53.4 nm

Weight

25 g

IoT Based Automatic Medicine Reminder

165

6.2 GSM (Worldwide Framework for Portable) GSM (Global framework for mobile communications)/GPRS (General Packet Radio Administration) is SIM900 MODEM. RS232 Quad-band GSM/GPRS electronics, chips in which frequencies range from 850, 900, 1800, and 1900 MHz. It could be very small in length and less difficult to apply with the GSM modem. The modem includes 3V3 and 5 V DC TTL interfaced with hardware which allows customers to flawlessly interface the hardware with 5 V microcontrollers (PIC, AVR, 8051, Arduino, and so on.) and 3V3 microcontrollers (ARM, ARM Cortex XX, and lots of more). The modem also can interface with a microcontroller with the aid of using the use of USART (Universal Synchronous Asynchronous Receiver and Transmitter) which highlights (Serial correspondence) [22].

6.3 LCD (Liquid Crystal Display) LCD (Liquid Crystal Display) is a flat panel show generation wherein it’s far usually utilized in TV’s and laptop display, cellular devices together with laptops, smartphones. LCDs have a huge variety of display lengths as compared to CRT (cathode ray tube), LCDs had been an essential soar so long as the discovery they supplanted, which comprise light-discharging diode (LED) and plasma shows. Small LCD display shows are generally utilized in LCD projectors and transportable customer gadgets such as small LCD screens are common in LCD projectors and user gadgets together with virtual cameras, watches, virtual clocks, calculators, and, which includes smartphones. LCD display show has been replaced with a big cathode ray tube (CRT) seen in most of the applications [23].

6.4 RTC (Real-Time Clock) RTC’s complete shape is a Real-Time Clock. RTC modules are digital tools that recollect TIME and DATE. That has batteries organized inside, which keeps the module running even with the absence of outside energy and keeps time, date updated. So we will have the best TIME and DATE from the RTC module on every occasion we need. So via the libraries which are used for the module DS3231 is best. Making use of libraries makes the device simple. You ought to download and use those libraries and get in touch with them in the programs. Once the header record is recorded the controller interacts with itself, offers the date and time. Morning timer can additionally be set or modified as a consequence of the use of libraries. Also, while the present day is going down, the RTC module chip withdraws the present day from battery supply related with it naturally. So the time might be modified. Later

166

R. Srikanteswara et al.

while the device restarts the controller can get the prevailing operating time from the module with no mistake [24].

7 Results The medicine reminder system is useful for the user who wants this and all related users. We conclude the result that our project is very useful for those people who are taking medicines regularly, a measure of medicine is very long and hard to remember for those users. Our product also promotes a stress-free life for both the patients and their loved ones which in turn results in a healthier life. Our product is very utilizable in that it can cure those victims’ illness and there will be no need of taking care of these types of patients, so a caretaker has no tension about their health and they will live a healthy and tension-free life [25].

7.1 Unit Testing See Figs. 7, 8, 9 and Tables 2, 3, 4, 5.

Fig. 7 Working system

IoT Based Automatic Medicine Reminder

167

Fig. 8 LCD display

8 Conclusion This medicine reminder aims to prevent an unhealthy and stress-free life for the patients or users who are taking medicines regularly and to dispense this system at a reasonable price and low cost also. Our project is recyclable by changing those other medicinal boxes that have only alerting systems and are non-usable or expensive compared to our product [25]. This pill dispensing system is also very easy to understand, user-friendly, and is sustainable so that the older generations also can use this product without the help of a tech-savvy person. This product has good scalability, is very reliable, and also sends regular notifications about the patient’s well-being through the Android application to their loved ones, hence they can live a worry-free life.

168 Fig. 9 Message displaying through Blynk app

R. Srikanteswara et al.

IoT Based Automatic Medicine Reminder Table 2 Unit test case for ICD

Table 3 Unit test case for RFID

Table 4 Unit test case for GSM

Table 5 Integration test case for RFID card and ICD dispIay

SI # test case: -

169 UTC-1

Name of test: -

ICD testing

Items being tested: -

ICD

SampIe input: -

Power suppIy

Expected output: -

ICD shouId dispIay “WeIcome to piIIbox” message

ActuaI output: -

ICD dispIays “WeIcome to piIIbox”

Remarks: -

Pass

SI # test case: -

UTC-2

Name of test: -

Checking of DC motor

Items being tested: -

DC motor

SampIe input: -

Run the DC motor

Expected output: -

DC Motor shouId run in both directions

ActuaI output: -

First it rotate in forward and next in reverse

Remarks: -

Pass

SI # test case: -

UTC-3

Name of test: -

N0deMCU testing

Items being tested: -

N0deMCU M0duIe

SampIe input: -

Power suppIy, Send the Message

Expected output: -

Message shouId be sent to the BIynk given in the program

ActuaI output: -

Message sent to the BIynk

Remarks: -

Pass

SI # test case: -

ITC-1

Name of test: -

Working of Arduino and ICD

Items being tested: -

DC motor and ICD dispIay

SampIe input: -

PiIIbox

Expected output: -

ICD shouId dispIay morning tabIet

ActuaI output: -

ICD dispIays morning tabIet

Remarks: -

Pass

170

R. Srikanteswara et al.

References 1. M.V. Moise, P.M. Svasta, A.G. Mazare, Programable IoT pills dispenser, in 2020 43rd International Spring Seminar on Electronics Technology (ISSE) (2020), pp. 1–4, https://doi.org/10. 1109/ISSE49702.2020.9121107 2. B. Ayshwarya, R. Velmurugan, Intelligent and safe medication box in health IoT platform for medication monitoring system with timely remainders, in 2021 7th International Conference on Advanced Computing and Communication Systems (ICACCS) (2021), pp. 1828–1831. https:// doi.org/10.1109/ICACCS51430.2021.9442017 3. B. Jimmy, J. Jose, Patient medication remainder: Measures in daily practice. Oman Med. J. (2019) 4. S. Bryant, Principles: Medicine dispensary. IOSR J. Electr. Electron. (1.03)6 (2014). Retrieved Febraury 5. D. Raskovic, T. Martin, E. Jovanov, Medical remainder applications for wearable computing. PC J. 47(4), 498506 (2020) 6. Google, “Activity”. Retrieved Jan (2014) Available: Google, “Fragments”. Retrieved Jan 2019 Available: Google, “Fragment Transaction” 7. H.-W. Kuo, Research and implementation of intelligent medical box, M.S. thesis, Department of Electrical Engineering, I-Shou University, Kaohsiung, TW (2017) 8. N.B. Othman, Ong, P. Ek., Pill dispenser with alarm through smart phone notification using Iot, in IEEE 5th Global Conference on Consumer Electronics (2016) 9. D.H. Mrityunjaya, K.J. Uttarkar, B. Teja, K. Hiremath, Automatic pill dispenser. Int. J. Adv. Res. Comput. (2018) 10. M. Viswanathan, C.E. Golin, C.D. Jones, M. Ashok, S.J. Blalock, R.C. Wines et al., Interventions to improve adherence to self-administered medications for chronic diseases in the United States: a systematic review. Ann. Intern. Med. 157(11), 785–95 (2012) 11. M.D. Lisa Rosenbaum, H. William, M.D. Shrank, Taking our medicine—Improving adherence in the accountability Era. N. Engl. J. Med. 369, 694–695 (2013) 12. “JONA”. J. Nurs. Adm. 39(5), 204–210 (2009) 13. G.B.-Z. Joram (three Dec 2010), Online clever tablet container allotting system, US 20090299522A1 14. S. Shinde, T. Kadaskar, P. Patil, R. Barathe, A smart pill box with remind and consumption using IOT. Int. Res. J. Eng. Technol. 4(12), 152–154 (2017) 15. A. Uttamrao (25 Jan 2015) Intelligent medication container, US 7877268 B2 16. C.G. Rodriguez-Gonzalez, A. Herranz-Alonso, V. Escudero Vilaplana, M.A. Ais-Larisgoitia, I. Iglesias-Peinado, M. Sanjurjo-Saez, Robotic dispensing improves patient safety, inventory management, and staff satisfaction in an outpatient hospital pharmacy. J. Eval. Clin. Pract. 25(1), 28–35 (2019) 17. F. Mattern, C. Floerkemeier, From the internet of computers to the internet of things. ETH Zurich, pp 1–6, Oct 2016 18. M.V. Moise, A.-M. Niculescu, A. Dumitra¸scu, Integration of internet of things technology into a pill dispenser, in 2020 IEEE 26th International Symposium for Design and Technology in Electronic Packaging (SIITME) (2020), pp. 270–273. https://doi.org/10.1109/SIITME50350. 2020.9292283 19. A.R. Biswas, R. Giaffreda, IoT and cloud convergence: Opportunities and challenges, in 2014 IEEE World Forum on Internet of Things (WF-IoT) (2014), pp. 375–376. https://doi.org/10. 1109/WF-IoT.2014.6803194 20. E. Peel, M. Douglas, J. Lawton, Self tracking of blood glucose in kind 2 diabetes: longitudinal qualitative have a look at of sufferers perspectives. BMJ 335(7618), 493 (2007) 21. K. Bhavya, B. Ragini, Assistant Professors Department of Engineering Karpaga Vinayaga College of Engineering and Technology, Padalam Tamilnadu, India” A Smart Medicine Box for Medication Management the usage of IOT 22. J.P. Solanke, S.K. Lakshman, Smart medicine and monitoring system for secure health the usage of IOT

IoT Based Automatic Medicine Reminder

171

23. M.V. Kondawar, A. Manusmare, Review on smart pill box monitored through internet with remind, secure and temperature controlled system 24. B. Jimmy, J. Jose, Patient medication adherence: Measures in daily practice. Oman Clin. J. (2016) 25. S. Bryant, OO principles: Encapsulation and decoupling. IOSR J. Electr. Electron. 1.03(6) (2014), Retrieved January

IoT Based Framework for the Detection of Face Mask and Temperature, to Restrict the Spread of Covids Ramya Srikanteswara , Anantashayana S. Hegde, K. Abhishek, R. Dilip Sai, and M. V. Gnanadeep

Abstract Coronavirus is a novel virus that is responsible for causing the disease, COVID-19, which is deadly. It was first detected in December 2019, in the city of Wuhan, China and due to its contagious nature, people all over the wor1d are now infected. COVID spreads very fast and easily through air, from one person to another, affecting almost the entire population of the world in a short span. Wearing a mask in public is very much necessary as a means of preventive measure against the viral disease. Moreover, body temperature is an important factor to be identified to determine whether an individual is affected from the virus. Manually checking if a person wears a mask in outdoors or determining the temperature of an individual in a crowded area, is a tedious task and requires an urgent need for solution. Internet technology introduction into the world is beneficial and it can transmit the data without any human interaction which is best suited for this Covid-19 situation. This article provides the road map to how this technology can be utilized for a better cause. In this work, an IoT based framework is designed to ensure the restriction of entry of a Covid affected individual into the premise, by detecting if the mask is worn and his temperature is normal, to avoid the spread of this disease. Moreover, using this method the safety of the staff in the checking process at the entry point is protected. Keywords Corona-virus · COVID-19 · Deep-learning · Face-mask-detection · Temperature detection · Tensor-Flow · Automatic functioning of gate

1 Introduction COVID is a contagious respiratory disease caused by Severe Acute Respiratory Syndrome CoronaVirus (SARS CoV). Currently COVID is spreading in all countries across the world quickly, affecting over 181 million peoples [1], and 3.9 million deaths are recorded based on the report from WHO (World Health Organization) on 28 June 2021. In order to avoid the worldwide disaster, a wise and clear-cut method R. Srikanteswara (B) · A. S. Hegde · K. Abhishek · R. D. Sai · M. V. Gnanadeep Nitte Meenakshi Institute of Technology, Bengaluru, India e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 I. J. Jacob et al. (eds.), Expert Clouds and Applications, Lecture Notes in Networks and Systems 444, https://doi.org/10.1007/978-981-19-2500-9_12

173

174

R. Srikanteswara et al.

to prevent the spreading of the COVID is instantly desired globally. Few familiar individual coronaviruses [2] that affect the earth are LN63, 1HKU, 430C and E229. Earlier similar viruses such as, COV- SARS, nCOV-2019 and COV-MERS, enervated people and affected wildlife, and now has changed toward individual coronaviruses [3]. People who have respirational difficulties can affect anyone near them, by their infectious droplets. Surrounds of polluted people will thus cause the spread, and the virus might never reach the close sides. To curb the metastasis [4], wearing a face mask while going to any public place is very much necessary. WHO is stressing on their points of wearing a face mask as their priority suggested by the health care assistants. The reason of using Internet of Things (IoT) for this model is important because IoT can transform the data without any human intervention. It is an intelligent system which can collect the data through cameras and sensors and analyze the data and send it to anywhere in the world. IoT can merge with other technology like deep learning, machine learning and AI to combat against COVID-19. Face mask image detection has become a significant job in international civilization. Mask recognition includes the processes of finding and locating the face in the image and the crucial process of identifying whether the identified face is covered using a mask or not. The problem is the object discovery to identify the objects. Face detection process deals with various characteristics to identify the face, like identification of colors, locating the points, joining the points to identify the objects, and also identifying different objects even if the objects are varied in sizes and so on [5]. Accuracy is more essential in Deep Learning, and tensor-flow is used to calculate their data. For face detection, a CNN model which contains a dataset of more than 60 k images of faces with or without mask is utilized. During training, the image are segregated as with mask and without mask. Next, a testing model is created which checks the given image. The image processing is done by extracting images bit by bit. The proposed method provides a simplified approach toward face mask detection by using the Deep Learning packages like TensorFlow, Keras and OpenCV. Main objectives of this proposed work are as follows: • Allowing a person to enter the premises only if he has a normal body temperature, and ensure he is wearing a mask; and automatically open the gate only if these conditions are satisfied. • Ensuring safety of the security person in charge, as a security person near the gate may not be required.

2 Related Background Covid has become one of the most life-threatening diseases. There are many techniques used to evade the virus. In one of the technique is detection of face. A face is recognized from a picture that has several different attributes. Consistent with [6], analysis into the detection of face needs an identification of expressions, following

IoT Based Framework for the Detection of Face Mask …

175

the face, and estimation-creation. Identifying the face in an image, is the major challenge, as the face modifications are different in color, size and shape and they aren’t absolute. It becomes a more difficult job in case of unclear pictures and gets delayed by another issue of concern, due to low quality picture capturing camera. Authors of [7] says, closure detection of face technique has 2 major challenges: (1) datasets of sizable volumes that contain pictures of masked and unmasked faces are not available. (2) facial expressions are excluded in the area bounded. Using the regionally linear embedding rule and the trained dictionaries on associate degree vastly, masked faces, synthesize of dull faces, many misplaced facial expressions are often regained, additionally the superiority of facial cues are often lessened to fine degree. Consistent with the work rumored in [8], there is a strict constraint that comes along with the Convolutional Neural Network (CNN) in laptop vision relating to the scale of the input image. Before fitting the pictures into the network, the pictures are reconfigured in order to overcome the inhibition. Generally, identifying the face from the picture with accuracy and identifying whether the person has a mask worn or not is a major challenge, however the proposed method identifies a face wearing a mask in motion. Although various methods exist, there is a need for a cost effective and efficient method for the prevention of spread of Covid. The proposed method considers both these important factors.

3 Proposed Work Dataset In the proposed method, two datasets are used for testing the technique. Dataset 1 [9] contains 1918 pictures of people wearing face masks and 1915 pictures of people not wearing face masks. The captured images are single face pictures in various surroundings, and with varieties of masks being worn by people in different colors as shown in Fig. 1. Dataset 2 [10] contains 1915 pictures of people without mask. In Fig. 2, the collections of faces are slightly tilted with multiple expressions on the faces. The main challenges faced while identifying people with or without masks are that, this method comprises of images in varying angles and lack of clarity. Indistinct moving faces in the video make it more difficult. The system can efficiently detect partially covered faces either with a mask, hair or hand. It considers the occlusion degree of four regions—nose, mouth, chin and eye to differentiate between annotated mask or face covered by hand. Therefore, people can use this loophole to get away from the sensors. Packages Incorporated a.

TENSORFLOW TensorFlow acts as an interface used for implementing the algorithms of Machine Learning (ML). It is also used in the implementation of the ML systems

176

R. Srikanteswara et al.

Fig. 1 Few pictures from dataset 1 with faces wearing masks

b.

with several fields of computer science, which contains study of sentiments, voice recognition, and taking out of physical information, computer vision, summarization of texts, recovery of information, computational discovery of drugs and also detection of flaws in order to follow research [11]. In the proposed model, the entire Convolutional Neural Network (CNN) architecture, which is sequential, uses TensorFlow as a backend. Also, it is used to change the size of the image, that is scaling them during the processing of the images. KERAS Keras provides the basic vital structure with easy assembly unit for the development and passage of ML measures. Scalability and cross platform capacities of TensorFlow are considered as an advantage. The layers and models are the core data buildings of Keras [12]. Using Keras, the layers in the CNN model are applied. The overall compilation of the model is done by the process of

IoT Based Framework for the Detection of Face Mask …

177

Fig. 2 Few pictures from dataset 2 with faces not wearing masks

c.

conversion of the vector classes into a matrix of binary format during image processing. OPENCV Open Source Computer Vision Library shortly known as openCV, is an open source computer vision and ML software library, which is employed to compare and differentiate, to identify objects and faces, to group motions in recordings, follow progressive modules, trace eye motion, path camera actions, eject red eyes from photos using flash notice relative pictures in an image database, understand landscape and discovered markers in order to overlay with exaggerated reality [13], in resizing and conversion of color in data images. OpenCV is used in the projected technique.

178

R. Srikanteswara et al.

Proposed Method In the proposed method, 2D convolution layers are linked to a dense-neuron layer, which is a cascade classifier and a pre-trained CNN. The process of imagepreprocessing involves data conversion from a given format to another format that is user friendly. The subject could be in any desired format which is more like tables, images, videos, graphs and so on. The data which is ordered, fits in a model called info-model, arranges and also records relationships among different objects [14]. This proposed method uses Numpy and openCV for image and video data arrangements. Conversion of image from RGB to Grayscale: Image recognition systems based on modern descriptors usually uses grayscale images, hence color image is converted to grayscale. While using robust descriptors, process of conversion from Color to Grayscale is a little complicated. In order to achieve a good routine, present nonessential data may increase the training dataset size. With the help of grayscale, the algorithm is rationalized and the computational requisites are diminished. Descriptors can be utilized, instead of working parallelly on colored images. [15]. The quality of image recognition can be implemented by constructing a novel framework with deep learning and Principal Component Analysis (PCA) to build an IoT image identification. This proposed research work has conducted many tests and delivered greatest image-based recognition results [16]. The face mask detection algorithm is as follows:

Deep Learning Deep-learning can be defined as a part of Machine Learning (ML) algorithms which uses several layers to extract features of high-level, progressively from raw-input. As

IoT Based Framework for the Detection of Face Mask …

179

an example of image-processing, the layers considered in the lower area may identify edges, while layers present in the higher area may identify the concept related to the topics in the life of human beings like faces, digits or letters. Deep learning an element of ML processes are reinforced by the use of Artificial Neural Networks (ANN) along with the representation learning. Learning may be categorized as supervised learning, semi-supervised learning or unsupervised learning [17]. Deep-learning architectures that comprises of Deep Neural Networks, Belief Networks, Persistent Neural Networks and Convolutional Neural Networks are applied to several fields like computer and machine vision, NLP, machine-translations, speech recognition, audio recognition, societal network riddling, bioinformatics, drug-designs, study of medical-images, review of material, and also programs of board-games. Wherever it is, it yields similar results surpassing the performance of human experts in some cases. Processing of information concepts and communication nodes in distributed systems, in the biological format in biological systems are the main inspiration for ANN [18]. However, neural networks are different from biological brains in many aspects. Specifically, the brains are dynamic and analogous in living organisms, whereas neural networks are symbolic and static. In case of deep learning, the word “deep” is used, as it utilizes several layers in a network. Previously, a universal classifier does not include linear perceptions. Therefore, it can be said that deep learning is a variation that deals with infinite layers of fixed size and accepting it for practical applications also for enhanced implementation, yet preserving back the theoretical popularity under mild situations. The layers in deep learning are allowed to be different from each other and also from biologically formed models. This model is for trainability, understandability and for efficiency; hence it can be mentioned as the “structured” part. Deep Neural Networks DNN has various types of neural networks that contain neurons, biases, functions, weights and synapses. The functioning of these parts is comparatively the same as that of the working of brains of humans, and it can be trained like other Machine Learning algorithms [19]. Considering an example: DNN which are trained in order to identify the animal, which is a cat breed, is studied and analyzed from the given image, also calculates the probability that the cat is of a particular breed. Users can examine generated outcomes and select the precise possibilities that the network must exhibit the above threshold values etc. Every measured manipulation is considered as a layer. A complex Deep Neural Network contains many layers; hence called “deep” networks. DNN are capable of modeling the complex relationships that are non-linear in nature. These neural network architectures create a compositional model in which the object is represented as compositional layers of primitives. Composition of features from lower layers is done using additional layers that effectively models data that are complex in nature with less units of measurement than performing as a shallow network. Deep network architectures contain several varieties of complex kinds of approaches. Each architecture reaches success in each of its respective domains.

180

R. Srikanteswara et al.

Comparing the performance of multiple architectures always is impossible, unless same data sets are used for evaluation. DNNs are networks where information flow occurs from input to the output layer that is in feed forward format, without looping back. Firstly, DNN creates a complete design of connections of all virtual neurons, and then assigns them with a random number chosen [20], or to any “weights”, in order to make the connections. The product of weights and inputs are found, and then an output value in the range 0 and 1 is returned. If a certain pattern is not found in the network, then the weights are adjusted using an algorithm. In this manner, few parameters are made more influential by using an algorithm. This process is done till it determines a correct mathematical method to process data completely. Recurrent Neural Networks (RNNs) are neural networks within which data flow takes place in any way and are also utilized for applications like language modeling. For a long period of time, the short-term memory is more effective. Convolutional Neural Networks (CNNs) are neural networks that can be used in computer-vision. It is used for Acoustic Modeling for speech recognition automatically. Convolution Neural Network Convolutional Neural Network shortly known as CNN, is also called as ConvNet which is a class of DNN. It is mostly applied to image processing. Most CNNs are only equivariant i.e., if a variation occurs, it occurs in equal proportions, as opposed to invariant that is no variance to translation [10]. Its applications include video and image recognition, image segmentation, natural language processing, image classification and financial time series. CNNs are perceptron constituted of several layers which are arranged grounded on some rules and regulations. The networks are completely connected, i.e., each neuron in layer 1 has a connection to any or all neurons of the upcoming layer. Entire connectivity in these fully connected networks [11] make them in peril of overfitting data (Fig. 3). Convolution Neural Networks takes a different approach toward regularization. It is done by considering the hierarchical patterns that are found in data and by assembling patterns [21] in increasing order of their complexity by using much simpler and smaller patterns that are embossed in their filters. Hence considering the connectivity and complexity measures, Convolution Neural Networks are on the lower extreme.

Fig. 3 Steps followed by convolutional neural network

IoT Based Framework for the Detection of Face Mask …

181

The idea of these networks is inspired from the nerves and neurons in living organism that acts as message carriers connecting all body parts to the brain. The connectivity pattern in that biological process, between neurons is similar to that of the organization [12] of the animal Visual Cortex. Individual Cortical neurons works only for restricted region of visual fields to stimuli known as the receptive field. Different neurons containing these receptive fields individually overlap completely so that they can cover the entire visual fields. CNN use less preprocessing compared to the other image classification algorithms. Hence networks can learn to enhance filters over automatic learning. In contrast, in case of traditional algorithms these filters are applied manually. Hence this feature is a major advantage in feature extraction technique, as there is independence from prior knowledge and from human interventions. Histogram of Oriented Gradients (HOG) HOG provides the feature descriptor technique used at deep learning for object detection in Computer Vision and Image Processing. In this procedure, the sum of the incidences of gradient orientation with local portions of an image is considered [15]. These concepts are relatable to shape contexts edge orientation histograms. The main difference is that it is computed on a compact network of consistently spaced cells, and for improved accuracy it used for overlapping local contrast normalization. In 1986, Robert K. Mc Connell explained the concepts behind HOG without using the word HOG in an application. In 1994, Mitsubishi Electric Research Laboratories used these concepts. Navneet Dalal [22] and Bill Trigg, scientists at French National Institute for Research in technology and Automation, at the Conference on CVPR (Computer Vision and Pattern Recognition) in 2005, presented their work on HOG descriptors, where these concepts became widespread because of their utilization. The main focus of Navneet Dalal and Bill Triggs, was on the detection of pedestrians in static images, as they wanted to include detection of humans in the videos. Moreover several varieties of vehicles and animals detection in static images, through their advances in their tests were performed (Fig. 4). Flow of the Proposed Work The framework works as a 2-factor authentication, to let a person enter the locale. As the first step, the face mask detection is performed, where a web cam captures the image of the person who wants to enter. The captured image is forwarded to the face mask detection algorithm, where the trained data set is used to compare the face of the person and determine whether the person is wearing a mask or not. Based on the result, if the person wears a mask, the face is bounded by a green colored rectangular box with accuracy and if the person does not wear a mask, then the face is bounded by a red colored rectangular box with accuracy, and also leads to a security alert sound, which indicates there is a person near the gate without a mask (Fig. 5). The second step of authentication is the temperature detection, where a hardware setup comprising of a microcontroller, temperature sensor, servo motor and LCD display are used. The temperature sensor senses the temperature of the person, which is forwarded to the LCD display by the microcontroller and is further used to

182

R. Srikanteswara et al.

Fig. 4 Images before and after application of HOG algorithm Flow of the proposed work Start

Capture image using web cam

Apply face mask detector

yes

Temperature detection

yes Gate opens

Check if temperature is normal

Fig. 5 Flow of the proposed work

Check if mask worn

no Gate closes

no

Sound alert

IoT Based Framework for the Detection of Face Mask …

183

determine if the servo motor has to open the gate or not. If the temperature detected by the temperature sensor is high, then the gate will not be open and the person is not allowed inside. If the temperature is normal, then the servo motor opens.

4 Results The proposed method performs two main functions to avoid the spread of Covid. It does a check if the person entering the organization has a mask on or not and checks the temperature as well. This can be checked by the admin on his system. Only when both the criteria are found to be fulfilled, the gate opens. Figures 6 and 7 are snapshots of mask detector detecting persons with and without mask. Fig. 6 Mask detector detecting a person wearing mask

Fig. 7 Mask detector detecting a person without mask

184

R. Srikanteswara et al.

Fig. 8 Temperature detected displayed in Fahrenheit using LCD display

The mask detection using keras and OpenCV method attain accuracy up to 95.77% and 94.58% respectively on two different datasets, where dataset 2 is more versatile than dataset 1. The optimized values of parameters are obtained using the Sequential Convolutional Neural Network model to detect the presence of masks correctly without causing over-fitting. Figure 8 indicates the temperature detected displayed in Fahrenheit using LCD display.

5 Conclusion The proposed technique mainly provides safety against the current deadly disease COVID-19, which is a contagious disease spreading through the air we breathe in and affecting the lungs, reflecting primary symptoms of cough, cold, fever, breathing problems and many more, ultimately leading to death. Even though the government has implemented lockdown, yet due to unavoidable reasons people insist to come out. As preliminary precautions, it is very much essential to keep social distancing, wash the hands regularly, circumvent touching nose and mouth, wear a mask before going to any public places, check the body temperature frequently and consult a doctor immediately if there are any symptoms. It is the responsibility of the institution/locale authority to ensure the people are healthy and are following these precautionary measures. Also, the safety of staff members involved in this process is also very much important. An Internet of Things (IoT) based biotelemetry can monitor the patients by analyzing the ECG signals. Information is classified using artificial neural module to obtain clinical decisions [23]. This method aims at allowing a person to enter the premise with normal temperature, and with a mask on. Allow or restrict of the entry of the person is controlled by automatic open/close gate based on the result of these factors. Since these precautionary measures are useful to restrict the affected people enter the premise, this protects the safety of healthy person and avoid the spreading of the disease. Moreover, it ensures the safety of all the people belonging to the institution including the

IoT Based Framework for the Detection of Face Mask …

185

staff authorities and members involved in the security process. This project has a vital role in reducing the contact of affected and unaffected person, which acts as a major step to fight against COVID-19. The basic Convolutional Neural Network (CNN) model using TensorFlow with Keras library, detects if the person has his mask on or off the face. The temperature is detected by the sensors and the gate automatically functions based on these two criteria. It ensures that no outsider enters the institutes/schools/companies/public places without a mask by creating an alarm. The model can be upgraded to mail the authority when it finds a person off-mask from his face.

References 1. WHO, Coronavirus disease 2019 covid-19. https://www.who.int/docs/default-source/corona viruse/situation-reports/20200812-covid-19-sitrep-205.pdf?sfvrsn=627c9aa8_2 (2020) 2. “Coronavirus disease 2019 (COVID-19)—Symptoms”, Centers for disease control and prevention, https://www.cdc.gov/coronavirus/2019-ncov/symptomstesting/symptoms.html. (2020) 3. Corna virus—Human coronavirus types—CDC, https://www.cdc.gov/coronavirus/types.html (2020) 4. WHO, Advice on the use of masks in the context of COVID-19: interim guidance (2020) 5. M. Jiang, X. Fan, H. Yan, RetinaMask: a face mask detector, https://arxiv.org/abs/2005.03950 (2020) 6. N. Dalal, B. Triggs, Histograms of oriented gradients for human detection, in 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR’05), vol. 1. (IEEE, 2005) 7. S. Ge, J. Li, Q. Ye, Z. Luo, in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2017, pp. 2682–2690 8. S. Ghosh, N. Das, M. Nasipuri, Reshaping inputs for convolutional neural network: Some common and uncommon methods. Pattern Recogn. 93, 79–94 (2019) 9. prajnasb/observatins, GitHub, https://github.com/prajnasb/observations/tree/master/experi ements/data (2020) 10. S. Bharathi et al., An automatic real-time face mask detection using CNN, in 2021 Smart Technologies, Communication and Robotics (STCR). (IEEE, 2021) 11. A. Das, M.W. Ansari, R. Basak, Covid-19 face mask detection using TensorFlow, Keras and OpenCV, in 2020 IEEE 17th India Council International Conference (INDICON) (IEEE, 2020). 12. X. Deng et al., A classification–detection approach of COVID-19 based on chest X-ray and CT by using keras pre-trained deep learning models. Comput. Model. Eng. Sci. 125(2), 579–596 (2020) 13. A. Dumala, A. Papasani, S. Vikkurty, COVID-19 face mask live detection using OpenCV, in Smart Computing Techniques and Applications (Springer, Singapore, 2021), pp. 347–352 14. B. Suvarnamukhi, M. Seshashayee, Big data concepts and techniques in data processing. Int. J. Comput. Sci. Eng. 6(10):712–714 (2018). https://doi.org/10.26438/ijcse/v6i10.712714 15. C. Kanan, G. Cottrell, Color-to-Grayscale https://www.researchgate.net/publication/221 755665 (2012) 16. I.J. Jacob, P.E. Darney, Design of deep learning algorithm for IoT application by image based recognition. J. ISMAC 3(03), 276–290 (2021) 17. R. Memisevic, Deep learning: Architectures, algorithms, applications, in Conference: 2015 IEEE Hot Chips 27 Symposium (HCS), Aug 2015. https://doi.org/10.1109/HOTCHIPS.2015. 7477319

186

R. Srikanteswara et al.

18. F. Hohman, M. Kahng, R. Pienta, D.H. Chau, Visual analytics in deep learning: An interrogative survey for the next frontiers, in IEEE Transactions on Visualization and Computer Graphics, vol. 25(8), pp. 2674–2693, 1 Aug 2019. https://doi.org/10.1109/TVCG.2018.2843369 19. H. Yi, S. Shiyu, D. Xiusheng, C. Zhigang, A study on deep neural networks framework, in 2016 IEEE Advanced Information Management, Communicates, Electronic and Automation Control Conference (IMCEC), 2016, pp. 1519–1522. https://doi.org/10.1109/IMCEC.2016.7867471 20. P. Sharma, A. Singh, Era of deep neural networks: A review, in 2017 8th International Conference on Computing, Communication and Networking Technologies (ICCCNT), 2017, pp. 1–5. https://doi.org/10.1109/ICCCNT.2017.8203938 21. T.S. Rao, S.A. Devi, P. Dileep, M.S. Ram, A Novel approach to detect face mask to control covid using deep learnig. https://ejmcm.com/article_2807_b1e004fc8cf0f8080144eb4707a 0b85a.pdf 22. R. Yamashita, M. Nishio, R. Do, K. Togashi, Convolutional neural networks: an overview and application in radiology. https://insightsimaging.springeropen.com/articles/10.1007/s13 244-018-0639-9 (2018) 23. V. Balasubramaniam, IoT based biotelemetry for smart health care monitoring system. J. Inf. Technol. Digit. World 2(3), 183–190 (2020)

Efficient Data Partitioning and Retrieval Using Modified ReDDE Technique Praveen M. Dhulavvagol, S. G. Totad, Nandan Bhandage, and Pradyumna Bilagi

Abstract I.T. industries and private organizations generate a massive volume of data every day. Storing and processing Big data is challenging due to scalability and performance issues. Nowadays, a distributed architecture is used to process Big data. In a distributed architecture, several nodes/systems communicate to store and process data in a distributed architecture. Search engines use distributed architecture to store and retrieve documents for the user query. Elasticsearch is an open-source search engine, which uses distributed architecture. The main goal of this paper is to configure elastic search clusters, implement the shard selection algorithms, and perform the comparative study analysis of the existing shard selection techniques with the proposed shard selection technique. The sharding technique is applied to partition and retrieve relevant data from the nodes. Shards are created on each data node of the cluster. Shard is the small unit of storage in the memory of the data node. Data is horizontally partitioned according to topic-based and stored on different shards. This paper proposes a Modified ReDDE shard selection algorithm that enhances the throughput by searching only the relevant shards in the distributed processing architecture instead of all the shards. The results interpret that the Modified ReDDE algorithm improves the performance parameters compared to existing shard selection techniques by 26%. Keywords Shard · Cluster · Node · Index · Elasticsearch · ReDDE

1 Introduction In today’s digital era, search engines are the medium for connecting to the digital world. The search engines serve the data which the user wants in some milliseconds. The data retrieved will be in terms of the millions that the data collection data centers P. M. Dhulavvagol (B) · S. G. Totad · N. Bhandage · P. Bilagi School of Computer Science and Engineering, KLE Technological University, Hubballi, India e-mail: [email protected] S. G. Totad e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 I. J. Jacob et al. (eds.), Expert Clouds and Applications, Lecture Notes in Networks and Systems 444, https://doi.org/10.1007/978-981-19-2500-9_13

187

188

P. M. Dhulavvagol et al.

have nowadays. The most searched result is the browser, the most relevant data for the search query. In general, the data is fetched from the storage by comparing the search query with the data in storage. This is called a traditional search technique that takes high throughput, high latency, and low fault tolerance, due to which the efficiency of a search engine is reduced. Now the data will be stored in distributed architecture where the huge server is located at different geographical places across the world. These servers communicate with each other and fetch the relevant data for the users. A load of searching for data in a single server has been reduced and distributed across multiple servers. This distributed architecture motivates the Elasticsearch search engine. Elasticsearch search engine follows the distributed architecture (master node- data node). One single server acts as the Master node, and all the other connected servers are called data nodes. Inturn, these data nodes are divided into shards, which means each is further divided into shards. Shards are the small storages where the actual data is stored and each shard belongs to different topics. When the user gives a search query the query is first passed to the master node, where it is connected to all nodes. This master node contains the information of each data node. That is the number of shards, replicas, and which topic is stored in which shard. The master node will search in those shards that belong to the query’s topic. The relevant files of the search query are fetched from shards and given to the user. This distributed architecture is called as Elasticsearch cluster.

2 Literature Survey Storing and retrieving a large dataset is challenging in Distributed architecture; therefore, the researchers have proposed some techniques to overcome the challenges [1]. Elasticsearch can be the better search engine for medium and large-scale data. CORI shard selection algorithm can be used as the shard selection and reported that the distributed architecture communicates in the search operation for a particular query. Here data nodes are only responsible few parts of the dataset in searching and storing. This algorithm will fetch the data from shards relevant to the search query. The parameters enhanced are latency, throughput and scalability. The CORI algorithm is the first lexicon-based algorithm and uses many assumptions and hard-coded values [2]. Dividing the data according to the topic-based motivates the cluster to search the data for the query in a few shards which contain the relevant data. If the documents have similarities, then they can be grouped. Kulkarni and Callan [2] described that the selective search can be the approach for dividing the data according to topic-based shards and searching in only a few shards that are relevant to the query. The experimental results showed that the selective search is more effective than the exhaustive method (traditional search method). The machine learning algorithm tries to divide the data according to their similarity and store them. While searching the machine learning algorithm fetches the data by similarity of the query and the data. Machine learning’s ability to learn and

Efficient Data Partitioning and Retrieval Using Modified ReDDE …

189

make predictions can take insights from a large dataset. L’Heureux [3] reported that Big data is the future. Humans produce 2.5 quintillion bytes every day, and storing, processing, fetching and maintaining efficiency is difficult. Machine learning can be the solution for this because of the ability of machine learning to analyze the data and store them according to the predictions made by the machine learning algorithms. Praveen et al. [4] explained that in distributed network the systems communicates with each other for information retrieval. The sharding technique can be applied for information retrieval where the scalability, consistency, and fault tolerance issues can be addressed. The paper’s main aim is to enhance processing and information retrieval in large-scale databases. The authors propose the ReDDE shard selection algorithm, focusing on the database collection and content similarity for information retrieval. The comparative analysis of the proposed algorithm (ReDDE) and existing shard selection algorithm (CORI) shows that the ReDDE shard selection algorithm is more efficient than CORI and reduces the overall cost by 28%. Federated search is technique of searching on a multiple text collection simultaneously. A query is submitted to the sub-collections, estimated to contain relevant data. Finally, the results from every collection is merged and listed. Further [5] identified the three major challenges of federated search. Those are collection selection, collection representation, and result merging. The authors explained the representation sets in Cooperative Environment, Lexicon Based Collection Selection and Federated Search Merging to overcome the challenges. Prior research under various conditions on CORI algorithm has proved to be an effective collection selection algorithm. Callan et al. [6] has reported that the CORI algorithm does not perform well in small and extensive databases mixed environments. Proposes a new collection selection algorithm, which is based on database size and database contents. The experimental results showed that the database size estimation is a more accurate method for a large database. The ReDDe algorithm showed better performance in collection ranking than the CORI algorithm [7]. The existing shard selection techniques CORI, exhaustive methods, and Rank-S have failed to optimize and efficiently retrieve the relevant data to the user query. We propose a Modified ReDDE shard selection technique to overcome these challenges, enhancing the performance parameters throughput, scalability, and latency.

3 Methodology The methodology for implementing the Modified ReDDE shard selection and ranking algorithm in a distributed architecture follows. The distributed architecture contains several nodes in which one node is considered a master node, and the remaining nodes are data nodes. Initially, the user query is passed to the master node. The master node contains CSI (Centralized Sample Index), where the query is matched. As a result of comparison, the master node creates a list of documents which are estimated to contain the relevant document to

190

P. M. Dhulavvagol et al.

Fig. 1 Proposed system architecture

the user query. The document with highest similarity score and shard which contains this document is considered as more relevant document and shard to the user. The unstructured data is fed into the master node where it first pre-processes the data and performs clustering operation on the pre-processed data. The steps followed to process the data are: • Pre-processed data is used for clustering the data according to topic-based. • The clustered data is loaded into the data nodes by the master node. • The user gives a query to the master node, where it analyzes to which cluster the query belongs. • A master node sends the query to that shard that contains the relevant data. • The data which is relevant to the query is fetched and displayed to the user The Fig. 1 shows the proposed system diagram. It mainly includes 6 modules.

3.1 Data Node The data node is a storage entity in cluster where the data is indexed and stored. The Fig. 2 shows the data node which mainly consists of two types of shards. • Primary shard—This shard contains the actual data. When we index documents then the documents will be stored in the primary shards. • Replica shard—This shard is the copy of the primary shards, which is present in the other data nodes

Efficient Data Partitioning and Retrieval Using Modified ReDDE …

191

Fig. 2 Data node in a distributed architecture

Fig. 3 Shard

3.2 Shard The shard is the unit at which Elasticsearch distributes data around the cluster. Shard contains the lucene index and segments (Fig. 3). Each elasticsearch index contains the shards as numbers specified during the indexing of the documents in elasticsearch. The shards contain Lucene index and segments. The Lucene index stores statistics about terms to make term-based search more efficient. Lucene’s index falls into the family of indexes known as an inverted index. This is because it can list, for a term, the documents that contain it. The segments store the actual text files. The Lucene index and segments are interconnected, when the query is matched by the Lucene index, it fetches the text files from the segments. ReDDE is the new resource selection algorithm which tries to estimate the relevant documents from large dataset. The algorithm tries to find the distribution of relevant data in dataset. The ReDDE algorithm considers the content similarity and size of the database for estimation of the relevant documents.

4 Implementation 4.1 Pre-Processing of Data Pre-processing consists of tokenizing, stemming, removal of stop words. Bringing down the data in such a format helps in getting results accurately.

192

P. M. Dhulavvagol et al.

Algorithm 1—Pre-processing of data Input- The data should be in text format and unstructured data. Output - Whole data is pre-processed and stored for future use def removewords (listofTokens,listof Words): return (token for token in list Tokens if token not in ListOfwords def applyStemming(listofTokens, stemmer): return (stemmer.stem(token) for token in listofToken) def twoletters(listofTokens) twoLetterWord=[ ] for token in Listoftokens: if len(token) ==2 or len(token)Token == 21 twoletterWord.append(token) return twoLetterword

4.2 Clustering of Data This involves dividing the data according to the topic-based with help of the kmeans clustering algorithm. Each cluster of data is loaded into each shard in data nodes through the master node. Algorithm 2—Clustering of data Input - All the text files should be preprocessed. Output - Clustering algorithm with term frequency and inverted document frequency(tf-Idf) clusters text files. def run_Means (max_k, data): max k + 1 kmeans results = dict() for k in range (2, max_k): kmeans = cluster. KMeans(n_clusters = k, init = ’k-means++’, n_init = 10, tol = 0.0001, in jobs = 1, random state = 1, algorithm = ‘full’) kmeans_results.update({k:kmeans.fit(data)}) return kmeans_results

4.3 Data Loading In this module, the data is loaded into the shards according to topics. Each shard contains a unique topic. Replica shards are loaded with data from other data nodes.

Efficient Data Partitioning and Retrieval Using Modified ReDDE …

193

Input—The data should be clustered according to topic-based. Specifying the number of shards and converting to JSON format. Output—Data is loaded into shards according to the topic-based.

4.4 Input Query The user enters the query in a search box and when the user presses the search button then relevant documents to the query are fetched. Input—Data should be present in data nodes according to topic-based. User gives the query. Output—Query is sent for retrieving relevant data.

4.5 Shard Selection Selecting the shards which contain the relevant document for the query can be implemented with the help of the CORI shard selection algorithm. Algorithm 3 Shard selection algorithm Input - Data should be present in data nodes according to topic-based and the user should give the query. Starts comparing the query with the ready created clusters. Output - Gives the shard number on which the relevant data is stored. def getroutingkeys (substr): routings=set ([ ]) for i in rangelo, len (files)): if filesl[i] in range0-7: routings routings union(set{Routing keys(u])) elif files[i] in range7-14: routings routings.union (set{Routing keys(u])) elif files[i] in range14-21: routings routings.union(set{Routing keys(u])) elif files[i] in range21-28: routings routings.union(set{Routing keys(u])) elif files[i] in range28-35: routings routings.union(set{Routing keys(u]))

194

P. M. Dhulavvagol et al.

4.6 Data Retrieval The data which is relevant to the query are fetched from the shards which are selected by the Modified ReDDE shard selection algorithm. The relevant documents are fetched based on the similarity score of the document to the user query.

5 Results and Discussion The experiment is performed on the 3 systems with Linux operating system, which has 32 GB RAM and 512 GB storage capacity. The dataset contains 12 GB of unstructured text files. The dataset is collected from Gutenberg dataset. The comparative study analysis of different shard selection algorithms, mainly CORI, ReDDE, and Modified ReDDE was carried out considering 10 and 20 queries, measuring the throughput and similarity score. The Table 1 interprets, as number of shards increase the time taken to retrieve the files is getting reduced. Among the three shard selection algorithm modified ReDDE gives better result as compared to CORI and ReDDE for 10 queries. From Fig. 4 we can observe that, as we increase the number of shards the retrieval time decreases. Initially, with only 1 shard, the time taken to retrieve the data is 864 ms. Similarly, as we increase the number of shards finally the time taken is 516 ms for 7 shards for 10 queries. The Table 2 interprets, as number of shards increase the time taken to retrieve the Table 1 Average time taken to retrieve files for 10 queries by shard selection algorithms Shard selection algorithm

No of shards

Multi-node un-clustered network. throughput (ms)

Multi-node clustered network. throughput (ms)

Score

CORI

1

4653

2611

10.050

3

3957

2304

9.568

5

3455

1804

13.685

7

3409

1661

15.263

1

3422

1365

11.256

3

3956

1217

10.878

5

3250

1113

13.686

7

2753

637

15.668

1

1533

864

12.056

3

1298

843

11.911

5

989

691

13.865

7

756

516

15.979

ReDDE

Modified ReDDE

Efficient Data Partitioning and Retrieval Using Modified ReDDE …

195

Fig. 4 Shard selection algorithms versus retrieval time

Table 2 Average time taken to retrieve files for 20 queries by shard selection algorithms Shard selection algorithms

CORI

ReDDE

Modi-fied ReDDE

No of shards

Multi-node un-clustered network.Throughput (ms)

Multi-node clustered network.Throughput (ms)

Score

1

6859

4583

10.050

3

6258

4257

9.568

5

5756

3758

13.685

7

5214

3599

15.263

1

5344

3591

11.256

3

5122

3052

10.878

5

4752

2651

13.686

7

4487

2257

15.668

1

2586

1576

12.056

3

2159

1322

11.911

5

1739

1049

13.865

7

1583

955

15.979

files is getting reduced. Among the three shard selection algorithm modified ReDDE gives better result as compared to CORI and ReDDE for 20 queries. The Fig. 5, shows the retrieval time for the proposed shard selection algorithm. Initially, with only 1 shard, the time taken to retrieve the data is 1576 ms. Similarly, as we increase the number of shards finally the time taken is 955 ms for 7 shards for 20 queries.

196

P. M. Dhulavvagol et al.

Fig. 5 Shard selection algorithm versus retrieval time

6 Conclusion Elasticsearch is used for storing and processing the large dataset. The shard selection algorithms help to enhance the throughput of the systems. The graphs and tables interpret that the retrieval time of the documents is related to the number of shards of a data node in the elasticsearch cluster. As we increase the number of shards in the cluster, the retrieval time of documents reduces in all the types of clusters. The shard selection algorithm selects the shard which is related to the query, therefore the document similarity score will be high in clustered data as compared to unclustered data. This is because as we increase the number of shards, documents in each shard reduce therefore searching time reduces and data is fetched faster for the given query. The results interpret that the modified ReDDE shard selection algorithm performance is better compared to the CORI and exhaustive shard selection algorithms by reducing overall cost by 26%.

References 1. P. Berglund, Shard selection in distributed collaborative search engines, a design, implementation and evaluation of shard selection in ElasticSearch, University of Gothenburg (2013) 2. A. Kulkarni, J. Callan, Selective Search: Efficient and Effective Search of Large Textual Collections (San Francisco State University, 2015) 3. A. L’Heureux, Machine Learning with Big Data: Challenges and Approaches (The University of Western Ontario, 2017) 4. M.D. Praveen, Vijay, S.G. Totad, Performance Analysis of Distributed Processing System using Shard Selection Technique in Elasticsearch (KLE Technological University, 2019) 5. E. Rodrigues, R. Morlay, Run Time Prediction for Big Data Iterative ML Algorithms: A KMeans Case Study (Faculty of Engineering, University of Porto, Porto, 2017) 6. J.P. Callan, Z. Lu, W.B. Croft, Searching distributed collections with inference networks, ˙ in Proceedings of the 18th Annual International ACM SIGIR Conference on Research and ˙ Development in Information Retrieval (ACM, 1995)

Efficient Data Partitioning and Retrieval Using Modified ReDDE …

197

7. P. Dhulavvagol, V. Bhajantri, S. Totad, Performance analysis of distributed processing system using shard selection techniques on elasticsearch. Procedia Comput. Sci. 167, 1626–1635 (2020). https://doi.org/10.1016/j.procs.2020.03.373

Stock Price Prediction Using Data Mining with Soft Computing Technique R. Suganya and S. Sathya

Abstract Stock costs forecast are interesting and valuable research theme. It provides huge revenue if the prediction favors’ and it makes a huge fall if it goes wrong. Developed nations economies are estimated by their stock trading sector. As of now, financial exchanges are viewed as a celebrated revenue generating field on the grounds that much of the time it gives large benefits with minimal loss and hence generally considered to be safe place of return. Financial exchange with its tremendous and dynamic data handling capacity is considered as a predominant place for profit lovers as well as for research scientists. In this paper, the concept of k-closest neighbor method was appled to select the data from the data set and later the selected datas are used to predict the future stock price using soft computing technique. In soft computing technique genetic algorithmic approach was implemented with a novelty for predicting and improving the performance of the result. Stocks of various companies are analyzed and the results are fine-tuned so the the trust worthable technique was developed. The experimental results show that this combined technique can well be suited for predicting the stock price. Keywords Soft computing · Long-short-filled term memory · Stock price · Data mining · Prediction · Clustering technique

1 Introduction Anticipating return on the securities exchange is a huge issue in monetary organizations and furthermore it is a challenging one. The share value forecast was a challenging method. It is found that for an organization’s share expenses don’t R. Suganya (B) Research Scholar, Department of Information Technology, School of Computing Sciences, Vels Institute of Science, Technology and Advanced Studies (VISTAS), Chennai, Tamilnadu, India e-mail: [email protected] S. Sathya Department of Information Technology, School of Computing Sciences, Vels Institute of Science, Technology and Advanced Studies, (VISTAS), Chennai, Tamilnadu, India e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 I. J. Jacob et al. (eds.), Expert Clouds and Applications, Lecture Notes in Networks and Systems 444, https://doi.org/10.1007/978-981-19-2500-9_14

199

200

R. Suganya and S. Sathya

actually rely exclusively upon the cash related status of the organization, yet furthermore on the nations socio-economic monetary circumstance. It is determined by the country’s specific monetary turn of events or items. So the present stock price forecast has always been a tedious and competitive one. Due to many reasons, for example, political events, company related news, catastrophic events share costs keep varying always. A great deal of exploration in finding stock costs or stock trend has been continuing for a long time. It includes expecting fundamental details and previous data that has been openly available. It is consistently a difficult undertaking in discovering the connection between the prescient information. The cost prevalently relies upon the purchase to sell proportion. The cost will rise sharply in case there is a very big demand in the purchasing conditions. The price will drop in case if there are a greater number of vendors than purchasers. Generally individuals trading are in concern with the exchange or brokers assistance. An intermediary information may help traders in making great stock records. For the vast majority of the stocks, most dealers have proposals dependent on organization data and what generally anticipated them. The effective market hypothesis shows that the stock costs reflect not only on the historical prospects but also on all related parameters. Stock markets are a consistently changing tumultuous business region where expectation assumes a significant part. Forecast always gives the current data related to the stock price. This can thus be utilized in client fields to execute to decide to buy or not the particular portions of particular script. To plan outcome and expectations, a clear predictive analysis is needed for the recorded information. These analysis should be result of the frequently observed and calculated expectations of the stock price. Order is a pack of buy and sell strategies. But the buy or sell depends on the predicted strategies. It’s a bunch of information that forms the characterization method and is utilized for the same. The most famous strategies, for example, decision based tree, linear programming, neural organization and insights are utilized by the above methods. The order of the contents are given below. Introduction was given in Sect. 1, Information Survey works are given in Sect. 2. The proposed model is represented in Sect. 3, Results and Discussion was given at Sect. 4 and atlast the conclusion was given in Sect. 5.

2 Related Work Examination led by Zhauo et al. [1, 2] sorted out some way to discover the inconsistency in exchanging the information between the rise and fall of the stock with respect to the volume of data traded in the stock exchange. They found that there are odd groups that determines the stock cost. The researchers [1] tried to do information mining and found that it is dependable with the group of similarities (k-implies). The ordinary return and profit rate is analyzed against the validity calculation and the estimate will either be a profit or loss. The creators Baviskar and Namdev [1, 3] focused on understanding stock commercial center related realities for purchase

Stock Price Prediction Using Data Mining …

201

and anticipate National Stock Exchange data to are expecting predetermination stock activities. Information mining has been definitely used to separate significant records from verifiable stock information to explore and to anticipate future characteristics. There are a few classifications of technical indicators used to make better analysis of stock. They are price based and volume defined. Weerachart and Benamas [1, 4] put forth their ideas which improves the info mining for analyzing and contributing the various stock levels. Furthermore the author used the gain factor for predicting the future value. The author used to find the presence of valid data, implementation of rank search method and wrapper selection using greedy approach. The author also introduced a novel step search technique. He then used subset elimination by incrementing the known value against the standards. By means of this the predictive quality was improved considerably. Qasim, Assif and Alnaqi [1, 5] utilized the choice tree grouping, and it’s one of the procedures of novel information mining. The CRISP-Data mining approach is utilized to build the proposed model. This review presents a proposition to utilize the choice tree classifier on stock verifiable costs to make choice standards that give financial exchange purchase or sell suggestions. The future value was not so accurate as there are many factors that influence it. Factors including political reasoning, demonetarization and other financial setbacks. Utilizing verifiable time series securities exchange information and information mining strategies, the creators Amgde and Kulgarni [1, 6] fostered a forecast model for foreseeing share market designs relying upon specialized analysis. The trial results got shown the capability of ARIMA model to anticipate the stock value records on transient premise. A financial exchange shows ventures and investment funds that are advantageous to improve the public economy’s effectiveness. R is a language of programming and a designs and handling climate. R-studio permits the client to run R scripts in a more easy to understand climate. The creators Navale et al. [1, 7] utilized man-made analysis and data mining to definitively expect the results. In automation, most examiners have used methods to achieve precision and results. In addition, the limits and execution were improved using the power of data mining. For improving the performance the combined effect of data mining and human intelligence were used together. Desai and Gandhi [1, 8], introduce a novel text mining way to deal with oversee measure the impact of predictable news on stock. They showed a model that predicts changes in stock value that is proportional to the impact of non-quantifiable data, that represents the wealth rather than mere data. This advancement is made to help financial executors with new developed models from obvious data that are most likely to impact. Prasanna and Ezhilmaran [1, 9] proved their work on the findings of the new stock costs. In this survey, the author attempted to produce some remarkable work utilizing information mining techniques to fortell the stock price. His work contributed such to the field of stock prediction and provoked others to concentrate on their performance. The journalists (Dr. P. K. Sahoo and Mr. Krishna Charlapally) of [1, 10] investigated the utilization of auto-backward way to deal with figure stock rates. Because of its straightforwardness and expansive agreeableness, the plan of auto relapse is utilized. They additionally played out an exploration on the auto-backward model’s viability. The strategy for Moore and Penrose is utilized to anticipate the relapse coefficient

202

R. Suganya and S. Sathya

coefficients. What’s more, they additionally concentrated on expectation precision by contrasting the anticipated qualities throughout some stretch of time with the real qualities. Charkha [11] used the typical neural network in conjunction with the stock market learning algorithm [12, 13] and the neruo network patterns for prediction. In this review, key parameters [14] and secret datas [15] are not used as a criteria for prediction [16] and hence it had broken the concept of data selection [17] in stock market analysis for future price. The high end concept of sigmoidal [18] neuto function with variable bias can also be used for classification.

3 Proposed Work Most of the existing models do not include the extern parameters such as political, social and environment factors. In this model, the external factors are considered so that the accuracy of the system can be improved considerably. Various factors that affect the stock market behavior are also considered. The power of data mining was used to select the required data and the soft computing methods are used to fine tune accuracy of the prediction. The research has been carried out by collecting the historical data of the script named CITI (City Union Bank) from NSE website (Fig. 1). Fig. 1 Schematic system architecture

Stock Price Prediction Using Data Mining …

203

Fig. 2 Proposed model for predicting the stock price

3.1 Procedure Step 1: Stock Data collection: The required data is collected from the NSE website for a period of 6 months. It is downloaded in the form of excel sheet. Step 2: Data Selection: Here the required parameters are selected and the other parameters are eliminated. It is the process of valid data selection for stock price prediction. Step 3: In this step, an effective classification method was applied. The effective data mining algorithm was implemented, i.e. k-means. Step 4: The resultant values of different algorithms are now trained by the soft computing model, i.e. algorithm based lstm method. Step 5: By combining the results of the data mining and soft computing approach the future stock price was found.

3.2 Proposed Model The proposed model, selects the required data from the data set. Among the parameters, totally 12 effective parameters were selected based on the technical analysis and also their impact toward the future prediction. For the selected data, separately the following data mining techniques have been implemented and the output was noted (Fig. 2). The data selection is done by means of technical analysis forum and further novel clustering technique of data mining concepts were executed and then the novel algorithm of lstm approach is utilized to predict the future outcome of the stock.

3.3 Clustering Algorithm The Clustering algorithm implemented for the selected data is of k-means type and the results are tabulated as a new data. It is executed as follows.

204

1. 2. 3. 4. 5.

R. Suganya and S. Sathya

Clusters the given data into k known groups. Select the random cluster centers. Objects can find their nearby cluster. Mean calculation of all objects in every cluster. Iterate the above 2, 3, 4 steps until same points are utilized.

3.4 Clustering Technique Clustering is the assignment of collection of a bunch of items so that articles in a similar gathering (called a bunch) are more like each other than to those in different gatherings (groups). Bunch is a gathering of articles that has a place with a similar class. As such, comparative items are gathered in one bunch and disparate articles are assembled in another group. While doing bunch investigation, first segment the arrangement of information into bunches dependent on information comparability and afterward dole out the names to the gatherings. It is unaided learning strategy which is utilized to bunch comparable examples on premise of elements. Based on this the results are tabulated in Table 1. Hence with the defined input data, three more sets of data for training to be given to the next phase i.e. LSTM a soft computing approach. The input for the proposed system is given below (Fig. 3). The above figures shows the various inputs that are considered for stock price prediction. Table 1 Shows the predictive comparing of the various models Date

Prediction using time series analysis

Prediction using Prediction technical using data analysis mining (formula based)

Proposed method (data mining and soft computing)

Actual value (Real value)

17/05/2021

165.34

167.89

168.75

169.25

167.95

18/05/2021

167.99

168.93

169.00

169.23

170.80

19/05/2021

168.92

169.01

169.06

169.68

169.95

20/05/2021

170.00

170.04

170.94

171.00

171.65

21/05/2021

173.21

173.88

174.33

174.98

175.05

24/05/2021

174.06

174.58

175.78

176.02

176.35

25/05/2021

171.84

172.09

172.64

173.01

173.10

26/05/2021

171.95

172.94

173.91

174.11

174.35

27/05/2021

170.03

171.96

172.89

173.39

173.90

28/05/2021

172.01

172.05

172.91

173.67

173.75

Stock Price Prediction Using Data Mining …

205

Fig. 3 Input for stock price prediction

3.5 Proposed Algorithm Step 1: Start the program. Step 2: Required data selection from the old traditional data set of a dealing stock. Step 3: Automatic downloading of the required data. Step 4: Perform feature analysis on the data between 0 and Step 5: Develop a model with nearly sixty time interval stamps and one essential output. Step 6: Build the effective re-trained network with that of the above Step 5. Step 7: Performing KNN, Step 7: Implement first LSTM layer and also perform value reduction. Step 8: Record the information available in the final layer. Step 9: Fine tune it by performing novel enchancements and the loss overcome technique to minimize error. Step 10: Predict and visualize the results using plotting techniques.

3.6 Implementation of Novel LSTM Approach A framework which, with the aid of long short time memory, learns online to anticipate close stock costs. The Long-Short-filled Term Memory is a preliminary example for the RNN methodology for deeper learning purposes, LSTM has input associations, as opposed to traditional feed-in neural network approach. This is one of the information technique used. But also on all types of file such as (e.g. sound

206

R. Suganya and S. Sathya

or movie file) for information. An important step is the collection of information from the market before the processing of the data. In this proposed framework the import of data from advertising clearance organizations such as BSE (Bombay Stock Exchange) and NSE is the main phase (National Stock Exchange). In order to be separated from multiple views, the data set used in the market expectancies must be used. Additionally, the information set adds up to enhance the data set with additional outside information. Most of the information includes the costs of stock for the previous year. Python packages are available for the stock price prediction. The following phase is the processing of the data; pre-processing is an important step forward in information mining, which requires the modification of rough information into a basic setup. The material obtained from the source is contradictory, fragmentary and contains errors. The pre-processing procedure will purify the information; the highlights have to be scaled in order to limit the factors. The model preparation includes cross-approval, a well-founded, anticipated model implementation with the information on preparation. In the computation itself, the target of tuning models is to unequivocally tune the estimation preparing and training.

3.7 Data Set The required data set was derived from the official website of NSE and it is for a duration of 8 months, from Nov 2020 to July 2021. Sample data for the script named MARUTHI SUZUKI was shown below in Fig. 4.

Fig. 4 Sample data set for the script maruthi suzuki

Stock Price Prediction Using Data Mining …

207

4 Results For example consider the bank script named CUB. From the above table it has been found that the proposed model of using data mining with soft computing technique proved to be accurate when compared with the other models (Figs. 5 and 6).

Fig. 5 Predicted results with the proposed model

Fig. 6 Result comparison of the predicted price vs actual price

208

R. Suganya and S. Sathya

Table 2 Performance analysis for different stocks Stock name

LSTM (RMS error value) existing model

Data mining with LSTM (RMS error value)-Proposed model

Indian Overseas Bank

0.49

0.17

Bank of Baroda

0.31

0.22

City union Bank

0.89

0.35

Axis Bank

0.56

0.12

From the above graph it is found that the proposed model gives a good range of accuracy in predicting the future price. It is found that the no. of sample increases the system accuracy will also keep increasing. The performance of this proposed system is compared with the results of the existing LSTM model of four different stocks and the results are tabulated (Table 2).

5 Conclusion An extended and a realistic research approach was conducted and was found with the result proofs that the proposed model gives a very good accuracy in predicting the future stock price. The above experiment further shows that the no. of layers improved the accuracy of the system will also get improved. Hence there is a very good scope for the future researchers in this field of stock price prediction.

References 1. P. Garg, S.K. Vishwakarma, An efficient prediction of share price using data mining techniques. Int. J. Eng. Adv. Technol. (IJEAT) 8(6), 3110–3115 (2019) 2. L. Zhao, L. Wang, Japan advanced ınstitute of science and technology “Price trend prediction of stock market using outlier data mining algorithm”, in 2015 IEEE Fifth International Conference on Big Data and Cloud Computing, (Baylor University, 2015) 3. S. Baviskar, N. Namdev, Analyzing and predicting stock market using data mining techniques. IJIRT 2(3), 76 (2015) 4. W. Lertyingyod, N. Benjamas, Stock price trend prediction using artificial neural network techniques. in IEEE 2016 5. Q. Al-Radaideh, A.A. Assaf, E. Alnagi, Predicting stock prices using data mining techniques, in The International Arab Conference on Information Technology (2013) 6. M.C. Angdi, A.P. Kulkarni, Time series data analysis for stock market prediction using data mining technique with R. Int. J. Adv. Res. Comput. Sci. 6(6) (2015) 7. G.S. Navale, N. Dudhwala, K. Jadhav, P. Gabda, B.K. Vihangam, Prediction of stock market using data mining and artificial ıntelligence. Int. J. Comput. Appl. 134 (2016) 8. R. Desai, S. Gandhi, Stock market prediction using data mining. IJEDR (Int. J. Eng. Dev. Res. 2(2) (2014) 9. S. Prasanna, D. Ezhilmaran, A survey of stock price prediction and estimation using data mining techniques. Int. J. Appl. Eng. Res. 11(6), 4097–4099 (2016). ISSN 0973-4562

Stock Price Prediction Using Data Mining …

209

10. P.K. Sahoo, K. Charlapally, Stock price prediction using regression analysis. Int. J. Sci. Eng. Res. 6(3), Mar-2015 11. P.R. Charkha, Stock price prediction and trend prediction using neural networks. IEEE (2008) 12. A.S. Kumar, R. Suganya, Stock price prediction using tech news based soft computing approach. Int. J. Adv. Trends Comput. Sci. Eng. 9(2), 2049–2054, Apr 2020 13. S. Sathya, R. Suganya, Stock price prediction using datamining with novel computing approach. Turk. J. Comput. Math. Educ. 12(10), 2345–2351 (2021) 14. C. Anand, Comparison of stock price prediction models using pre-trained neural networks. J. Ubiquitous Comput. Commun. Technol. (UCCT) 3(2), 122–134 (2021) 15. P. Karthigaikumar, Industrial quality prediction system through data mining algorithm. J. Electron. Inf. 3(2), 126–137 (2021) 16. M. Tripathi, Sentiment analysis of nepali COVID19 tweets using NB, SVM and LSTM. J. Artif. Intell. 3(03), 151–168 (2021) 17. J.I.-Z. Chen, K.-L. Lai, Deep convolution neural network model for credit-card fraud detection and alert. J. Artif. Intell. 3(02), 101–112 (2021) 18. S.R. Mugunthan, T. Vijayakumar, Design of ımproved version of sigmoidal function with biases for classification task in ELM domain. J. Soft Comput. Paradigm (JSCP) 3(02), 70–82 (2021)

A Complete Analysis of Communication Protocols Used in IoT Priya Matta, Sanjeev Kukreti, and Sonal Sharma

Abstract With the proliferation of wireless networking, computer miniaturization, incorporation of computing technologies into every object, and the emergence of Internet connectivity everywhere, Internet-of-Things (IoT) has become the world’s most influential framework. IoT like all other network-based paradigms, deals with communication. Besides the expansion of connectivity and growth in the count of connected devices, there is an increased requirement of efficient and better solutions to connect them, make them communicate without a fail. To accomplish this communication, there are a number of well-defined and standardized set of rules, known as protocols. Various researchers have defined different layers in an IoT System. Each layer has one or the other protocol. In each layer, one can find a number of protocols, forming a protocol family for a specific layer. These protocols are discussed in sequence corresponding to the different layers. The technical aspects like expenses in terms of size and speed, and suitability to the application domain is elaborated in detail. Some of the majorly discussed protocols are 6LowPAN, IPv6, RPL, Wifi, Bluetooth, mDNS, DNS-SD, MQTT, CoAP, AMQP, Websocket, OMA-DM. This paper also offers the wide-ranging and most inclusive features of above-mentioned renowned protocols available in the market, with a proper comparison among them. Keywords Wireless-networking · Internet-of-Things · Communication · Protocols · Application domain

1 Introduction Internet of Things is a paradigm where interrelated objects are interconnected to each other via a global network, the Internet. The things that are present in this type of P. Matta (B) · S. Kukreti Department of Department of Computer Science and Engineering, Graphic Era University, Dehradun, India e-mail: [email protected] S. Sharma Uttaranchal University, Dehradun, India © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 I. J. Jacob et al. (eds.), Expert Clouds and Applications, Lecture Notes in Networks and Systems 444, https://doi.org/10.1007/978-981-19-2500-9_15

211

212

P. Matta et al.

scenario consist of circuits, electronics, software, sensors and various types of tools that are used for connecting as well as exchanging the data with other devices. In this era of technology, we all are surrounded with smart gadgets around us like smart phones, smart watches, smart home appliances etc. Every time, somehow, we are always connected to Internet by different types of means, the main reasons behind this is, that IoT is playing a vital role in our lives as it made the things easier, fast, precise, and convenient that later on help us in our work to make it easier. According to Elhadi et al. [1], “the concept of a network of smart devices was discussed as early as 1982, with a modified Coke machine at Carnegie Mellon University becoming the first internet-connected appliance, able to report its inventory and whether newly loaded drinks were cold.” In 1999, Kevin Ashton presented with a new terminology, Internet-of-Things, to define a scenario in which all the things existing in the physical world are connected via the Internet. The expansion of Internet connectivity from the computing devices to comparatively unintelligent devices like chair, table, dryer, refrigerator, is the main aim behind the concept of IoT. According to Sungeetha et al. [2], “users and things can be connected at all times and locations using IoT.” According to Jacob et al. [3], “IoT is an ecosystem comprised of multiple devices and connections, a large number of users, and a massive amount of data.” IoT converts everything to a “smart-thing” and therefore, advances every facet of our living. IoT is usually described as “dynamic global network infrastructure with selfconfiguring capabilities based on standards and communication protocols.” For largescale scenarios Hanes et al. [4] described the paradigm as, “Internet of Things envisions a self-configuring, adaptive, complex network that interconnects things to the Internet through the utilization of standard communication protocols.” According to Arseni et al. [5] “following the complete trend, the IT paradigm called IoT aims to group each technological end-point that has the ability to communicate, under the same umbrella.” According to Miorandi et al. [6], “since the first day it was conceived, the IoT has been considered as the technology for seamlessly integrating classical networks and networked objects.” These smart objects are interconnected to each other with embedded electronics, computing capabilities, ability to sense and correspondingly actuate, and can have a unique existence in the physical as well as information world. After introducing the concept of IoT, the paper follows four more sections. In second section, we have discussed the characteristics of IoT. Section third discusses background and related work. Next section contains some of the already proposed architectures for an IoT system. Different layers have been identified in the same section. Following the identification of layers, their respective protocols are discussed in detail. Section five is conclusion of the paper.

2 Characteristics of IoT The concept of “smart things” targets at implementing technological tools in each and every “thing,” that is physically existing and feasible in every domain. At every

A Complete Analysis of Communication Protocols …

213

level and aspect of connected things, IoT should be considered as a major constituent. According to Chen and Kunz [7], since the IoT is a set of tools and technologies, it is necessary to take into account each level and aspect of a connected object, modeled as a system in its own right. Although, there are various components that assemble together to design and develop an IoT system, but the complete IoT system has some key characteristics, that are mentioned below. Connectivity: In IoT, connectivity means all the devices are able to communicate with each other, exchange the data among themselves. They are connected to each other via any type of network, mean they are not required to be connected via any predominantly major providers. Devices can make their own network that can exist on a much smaller and cheaper scale while still being functional. IoT creates these small networks between its system devices. Small devices: Nowadays the size of devices is decreasing, devices are available at low costs, and are becoming more powerful over time. IoT is working on the area of making devices small and making it more precise, scalable, portable, and versatile. Sensors and actuators: Micro-Electrical-Mechanical-Systems (MEMS) are unavoidable components of an IoT system. Sensors as well as actuators are those MEMS which convert functional energy to electrical signals and electrical signals to functional energy respectively. These form a crucial component that results into a transition of passive network of things to an active network of smart devices. Convenient control: The devices at home, office, small-scale or large-scale industries, hospitals, schools, institutes can be conveniently handled and controlled. Active engagement: All the connected technology, product, or services have to be actively engaged with each other to pursue as an active component of real time IoT system. All of them must possess active engagement with each other by the help of IoT. Artificial intelligence—Today, IoT has made every device that much smart that it knows how to react or respond over a particular situation. For example, if there is a refrigerator without butter and bread, it can itself order the required amount of bread and butter from the appropriate shopping complex.

3 Background and Related Work Whenever a paradigm emerges, its components, underlying platforms, supporting technologies are the major considerations. Similarly, whenever a new system of that paradigm is to be designed and developed, the major focus is on the architecture of the system. As there is no single agreed generalized architecture of an IoT system, therefore, various researchers proposed different architectures for different applications. Various suggestions are proposed to design and develop a domain-specific IoT architecture. Static component of the architecture does not allow it to serve the conditions where the requirement of diverse applications is completely dynamic in nature. Beier et al. [8] proposed that, “EPC global IoT architecture mainly focuses

214

P. Matta et al.

on the RFID network and smart logistics system can be considered as solution for architecture.” According to some researchers, IoT model can be easily defined in terms of different layers. For example, Gronbaek [9] and Dai et al. [10] have proposed that, “architectural model similar to the open systems interconnection (OSI) architecture are fruitful on IoT too.” Gronbaek [9] also declared that, “the architecture will support ubiquitous services on an end-to-end basis.” They deliberated it on the basis of four levels: Things layer, Adaptation layer, Internet layer, and Application layer. Tan and Wang [11] proposed, “a five-layered architecture to represent an IoT system completely, consisting of Application layer, Middleware layer, Coordination Backbone layer, Access layer, Edge Technology layer.” Ma [12] conversed that, “the architecture of IoT should be studied from the viewpoint of users, network providers, application developers and service providers, providing the basis for defining a wide variety of interfaces, protocols and standards.” Jing et al. [13] proposed, “a threelayered architecture is enough to handle all the tasks of an IoT system. The three layers are namely: Perception layer, Transportation layer and Application layer.” Similarly, Bandyopadhyay and Sen [14] proposed, “a generic five-layer architecture consisting of edge technology layer, access gateway layer, internet layer, middleware layer, and application layer, can describe the overall design of IoT.” Sarkar et al. [15] offered that, “the functionalities of IoT infrastructure are grouped into three layers, which are: (1) virtual object layer (VOL); (2) composite virtual object layer (CVOL); and (3) service layer (SL).” According to Ning and Wang [16], “a centralized web-resource-based architecture can be generated for decoupling the application development from the domain of heterogeneous devices.” They gave the concept of using a thin client, where a thin server can be used as a server and does not contain any application logic. And therefore, such an IoT model can support a variety of IoT applications. Castellani et al. [] proposed a specialized architecture, designed and developed for the application implemented in smart offices. It includes smart door for authenticated entry using unique identification techniques like RFID and secure network. The proposed model is composed of three types of nodes, namely, Base Station Node (BSN), Mobile Node (MN), and Specialized Node (SN). Yashiro et al. [17] proposed uID architecture, that can easily be reflected as a welldefined platform for things-oriented as well as semantic-oriented IoT systems. Its two main constituents are uCode and uID database. According to them, “concept of uIDCoAP design is to mitigate the burdens of manufacturers to add IoT communication functions to their existing products.” Ungurean [18] proposed an architecture based on OPC.NET specifications that is based on two main modules: the data server and the HMI application. According to them, “the data server acquires data from a network of sensors (fieldbus) and sends commands to the actuators (such as relays, electro valves, etc.) that are connected on the fieldbuses.” According to Rahman et al. [19], the architecture of IoT consists of five layers namely: Physical layer, MAC layer, adaptation layer, network layer, and application layer.

A Complete Analysis of Communication Protocols …

215

4 IoT Architecture Many researchers proposed three-layered architecture, while others proposed fourlayered and five-layered architecture too. The different architectures proposed by different researchers [20–23] are given below (Figs. 1, 2, 3 and 4): A three-layer architecture is the most accepted IoT architecture, that covers almost all the functionalities of an IoT system. As the name suggests, it has three layers namely, perception layer, network layer, and application layer. Perception layer is also referred as physical layer, as it deals with all the hardware components and physical devices. Sometimes it is also termed as sensing layer, as it contains the sensors which are embedded into physical things. The main task of this layer is to gather data from smart devices and transfer the collected data to the network layer. Network layers forms the second layer of the architecture. It basically acts as an interface between perception layer and application layer. The major issues of consideration in this layer are connection technologies, bandwidth, and secured connection. The connectivity may be wired or wireless, connection may be simplex, half duplex or full duplex. Application layer forms the last layer that receives the data from network layer, analyze it and process it to draw decisions. This analysis sends the response and result back to the perception layer via network layer, to act accordingly.

Fig. 1 Three-layered architecture [20]

216 Fig. 2 Three-layered architecture [21]

Fig. 3 Three-layered architecture [22]

P. Matta et al.

A Complete Analysis of Communication Protocols …

217

Fig. 4 Four-layered architecture [23]

5 IoT Protocols Whenever a system is composed of multiple layers, there are some well-defined rules of communication among those layers. Each layer has to follow a set of rules to communicate with the other layer as well as is peer on the other end. The communication among the smart devices also requires some communication rules. To understand and accomplish this communication, there are a number of well-defined and standardized set of rules. These set of rules are referred as protocols. They enable various devices embedded in different systems and using different networks to communicate each other. The massive Internet-of-Things system is the more complex as the protocols are required to ensure seamless and flawless communication. While discussing a general three-layered architecture of IoT, one can find a number of protocols, forming a protocol family for a specific layer. In previous section, we have identified a number of layers. Each layer will have its own set of rules and regulations, or precisely termed as protocols. These protocols are discussed in sequence corresponding to the different layers. CoAP: Internet Engineering Task Force (IETF) designed and developed Constrained Application Protocol (CoAP). According to R6, “CoAP has been defined as a technology enabler to allow applications to interact with physical objects.” CoAP also enables the merging of constrained devices into the IoT via constrained networks. This integration is possible even if the network availability as well as bandwidth is quite low. The base of CoAP is UDP, therefore overall implementation of CoAP is considered very lightweight. CoAP make use of all the commands provided by HTTP, therefore, efficiently supports client/server architecture. According to Rodrigues

218

P. Matta et al.

[24], “Where TCP-based protocols like MQTT fail to exchange information and communicate effectively, CoAP can continue to work.” According to Yashiro et al. [25], “CoAP CHTTP Mapping enables CoAP clients to access resources on HTTP servers through a reverse proxy that translates the HTTP Status codes to the Response codes of CoAP.” Some of the well-known characteristics of CoAP are as follows: • • • • •

Can work with the networks having low bandwidth. CoAP is a request/response protocol CoAP makes proper utilization of synchronous as well as asynchronous messages. CoAP runs over UDP Lowers down the overheads as encountered by TCP

MQTT: Message Queuing Telemetry Transport (MQTT), designed and developed by IBM is emerged as universal protocol for all kind of IoT projects. It is basically meant for lightweight M2M communications. MQTT functions by using TCP as its major base and is categorized as asynchronous publish/subscribe protocol. Instead of preferring request/response protocols, IoT systems favor publish/subscribe protocols, as the requirements of IoT projects can easily be met by publish/subscribe protocols. According to [23], “Publish/subscribe protocols meet better the IoT requirements than request/response since clients do not have to request updates thus, the network bandwidth is decreasing and the need for using computational resources is dropping.” Some of the well-known characteristics of MQTT are as follows: • • • •

Low energy requirement. The technique used by MQTT is “messaging-passing.” Size of data packets is very less, resulting in less bandwidth requirements. MQTT is a lightweight protocol, therefore, generally implementable in all kinds of projects. • MQTT runs over TCP. • A well-known example: MQTT is being used by Facebook. AMQP: To facilitate encrypted messaging among organizations and applications, and for asynchronous messaging by wire an open source published standard Advanced Message Queuing Protocol (AMQP) was introduced. According to Yokotani and Sasaki [26], “Study shows that comparing AMQP with the REST, AMQP can send a larger number of messages per second.” Because of its features like portability, secure, efficiency, and support of multichannel, AMQP is used extensively in IoT device management and client/server messaging. Even after network disruptions AMQP’s store and forward feature ensures reliability and is its key advantage. According to Yashiro et al. [25], “It ensures reliability with the following message-delivery guarantees: • At most once: means that a message is sent once either if it is delivered or not. • At least once: means that a message will be definitely delivered one time, possibly more. • Exactly once: means that a message will be delivered only one time.”

A Complete Analysis of Communication Protocols …

219

Some of the well-known characteristics of AMQP are as follows: • Can ensure reliability even in disrupted network. • Used to facilitate encrypted messaging. • It is open-sourced published standard. WEBSOCKET: As an enhancement to a standard HTTP connection the WebSocket was developed as a bi-directional communication protocol. The WebSocket’s base is TCP, and has newly come with the introduction of HTML5. For communication among client and server the technique which is used is full-duplex message-based. WebSocket is a kind of protocol which is not a request/response and publish/subscribe protocol. For establishment of a WebSocket session, a handshake with a server is initialized by the client. For handling the WebSocket sessions along with the HTTP connections over the same port by the web servers, the handshake in itself is identical to HTTP. Overhead in WebSocket messages are of only 2 bytes at the time of the session. According to Yashiro et al. [25], “It is designed for real-time communication, it is secure, it minimizes overhead and with the use of WAMP it can provide efficient messaging systems.” Some of the well-known characteristics of WebSocket are as follows: • It is an upgrade to HTTP. • It is a bi-directional communication protocol. • Operates over TCP. OMA DM: For the constrained devices having limited bandwidth, the Open Mobile Alliance Device Management (OMA DM) was designed and released in 2003. Its specifications were predecessor of LWM2M which were developed for mobile phones, tablets, and PDAs. The OMA DM gives many functions for the purpose of mobile device management and supports M2M communication through the range of protocols consisting HTTP, WAP, or SMS. The language which is defined by OMA-DM and is heavily dependent on it is SyncML. According to Ren et al. [27], “The typical underlying transport protocol is HTTP. This makes OMA-DM in unaltered form infeasible for constrained devices.” So basically, to support firmware updates, configuration, fault management and provisioning, OMA-DM should be preferred to IoT devices. Some of the well-known characteristics of OMA DM are as follows: • Designed for constrained devices. • Supports M2M communication. • Dependent on SyncML. Bluetooth: Bluetooth is the most preferred protocol for the short-range communication and building Personal Area Network (PAN). Bluetooth for transmission of data uses a 2.4 GHz wireless radio wave link and is commonly used protocol in IoT for wireless data transfer. Bluetooth really fits in for making wireless data transfer very cheap, short range, less power consumptive between the computing devices. The major section of devices where the Bluetooth protocol is taken on work are smartphones, smart wearable devices based on PAN and other portable devices,

220

P. Matta et al.

where without using big amount of memory and power small segments of data is transferred. There is another version of Bluetooth known as Bluetooth Low Energy (BLE) which consumes very less energy and has an important part in connecting the IoT devices. Some of the well-known characteristics of Bluetooth are as follows: • Uses 2.4 GHz frequency radio wave link for data transfer. • Very cheap, short ranged and less power consumptive. • Used in smartphones, smart wearable devices etc. IPv6: To tackle the long-anticipated problem of IPv4 address exhaustion, Internet Engineering Task Force (IETF) developed the IPv6 protocol. The count of available address space in IPv6 is 2^128 which in comparison to IPv4 are much more. IPv6 can facilitate up to 340 undecillion unique IP addresses. IPv6 supports auto-configuration, integrated security, and a variety of new mobility features, which in turn enables a higher degree of network complexity. The architecture of IPv6 and IPv4 is mostly similar. IPv6 is defenseless to malicious activities like IP scanning just because it provides us more address space. 6LowPAN: 6LowPAN is IP based technology which came into existence in 2007. Its an IPv6 low power wireless Personal Area Network that interpret encapsulation and header compression mechanisms. It was introduced because a low-powered energy IoT protocol was extensively required at that time, that’s the reason why it replaced IPv6. It has a major role in IoT wireless communication. According to Viswanath et al. [28], “It acts by supporting addresses with different lengths, low bandwidth, star and mesh topologies, battery supplied devices, low cost, large number of devices, unknown node positions, high unreliability, and long idle periods during when communications interfaces are turned off to save energy.” Some of the well-known characteristics of 6LowPAN are as follows: • IP based technology. • Interpret encapsulation and header compression mechanisms. Wi-Fi: It is a wireless protocol that operates on 2.4/5 GHz frequency and was developed to replace Ethernet using wireless communication, to provide easy to implement and easy to use short ranged wireless connectivity. Spectrum cost of Wi-Fi is zero. Wi-Fi is now an undeniable choice for IoT connectivity due to its zero-spectrum cost and the coverage of Wi-Fi is almost everywhere, but it not a correct choice always. Right now 802.11n is the commonly used Wi-Fi in home and business, it offers throughput of about hundreds of megabit/sec, which is good for file transmission but for various IoT application it is lot power consuming. Some of the well-known characteristics of Wi-Fi are as follows: • Easy to implement and use. • Provides throughput of hundreds of megabit/sec. RPL: Routing Protocol for Low-Power and Lossy Network is a distance vector routing protocol which runs over IPv6 and as its name suggests is developed for Low-Power and Lossy Network (LLNs). Message confidentiality and integrity is also

A Complete Analysis of Communication Protocols …

221

supported by RPL. This protocol works on IEEE 802.15.4. If link layer mechanisms are not accessible and inappropriate, RPL can use its own mechanism. According to Viswanath et al. [26], “Network devices running RPL protocol are connected with no present cycles. Due to that, the Destination Oriented Directed Acyclic Graph (DODAG) was built and is called a DODAG root.” mDNS, DNS-SD: For the desktop systems having nearly no limit of bandwidth Multicast Domain Name System (mDNS), Domain Name System Service Discovery (DNS-SD) were developed, but it lacks optimization for smart object networks which have low data rate. Even without any help from the servers in local network, mDNS allows to map domain name to network address. During the usage of DNS resource records, DNS SD facilitates to discover and to broadcast services in a network.

6 Conclusion With the vast emergence of IoT in day today life, it is highly required that the IoT systems should be highly efficient. Their efficiency is the actual major to their appropriate use. In any IoT system, the most critical part is communication. IoT like all other network-based paradigms, deals with communication. To accomplish any communication perfectly, there should be a number of well-defined and standardized set of rules, known as protocols. Time to time various protocols are defined and designed by researchers for IoT. Some of the well-known protocols are discussed in this work. Some of the majorly discussed protocols are 6LowPAN, IPv6, EPC, Wifi, Bluetooth, mDNS, DNS-SD, MQTT, CoAP, AMQP, OMA-DM, JSON-LD. Their technical specifications, advantages, and disadvantages are also discussed. This paper can give a brief overview to those researchers, who are going to develop IoT systems for different applications.

References 1. S. Elhadi et al., Comparative study of IoT protocols. Smart Appl. Data Anal. Smart Cities (SADASC’18) (2018) 2. A. Sungheetha, R. Sharma, Real time monitoring and fire detection using internet of things and cloud based drones. J. Soft Comput. Paradigm (JSCP) 2(03), 168–174 (2020) 3. I.J. Jacob, P.E. Darney, Design of deep learning algorithm for IoT application by image based recognition. J. ISMAC 3(03), 276–290 (2021) 4. D. Hanes, G. Salgueiro, P. Grossetete, R. Barton, J. Henry. IoT fundamentals: Networking technologies, protocols, and use cases for the internet of things. Cisco Press (2017) 5. S.-C. Arseni et al., Analysis of the security solutions implemented in current Internet of Things platforms, in 2015 Conference Grid, Cloud & High Performance Computing in Science (ROLCG) (IEEE, 2015) 6. D. Miorandi, S. Sicari, F. De Pellegrini, I. Chlamtac, Internet of things: Vision, applications and research challenges. Ad Hoc Netw. 10(7), 1497–1516 (2012)

222

P. Matta et al.

7. Y. Chen, T. Kunz, Performance evaluation of IoT protocols under a constrained wireless access network, in 2016 International Conference on Selected Topics in Mobile and Wireless Networking (MoWNeT) (IEEE, 2016) 8. S. Beier et al., Discovery services-enabling RFID traceability in EPC global networks, COMAD 6 (2006) 9. I. Gronbaek, Architecture for the Internet of Things (IoT): API and interconnect, in 2008 Second International Conference on Sensor Technologies and Applications (sensorcomm 2008) (IEEE, 2008) 10. G.P. Dai, Y. Wang, Design on architecture of internet of things, in Advances in Computer Science and Information Engineering (Springer, Berlin, Heidelberg, 2012). pp. 1–7 11. L. Tan, N. Wang, Future internet: The internet of things, in 2010 3rd International Conference on Advanced Computer Theory and Engineering (ICACTE), vol. 5 (IEEE, 2010) 12. H.D. Ma, Internet of things: Objectives and scientific challenges. J. Comput. Sci. Technol. 26(6), 919–924 (2011). https://doi.org/10.1007/s11390-011-1189-5 13. Q. Jing, A.V. Vasilakos, J. Lu, Security of the internet of things: Perspectives and challenges. Wirel. Netw. Springer 20(8), 2481–2501 (2014) 14. D. Bandyopadhyay, J. Sen, Internet of things: Applications and challenges in technology and standardization. Wireless Pers. Commun. 58(1), 49–69 (2011). https://doi.org/10.1007/s11277011-0288-5 15. C. Sarkar, A.U. Akshay, R.V. Prasad, A. Rahim, R. Neisse, G. Baldini, DIAT: A scalable distributed architecture for IoT. IEEE Internet Things J. 2(3), 230–239 (2015). https://doi.org/ 10.1109/JIOT.2014.2387155 16. H. Ning, Z. Wang, Future Internet of Things architecture: Like mankind neural system or social organization framework? IEEE Commun. Let. 15(4), 461–463 (2011) 17. A.P. Castellani, N. Bui, P. Casari, M. Rossi, Z. Shelby, M. Zorzi, Architecture and protocols for the Internet of Things: A case study, in Proceedings of 8th IEEE International Conference on Pervasive Computing and Communication Workshops (PERCOM). (Mainheim, Germany, 2010), pp. 678–683 18. T. Yashiro, S. Kobayashi, N. Koshizuka, Sakamura, K., An Internet of Things (IoT) architecture for embedded appliances, in 2013 IEEE Region 10 Humanitarian Technology Conference, (Sendai, 2013), pp. 314–319 19. I. Ungurean, N.-C. Gaitan, V.G. Gaitan, An IoT architecture for things from industrial environment, in 2014 10th International Conference on Communications (COMM) (IEEE, 2014) 20. A. Hakiri, P. Berthou, A. Gokhale, S. Abdellatif, Publish/subscribe-enabled software defined networking for efficient and scalable IoT communications. IEEE Commun. Mag. 53(9), 48–54 (2015) 21. P. Lea, Internet of Things for architects: Architecting IoT solutions by implementing sensors, communication infrastructure, edge computing, analytics, and security. Packt Publishing Ltd. (2018) 22. J. Frahim, C. Pignataro, J. Apcar, M. Morrow, Securing the internet of things: a proposed framework. Cisco White Paper (2015) 23. Serving at the edge: A scalable IoT architecture based on transparent computing 24. T.G. Rodrigues et al., Hybrid method for minimizing service delay in edge cloud computing through VM migration and transmission power control. IEEE Trans. Comput. 66(5), 810–819 (2017) 25. T. Yashiro et al., An Internet of Things (IoT) architecture for embedded appliances, in 2013 IEEE Region 10 Humanitarian Technology Conference (IEEE, 2013) 26. T. Yokotani, Y. Sasaki,Transfer protocols of tiny data blocks in IoT and their performance evaluation, in 2016 IEEE 3rd World Forum on Internet of Things (WF-IoT) (Reston, VA, 2016), pp. 54–57 27. J. Ren, H. Guo, C. Xu, Y. Zhang, Serving at the edge: A scalable IoT architecture based on transparent computing. IEEE Netw. 31(5), 96–105 (2017)

A Complete Analysis of Communication Protocols …

223

28. S.K. Viswanath et al., System design of the internet of things for residential smart grid. IEEE Wirel. Commun. 23(5), 90–98 (2016). https://doi.org/10.1109/MWC.2016.7721747 29. R.A. Rahman, B. Shah, Security analysis of IoT protocols: A focus in CoAP, in 2016 3rd MEC International Conference on Big Data and Smart City (ICBDSC) (2016), pp. 1–7

Image Captioning Using Deep Learning Model Disha Patel, Ankita Gandhi, and Zubin Bhaidasna

Abstract Image Captioning means that the natural language descriptions are generated automatically based on the content of an image. It’s an important aspect of scene comprehension since it integrates computer vision and natural language processing knowledge. Numerous methods and algorithms are developed by researchers to increase the accuracy of image captioning. However, it is one of the major questions for future researchers to get optimized result in captioning an image. Furthermore, there are thousands of gray-scale images that are captioned. In this proposed work, different pre-trained models are used to extract features of images through Convolutional Neural Network (CNN) for colored images and gray-scale images from dataset and then, the extracted features are fed into LSTM, which generates caption for images. At last, the model’s accuracy of color and gray-scale images are studied to determine the model’s capability in captioning both types of images. Keywords CNN—Convolutional neural network · LSTM—Long short-term memory · R, G, B—(Red, Green, Blue) · NLP—Natural language processing

1 Introduction Deep learning is a machine learning and Artificial Intelligence (AI) technique that mimics how humans acquire knowledge. Deep learning is highly useful for data scientists who are concerned with gathering, analyzing, and interpreting massive amounts of data; it speeds up and simplifies the process. Image processing [1] is a technique for performing operations on images in order to improve them or extract relevant information from them. In science and industry, D. Patel (B) · A. Gandhi · Z. Bhaidasna Parul Institute of Engineering and Technology, Vadodara, India e-mail: [email protected] A. Gandhi e-mail: [email protected] Z. Bhaidasna e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 I. J. Jacob et al. (eds.), Expert Clouds and Applications, Lecture Notes in Networks and Systems 444, https://doi.org/10.1007/978-981-19-2500-9_16

225

226

D. Patel et al.

image processing [1] has been and will continue to play an essential role. Number of applications, including visual recognition, scene understanding, etc., uses image processing. Caption generation [2] is a fascinating artificial intelligence challenge that involves generating a descriptive text for a given image. It uses two computer vision approaches to comprehend the image’s content, as well as a language model from the field of Natural Language Processing (NLP) to convert the image’s comprehension into words in the correct order [2]. Computer vision, NLP, and machine learning are all used to solve the problem of image caption [3]. Caption generation is a difficult artificial intelligence task in which a textual explanation for a given photograph must be generated. Image captioning is a solution to the real-life problems faced by people, especially for the vision and hearing-impaired person. This concept is not only implemented with images but can also be used with videos. There are also other applications like making a new dataset in virtual assistant, image indexing, etc.

2 Review of Literature Here is the summary of relevant papers with their detail in literature survey table. In [2], the author proposed a hybrid method that employs a multilayer CNN to produce features and an LSTM to arrange words into sentences. In that work, CNN compares the training photos to produce a correct caption for an image. It outperformed state-of-the-art models when using Bleu Metrics, a technique for evaluating model performance (Fig. 1). Firstly, in the above figure, the images are taken which are given to a pre-trained model for its features extraction. On the other side, the captions given in datasets are utilized and pre-processing is done (finding out tokens, vocabulary, etc). After that, both the outputs i.e., extracted features as well as the output of all processed data, are given while constructing the model for image captioning with RNN. At last, the caption is generated by the model for the respective image. A novel technique was presented in study [4]. The major goal was to caption an image directly from a gray-scale photograph. With the help of Inception V3, the features of the photos were extracted using a pre-trained model. Following that, those characteristics were fed into the LSTM (Long Short-Term Memory) model. However, it achieved a 46% total accuracy (Fig. 2). The flow of the whole system is seen in the above diagram. Here, Inception V3 is used as pre-trained model and its features are taken as first inputs in the model, which is followed by Dropout and Dense layer. Secondly, the embedded layer is taken after the second input. It is followed by the Dropout layer with probability of 0.5 and last, LSTM layer of 256 units is added. At last, Decoder model is made with the extracted inputs and LSTM which passes through Dense layer of 256 units having ReLU as activation function. Finally, SoftMax layer is added to the output probability.

Image Captioning Using Deep Learning Model

227

Fig. 1 Architecture of image captioning [2]

The paper [1] demonstrated a way for captioning both photos and videos that is comparable. CNN and RNN (LSTM), as well as GAN, are utilized to caption images. CNN and RNN were employed in the generation of captions for videos, and LSTM model with GAN was used where videos were segmented into frames of pictures to be processed. The work had a quick rundown of picture and video captioning. In the research work [5], MLADIC is a multitask system that used a dual learning technique to maximize two linked objectives: image captioning and text-to-image synthesis. Encoder-decoder model (CNN-LSTM) was used for image captioning, and C-GAN was used for picture synthesis (conditional generative adversarial network). One is the target domain, which included the flicker30k and oxford-102 datasets, while the other is the source domain, which included the MS-COCO dataset. In comparison to the target domain, MLADIC performed better. In future, large-scale unpaired photos and text can be gathered from a variety of web sources, such as Pinterest for photographs and Wikipedia for content. The research work [6] introduced a new picture captioning difficulty, namely, characterizing an image under a specific theme. A cross-model embedding of an image was taught to tackle this challenge. On the MS-COCO and Flickr30K datasets, the suggested method has performed good result for both captioning image-retrieval and caption production. The new framework gives users more control over how captions for photographs are generated, which could lead to some interesting applications. It was found to be superior to other approaches. It delivered good results on both datasets. In [7], captioning is a project that described the content of RSIs. To describe RSL 5 sentences were used in caption dataset. In previous approaches, five sentences were given separately, which could result in an unclear sentence. To treat five sentences as a whole, a collection of words with the same topic terms should contain the same

228

D. Patel et al.

Fig. 2 Schematic diagram of LSTM network [4]

information in all five sentences. In the training phase, the first topic repository was supplied to record subject words. The topic terms were then obtained from the topic repository for testing purposes. Finally, topic words were fed into a recurrent memory network to help generate sentences. The test image’s topic words can also be altered. The question of how to limit the recurrence of these situations will be addressed in future research. In the research work [8], to resolve these issues, an online positive recall and missing concepts mining method were offered in this study. The loss of distinct samples based on their online positive recall forecasts was re-weighted. For missing concept mining, there were two stages. An element-by-element selection procedure was used to determine the most appropriate concepts at each time for caption generation, so that the image description can be described very precisely. The relationship between concepts would be clearly extracted in future work in order to increase performance.

Image Captioning Using Deep Learning Model

229

The paper [3] proposed a multilayer dense attention model for image caption. The image features for the coding part were extracted using a faster recurrent convolutional neural network, the multilayer dense attention model which was decoded using the LSTM-attend, and using LSTM-attend, the description text was generated. Parameters were defined using strategy gradient optimization in reinforcement learning. The dense attention methods were used in coding stage to effectively avoid non-salient information interference, and the relevant descriptive text for the decoding process is output. As a result, photos were analyzed and text was generated with ease. In the research paper [9], the old way of text generation was obsolete. It did not have any connections, such as those between famous people, institutions, and buildings. In the study, the user focused on annotating a news visual, which can aid in gaining real-world knowledge about the story behind the entire image. The researcher proposed a fresh technique that captions the image more precisely to a life-like explanation by exploiting the semantic importance of the listed elements. Furthermore, it extracted text from news and examines the relationships between them. In particular, the user recommended to take out some information from news using a phrase correlation analysis algorithm. There is no Optimal Image Text recognition Training Set in the work [10]. An unique method entitled sglstm was designed to overcome the unavoidable imbalance of message in datasets. FLICKERNYC was built from Flicker Dataset, and a unique guiding textual feature was acquired using a multimodal LSTM model. Image content and description were inextricably linked while teaching mlstm. The leading information was then used as extra input to the network during sglstm development, together with the picture representations and corresponding ground truth descriptions. These inputs were passed to the multimodal block and the caption was generated. In [11], a new system “RSI” was introduced that generated captions using finding out the relationships between objects with their attributes, which is present in RS images with caption. The first step was to encode image visual features, then generate the caption. Second step was to convert that generated caption into meaningful feature vectors. Last step was to measure the comparison between both the vectors of the query picture’s caption as well as those of archive photos and then retrieve the most comparable photos to the query image. They intend to improve the captioning area in future. In [12], Prior RL-based picture captioning strategies concentrated mostly on a specific policy network and reward function—an approach that is unsuitable for multilevel (keyword and phrase) and multimodal image captioning. A novel image labeling framework was suggested for optimization that can be readily combined with RNN-based captioning models, language metrics, or visual semantic functions. It is comprised of two components: a multilevel policy network that modifies the word and sentence-level rules for word formation simultaneously, and a multilevel reward function that uses both a vision language reward and a language reward cooperatively. The research work [13] supervised learning model that integrates multiple deep learning approaches in order to investigate feasibility of capturing the difference between two image characteristics in order to generate a language model probability

230

D. Patel et al.

distribution. The feature of an image pair was extracted using deep Siamese convolutional neural network, and then salient regions of feature vector were detected using an attention mechanism, allowing a bidirectional LSTM decoder to generate a matching and semantically associated caption sequence. The model compared a pair of images and corresponding caption. In the paper [14], the system was typically a top-down based mechanism. These models were inquired from image features using hidden LSTM, rather being optimized. A paradigm called VAA (Visual Aligning Attention) was presented to address issues caused by top-down based mechanisms, such as losing concentration and resulting in decreased accuracy. During the training phase, a well-designed visual positioning loss optimized the attention layer. By directly measuring the feature comparison of attended image characteristics and corresponding words inserting vectors, the visual alignment loss can be calculated. It filtered out the non-visual words in sentences from the visual vocabulary. In [15], how to effectively increase the “vertical depth” of encoder-decoder remains to be solved. For picture captioning, a new technique called Deep Hierarchical Encoder-Decoder Network (DHEDN) was created. The structure was exposed to different encoders and decoders in this method. In caption generation, it has the ability to combine high-level semantics vision and language. It has three layers in its model. The first component is an LSTM decoder. The middle layer has an encoderdecoder to improve the top layer’s decoding ability, while the bottom layer had an LSTM for textual input encoding. The Qualitative studies showed how this model “translates” an image into a sentence and a visualization showed how the unseen states of different hierarchical LSTMs evolve over time. In research [16], CNN is image decoder which extracted features from image. A RNM is caption decoder which generated output caption according to features. However, it represented the co-attention that distinguishes intermodal relations while ignoring the self-attention that distinguishes intramodal interactions. Multimodal Transformer Model, which is evolved from Transformer Model, overcomes it. In this way, it captured both inter- and intra-model interaction in a single attention block. It can make difficult decisions and generate captions. Furthermore, the multiview visual feature introduced in the MT model can improve its performance.

3 Literature Table Title

Journal name

Year

Data set

Algorithm

Image captioning—a deep learning approach [2]

International Journal of Applied Engineering Research Open Access

2018

Flickr8k, Flickr30k

CNN, LSTM – Accuracy-68% – Limited dataset is used – Accuracy can be improved

Limitations

(continued)

Image Captioning Using Deep Learning Model

231

(continued) Title

Year

Data set

Algorithm

Deep cap: A deep IEEE learning model to caption black and white images [4]

Journal name

2020

Flickr8k

CNN, LSTM – Accuracy-45.77% – Accuracy can improve in future using alternate method

Limitations

Automatic image IEEE Access and video caption generation with deep learning: A concise review and algorithmic overlap [1]

2020

Flickr8k,9k, 30k and MSCOCO

CNN, RNN/LSTM

– Shown concise review – Any type of implementation is not done

– Large-scale dataset for better performance – For text they have collected data from different sources like Pinterest for images and Wikipedia for text – Can improve performance of their model

Multitask learning for cross-Domain image captioning [5]

IEEE 2019 Transactions on Multimedia

Flickr30k, Oxford- 102, MS-COCO

CNN, CGAN, RNWLSTM

Topic-oriented image captioning Based on order-embedding [6]

IEEE Transactions on Image Processing

2019

Flickr30k, MS-COCO

CNN, RNN/LSTM

Retrieval topic recurrent memory network for RSI captioning [7]

IEEE Journal of Selected Topics in Applied Earth

2020

UCMcaptions, RSICD

Memory Networks, RTRMN

– Some images have meaningless action so they try to solve that error

More is better: precise and detailed image captioning using online positive recall and missing concepts mining [8]

IEEE Transactions on Image Processing

2019

MS-COCO, MS-COCO Online Test Server

FCN, MIL

– Need to extract relation between concepts to improve performance

Multilayer dense IEEE Access Open Access attention model for image caption [3]

2019

Chinese AI’ Flickr8k, 30k, MSCOCO

Faster R-CNN ,

– Accuracy-68% – Have to implement on larger dataset

LSTM

(continued)

232

D. Patel et al.

(continued) Journal name

Year

Data set

Algorithm

Context-driven IEEE Access image caption with global semantic relations of the named entities [9]

2020

Good news, Breaking news

CNN, LSTM – Few outcomes have average score compare to others

Self-guiding multimodal LSTM—When we do not have a perfect training dataset for image captioning [10]

IEEE Transactions on Image Processing

2019

Flickr NYC

m LSTM, sg LSTM

– Test this model on another dataset instead the one which they have used

Toward remote sensing image retrieval under a deep image captioning perspective [11]

IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing

2020

UAV image captioning dataset, RSICD dataset

CNN, RNNLSTM

– Improve captioning block for better performance

Multi-level policy IEEE 2020 and reward-based Transactions deep on Multimedia reinforcement learning framework for image captioning [12]

Flickr30k, MS-COCO

CNN, Multi-layer LSTM

– Implement multi-agent algorithm to train the model

Caption net: Automatic endto-end siamese difference captioning model with attention [13]

Title

Limitations

IEEE Access

2019

Spot-the-diff baseline

Deep CNN, LSTM

– More accurate features is need to obtain – Accuracy is not so good

VAA: Visual IEEE Access aligning attention model for remote sensing image captioning [14]

2019

UCMcaption, SydneyCaptions

VAA Model

– Accuracy-81% – Can improve VAA model for complicated images

Flickr8K, 30k, MSCOCO

LSTM

– More dataset can take to test model’s performance

Deep hierarchical IEEE 2020 encoder- decoder Transactions network for on Multimedia image captioning [15]

(continued)

Image Captioning Using Deep Learning Model

233

(continued) Title

Journal name

Year

Data set

Algorithm

Limitations

Multimodal transformer with multi-view visual representation for image captioning [16]

IEEE Transactions on Circuits and Systems for Video Technology

2020

MS-COCO 2015 dataset

Multimodal transfer

– Can go for large dataset to test the model

4 Proposed work There are several methods and models that are created to caption the image. Yet captioning image is the complex task. Some of the researchers have implemented different methods to caption color image while 1% of researchers have captioned gray-scale image. Overall, proper accuracy has not been attained due to several factors such as models, training, inputs, and outputs, etc. So, in this proposed work, a new approach that can caption both color and gray-scale images has been proposed. Additionally, this work can be further utilized in social media platforms for captioning purpose. In future, one possible outcome can be a dataset named Flicker30k is used to train model now, however, later a larger dataset can be opted. The Flickr30k dataset includes 31,000 photos from Flickr, as well as 5 reference sentences contributed by human annotators. The similar dataset is also used for gray-scale images, by converting the current dataset’s images into gray-scale. For feature extraction, pretrained model i.e., Inception V3, VGG16 or VGG19 are used. They are highly trained models with thousands of images. They are basically used to identify any object in the image and are highly accurate. Firstly, the features are obtained from the image datasets of the colored one. While, for the gray-scale captioning, current dataset will be converted into gray-scale, so that, model will easily understand how to caption the gray-scale images as they only have 1-channel. The images are then separated and fed to the pre-trained (Inception) model accordingly. If there are color images which is of 3-channel, it gives in RGB channel form to the model. However, Inception Model takes input of 3-channels at a time, hence 1-channel of gray-scale was stacked 3 times to it. Later, the extracted features are used as input to LSTM, which generates the caption of an image. From the above procedure, the good performance of model is gained, where it can identify both types of images. With the help of proper methods, training more data and epochs and use of proper technique in the model, the performance of image captioning can be improved more than other methods.

234

D. Patel et al.

5 Observation and Result Many approaches and algorithms have been presented and used to increase the accuracy of captioning the picture during the research of various studies. Some attempted to improve accuracy, while others introduced a new technique for a variety of reasons. Making a new set of data is one of these goals. Further, the Attention Mechanism is applied in a few articles [3, 14, 16]. With this strategy, their major objective was to increase the model’s accuracy for captioning. On the other hand, other researchers utilized the Encoder-Decoder Framework [15], which included building a new network that was accessible to distinct encoders and decoders. The research [10] that caught our attention aimed to create a new dataset from the Flicker dataset called “FlickerNYC.” The researchers [10] created two approaches (sgLSTM and mLSTM) for improving caption descriptions in the existing dataset. However, no method has previously been proposed that can create both color and gray-scale images.

6 Conclusion This paper explains what image captioning is and its benefits. A different approach to caption the colorized images and gray-scale images has been framed and tested it with different datasets and pre-trained models. At last, model’s accuracy has been determined to find out that this model is the most suitable model to caption both types of images. However, there are few challenges to overcome. Firstly, the irrelevant use of words in caption, i.e., captioning which is not present in a scene, is the main problem. Secondly, evaluation is the problem where automatic testing is not as good as human testing. Computer vision systems will become more reliable as automatic picture captioning and scene interpretation improve, making them more useful as personal assistants for visually impaired persons and in enhancing their day-to-day lives. From the above survey and study, the benefits as well as challenges to overcome in field of image processing are evident.

References 1. S. Amirian, K. Rasheed, T. Taha, H. Arabnia, Automatic image and video caption generation with deep learning: A concise review and algorithmic overlap. IEEE Access 8, 218386–218400 (2020) 2. D.S. Lakshminarasimhan Srinivasan, A.L. Amutha, Image captioning—A deep learning approach. Int. J. Appl. Eng. Res. Open Access 3. K. Wang, X. Zhang, F. Wang, T. Wu, C. Chen, Multilayer dense attention model for image caption. IEEE Access 7, 66358–66368 (2019)

Image Captioning Using Deep Learning Model

235

4. V. Pandit, R. Gulati, C. Singla, S. Singh, DeepCap: A deep learning model to caption black and white images, in 2020 10th International Conference on Cloud Computing, Data Science & Engineering (Confluence) 5. M. Yang, W. Zhao, W. Xu, Y. Feng, Z. Zhao, X. Chen, K. Lei, Multitask learning for crossdomain image captioning. IEEE Trans. Multimedia 21(4), 1047–1061 (2019) 6. N. Yu, X. Hu, B. Song, J. Yang, J. Zhang, Topic-oriented image captioning based on orderembedding. IEEE Trans. Image Process. 28(6), 2743–2754 (2019) 7. B. Wang, X. Zheng, B. Qu, X. Lu, Retrieval topic recurrent memory network for remote sensing image captioning. IEEE J. Sel. Topics Appl. Earth Observations Remote Sens. 13, 256–270 (2020) 8. M. Zhang, Y. Yang, H. Zhang, Y. Ji, H. Shen, T. Chua, More is better: Precise and detailed image captioning using online positive recall and missing concepts mining. IEEE Trans. Image Process. 28(1), 32–44 (2019) 9. Y. Jing, X. Zhiwei, G. Guanglai, Context-driven image caption with global semantic relations of the named entities. IEEE Access 8, 143584–143594 (2020) 10. Y. Xian, Y. Tian, Self-guiding multimodal LSTM—when we do not have a perfect training dataset for image captioning. IEEE Trans. Image Process. 28(11), 5241–5252 (2019) 11. G. Hoxha, F. Melgani, B. Demir, Toward remote sensing image retrieval under a deep image captioning perspective. IEEE J. Sel. Top. Appl. Earth Observations Remote Sens. 13, 4462– 4475 (2020) 12. N. Xu, H. Zhang, A. Liu, W. Nie, Y. Su, J. Nie, Y. Zhang, Multi-level policy and reward-based deep reinforcement learning framework for image captioning. IEEE Trans. Multimedia 22(5), 1372–1383 (2020) 13. L. Yang, H. Wang, P. Tang, Q. Li, CaptionNet: A tailor-made recurrent neural network for generating image descriptions. IEEE Trans. Multimedia 23, 835–845 (2021) 14. Z. Zhang, W. Zhang, W. Diao, M. Yan, X. Gao, X. Sun, VAA: Visual aligning attention model for remote sensing image captioning. IEEE Access 7, 137355–137364 (2019) 15. X. Xiao, L. Wang, K. Ding, S. Xiang, C. Pan, Deep hierarchical encoder–decoder network for image captioning. IEEE Trans. Multimedia 21(11), 2942–2956 (2019) 16. J. Yu, J. Li, Z. Yu, Q. Huang, Multimodal transformer with multi-view visual representation for image captioning. IEEE Trans. Circuits Syst. Video Technol. 30(12), 4467–4480 (2020)

Recommender System Using Knowledge Graph and Ontology: A Survey Warisahmed Bunglawala, Jaimeel Shah, and Darshna Parmar

Abstract In recent years, Users find it challenging to choose what interests them among the many options offered due to the abundance of information. And finding those choices from data itself has also becomes very challenging task for organizations. To handle this task, recommendation system is an important field of research in computer science. Despite several efforts to make RS more efficient and personalized, it still faces issues like cold start, data sparsity, etc. And as it is designed to be readable by humans only, computer cannot process nor can interpret the data in it. Ontology facilitates the knowledge sharing, reuse, communication, collaboration, and construction of knowledge rich and intensive systems. Adding semantically empowered techniques to recommender systems can significantly improve the overall quality of recommendations. There has been a lot of interest in creating recommendations using knowledge graphs as a side information. By this, we not only overcome the issues of traditional RS but also provide a flexible structure, which naturally allows integration of multiple entities all together. It is also helpful in explanation for recommended items. So, in this survey we collected recently published research papers on this particular field to enhance RS. We provided a fine-grained information on this topic along with the explanation on how to use ontology approach for building a KG and challenges faced by both RS and KG systems. Certain crucial datasets and tools are also offered for a better understanding of accessibility. Keywords Recommender system · Knowledge graph · Ontology · Top-down approach · Bottom-up approach · Challenges in KG · Challenges in RS

W. Bunglawala (B) · J. Shah Parul Institute of Engineering and Technology, Vadodara, India e-mail: [email protected] J. Shah e-mail: [email protected] D. Parmar Prof., Parul Institute of Engineering and Technology, Vadodara, India e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 I. J. Jacob et al. (eds.), Expert Clouds and Applications, Lecture Notes in Networks and Systems 444, https://doi.org/10.1007/978-981-19-2500-9_17

237

238

W. Bunglawala et al.

1 Introduction The rapid progress of digital technology has resulted in a massive increase in data. Data are generated in large quantities on social media platforms like Twitter, Instagram, and Facebook. Furthermore, research and publications are also increasing day by day [1]. With this advancement of the data, it brings advantages and disadvantages both. Advantage can be described as we have a lot of data available within seconds. Disadvantage can be stated as abundance of data has increased and due to that we are unable to get most relevant and required information. To overcome this disadvantage, recommendation system has been developed and still in research trend. A recommendation system is a type of information filtering system that attempts to predict a user’s “rating” or “preference” for an item. The task of a recommendation system can be divided into two parts: (I) estimating a value of prediction for an item and (ii) recommending users about items [1]. To accomplish this objective, there are a variety of ways available, the most frequent or popular of which being Content-based Recommendation Systems and Collaborative Recommendation Systems. Hybrid Recommendation System is also developed by merging those techniques [1–3]. Recommendation systems require constant change because of the exponential growth of data and knowledge. In recent years, introducing a knowledge graph as a side information in recommendation system has attracted a lot of researchers and organizations [4, 5]. First, Knowledge graph is introduced by goggle in May 2012 [6]. There are lot of definitions on knowledge graph is available according to its usage. For ease of comprehension, a knowledge graph is a heterogeneous graph in which the nodes represent entities and the edges reflect relationships between them [2]. Knowledge graph has the advantage of having a flexible basic graph structure and providing a model of how everything is connected. In this way, knowledge graph is of great use in RS as side information and to help with explanation and integration of large data. Ontologies are the foundation of a knowledge graph’s formal semantics. They can be thought of as the graph’s data schema. They serve as a formal contract between the knowledge graph’s creators and its users over the meaning of the data included inside it. A user could be another person or a program that needs to interpret facts in a reliable and exact manner. Ontologies ensure that data and its meanings are understood by everyone [7]. So, we can say that ontology represents structure or schema and it has power to rules and reasoning, while knowledge graph captures the data. Motivation behind this survey is to state the latest research in the area of recommender system as it is very important field of research now a days and in upcoming future. And to handle the abundance of data that are generating nowadays we need some common and more scalable structure to handle it, so concept of knowledge graph and ontology is considered. That’s why study not only include basic information regarding system but also gives an overview on basic steps to follow for building the system. And study also include information available on datasets and tools that can be used. Also, this type of system cannot only be used for ecommerce. But it

Recommender System Using Knowledge Graph …

239

can be majorly beneficial to the situation like COVID-19 where we want related data available within seconds for data discovery or in case of emergency. So, we reviewed some research and articles combining these technologies.

2 Background and Related Work 2.1 Concept of Knowledge Graph Although some people have attempted to establish a formal definition of knowledge graph, none of them can be said the standard definition. The phrase “Knowledge Graph” can be interpreted in a variety of ways. As an alternative to the definition, the following features of a knowledge graph can be presented [8]: • It primarily describes real-world things and their interrelationships in the form of a graph. • In a schema, defines the classes and characteristics of entities. • Allows for the possible interconnection of arbitrary things. • Covers a wide range of topics. As shown in the figure below, entity is a thing present in a real world while concept is something define as a collection of individuals which have a same characteristic. Literal can be defined as a nothing but a specific value or strings of some relations. And the edge between entity and concepts can be define as a relation. For example, Yao Ming is individual entity and Basketball is a concept as so many players out there play basketball such as Kobe Bryant and Stephan Curry. While Yao Ming height can be define as a “2.29 m”, so this specific value can be said literal and on the other hand, Yao Ming has a wife Ye Li so wife is a relation between those two entities (Fig. 1). Fig. 1 Knowledge graph [8]

240

W. Bunglawala et al.

It’s worth noting that there are two sorts of knowledge in KG: schematic knowledge and factual knowledge [8]. Statements concerning concepts and qualities, such as (Asian Country, subClassOf(), Country), make up schematic knowledge. While factual information is made up of claims regarding specific situations, the triples in the graph above are all factual knowledge. The majority of the KG has a great quantity of factual information and a minor amount of schematic knowledge. The logical foundation of knowledge graphs is based on ontology languages such as Resource Description Framework (RDF) and Ontology Web Language (OWL), which are W3C-recommended creations (World Wide Web Consortium). RDF can be used to represent rich and complicated knowledge about entities, properties, and relationships, while owl can represent both schematic and factual knowledge. So, using ontology as a basic logical foundation, a knowledge graph can be formed in one of two ways: bottom-up or top-down.

2.2 Literature Work As we have reviewed latest work related to recommender system using Knowledge graph, and ontology, various researchers have used various model and techniques. Some of the reviewed research and articles are mentioned below. In paper [4], authors have mentioned MKR model for KG enhanced RS and within a short period of time one more research [9] mentioned extended MKR model. the system flow mentioned in the paper is, feature extraction—in this general feature are extracted using MLP layer and Text feature extraction is based on the Text CNN. After the feature extraction, recommendation module is there in that they have taken u and v where u is users and v is items as an input for this layer and predicted the probability of the user u engaging item v. After this knowledge graph embedding layer is there as a side information. And then cross-compression unit is mentioned, in which two parts are there, one is crossing part and another is compression part. Through cross-compression units, SI-MKR can adaptively adjust the weights of knowledge transfer and learn the relevance between the two. In one research [10], authors applied a hierarchical design based on heterogeneous input features to recommendation systems to learn text features, behavior features, graph structured features, and spatio-temporal features from massive data. They introduced the classification model design of recommendation systems, which is divided into three layers: feature input, feature learning, and output layers. And they also mentioned the evaluation indexes like mean absolute error (MAE) and root mean square error (RMSE) for evaluating the RS. Experimental comparison and future development direction of recommendation systems are also mentioned. Paper [6] written by authors Jiangzhou Liu, Li Duan presents the basic knowledge of recommendation system and knowledge graph. After that they mentioned the key methods that used in the recommendation system with KG which includes path-based method, embedding-based method and hybrid method further more they mentioned the user interest model, after that they have provided some basic future

Recommender System Using Knowledge Graph …

241

directions which include combination with graph neural network, enhanced representation of KG, KG completion and corrections. With same title paper [5] written by authors Qingyu Guo and others categorized recommendation methods based on knowledge graph into three categories: embedding based, connection-based methods and propagation-based methods and mentioned pros and cons on algorithm used in these methods. Useful dataset is mentioned and categorized into different categories such as movies and books. One of the detailed papers [11] mentioned KG based RS filtering approaches into categories like ontology based, linked open database, embedding based and path based. In result, they classified into two categories such as KG and Semantic Web and second is KG and AI methods. In first category, top six approaches mentioned some of them are KG and linked data, KGE and Ontologies, and in second category, they compared different filtering approach and mentioned that hybrid system is most common. Future direction mentioned by the authors are interpretability of RS, explainable recommendation and KG based dynamic RS. In other paper [12], authors mentioned that there are two types of approach, bottom-up and top-down approach. Paper describes bottom-up approach for knowledge graph creation. In that, they first shown the architecture with different layers like knowledge extraction, knowledge fusion, storage of knowledge graph, and retrieval of KG have been described in details with their methods and tools available for it. In one paper [13] authors created a scenic spot knowledge graph based on ontology, paper defines the concept of ontology on why and how should we use ontology so that the purpose can be served is greatly explained in this paper. Also, they present architecture that includes steps like data gathering and ontology building, entity alignment, and knowledge graph storage tool. They used neo4j for storage and mentioned that it is one of the great databases that stores structured data in the form of network. For the evaluation purpose they also describe precision and recall matrices. There model outperform the string similarity method. In paper [2], authors mentioned that graph database is more efficient and expressive so they used a property graph. In that, they represent a multi-layer graph model and constructed a knowledge graph and returned the various top end N recommendation. So, they mentioned five-layer model in which layer one is for users and details, layer two is for needs, layer three mentioned features and related details while the layer four comprised of all nodes related to various items specifications and its associated details. The layer five comprised of all nodes related to various items and its associated details. The construction of layer two, three, and four can be carried out based of preoccupied knowledge. In the process, a system model is defined as a combination of different recommendation techniques hence can be called hybrid model RS, so that more efficient top-N recommendation can be done. In survey paper [14] along with review, some great future directions are mentioned such as bringing in more side information into knowledge graph so that power can be enhanced, also connecting social networks to know how social influence affect the recommendation, explainable recommendation, and GCN are also in trend. For the purpose of explainable reasoning over KG for RS, one of the papers [15] has mentioned new model named KPRN-knowledge aware path recurrent network.

242

W. Bunglawala et al.

Model allows effective reasoning over path to infer underlying rationale of user-item interaction. Also, they designed new weighted pooling operation to discriminate the strength of different paths in connecting users with an item. Datasets used in this paper relate to music and movie. They also used LSTM to capture the sequential dependencies. In paper [16], authors describe the combination of ontology and collaborative filtering for mooc’s recommender system. They mentioned basic components of personalized system as (1) techniques, (2) item, and (3) personalization. Proposed method includes hybrid method as mentioned above and for computing similarity and mooc’s similarity extended cosine similarity used and for learners’ similarity PCC is used. At last, algorithm for generating recommendation is mentioned. We also studied papers other than above, and some of that focuses on solving problems using KG related to COVID-19 pandemic that worth mentioning. They are listed below. • Cone-KG: A Semantic Knowledge Graph with News Content and Social Context for Studying COVID-19 News Articles on social media [17]. • Open Research Knowledge Graph: Next Generation Infrastructure for Semantic Scholarly Knowledge [18]. • COVID-19 Knowledge Graph: Accelerating Information Retrieval and Discovery for Scientific Literature [1].

3 Observation and Discussion During survey, we have read many recently published papers and articles, and we find some very useful details which we present in this section as simply as possible. We explain some basic terms and challenges that one need to understand for the work. And some of the useful dataset and tools are also mentioned to get started. Most of the papers mentioned traditional algorithm as content based, collaborative and hybrid algorithm and then some personalized algorithm also mentioned such as demographic based, community based, and knowledge-based algorithm. In knowledge-based algorithm, some papers mentioned techniques or approach to achieve the recommendation using knowledge graph that are divided into four categories. For better understanding, see Fig. 2. Among these categories, ontology-based approach is popular due to the fact that it facilitates knowledge sharing, reusing, and highly rich knowledge with semantics.

3.1 Ontology-Based While creating ontology-based knowledge graph two types of approach are there, top-down approach and bottom-up approach. None of these two methods is better than each other it’s depends on the view of the developer. It may be easier for the

Recommender System Using Knowledge Graph …

243

Fig. 2 Recommendation techniques

developer to follow the top-down method if they have a more systematic top-down perspective of the domain. Or if they have better understanding at data level, they might follow bottom-up approach. The combine technique, on the other hand, is easier for most ontology developers since it leverages the notion “in the middle,” which is a more descriptive concept in domain ontology. For simplicity, we represent bottom-up architecture for KG using ontology just to understand the terms in overall process, as steps can be altered on base of approach developer use. To build an ontology, there are some useful software available which helps us in construction as well as in visualization of ontology, those are protégé, NeOn Toolkit, SWOOP, Neologism, and Vitro [19]. In bottom-up approach, from Linked Open Data (LOD) or other knowledge resources, we extract knowledge instances. After knowledge fusing, to generate the entire KGs, the top-level ontologism is built using knowledge instances [12]. Bottom-up approach of KG is and iterative update process, which includes knowledge acquisition, knowledge fusion, knowledge storage, and retrieval. For better understanding, let us look at the following architecture of bottom-up approach. Structured data, unstructured data, and semi-structured data are the three basic sources of knowledge acquisition, as shown in Fig. 3. Attribute, relation, and entity extraction are all types of knowledge extraction. Following that, knowledge fusion can be defined as an iterative process in which we build the ontology and regularly review it for higher quality. NoSQL databases are more popular for storing and retrieving knowledge graphs. In knowledge extraction, extracted knowledge is usually presented in machine readable formats such as RDF and JsonLD. There are many tools available for knowledge extraction depending on the needs and functions, some of them are Stanford NER, OpenNLP, AIDA, Open Calais, and Wikimeta. While we can extract knowledge from any sources such a website or any record and datasets available, nowadays most of the instances extracted from DBpedia or Yago, and Wikipedia. For

244

W. Bunglawala et al.

Fig. 3 Bottom-up approach for KG using ontology [12]

semi-structured and unstructured data sources, we need entity extraction, relation extraction, and attribute extraction. • Entity extraction is the process of identifying an entity from a large amount of data and categorizing it into predetermined categories such as person, place, or location. • Relationships among those entities are analyzed after entity extraction to conceptually extract relations. • Attribute extraction is to define the intentional semantics of the entity and it is important for defining the concepts of entity more clearly. The purpose of knowledge fusion is to achieve entity alignment and ontology creation. And it is an iterative process. Entity matching is another name for entity alignment. The goal of entity matching is to determine whether or not various entities refer to same real-world object. It’s worth noting that entity alignment typically relies on external sources like manually created corpora or Wikipedia links. After that, ontology construction and evaluation step is there, in that we create the ontology and constantly evaluate it for better performance of the application. To ensure, the KG’s quality, general ontologies such as FOAF and general meta-data from schema.org are required. In terms of KG storage, it is often saved in a NoSQL database. There are two basic storage types: RDF-based storage and graph database storage. The benefit of using RDF is that it improves the efficiency of querying and merge-join for triple patterns. Better query results on the other hand, come at a high expense in terms of storage capacity. Some popular RDF-based data storage is 4store, RDF Store, TripleT and so on, most of the native storage system provides SPARQL or similar like query languages. Graph-based storage, on the other hand, has the advantage of providing excellent graph query languages and supporting a wide

Recommender System Using Knowledge Graph …

245

range of graph mining methods. They do have drawbacks such as slow knowledge updates, expensive maintenance costs, and distributed knowledge inconsistency. SPARQL is a popular query language for retrieving information, and practically, every large-scale knowledge graph system has a SPARQL query endpoint. SPARQL generates output in JSON, JSON-LD, RDF/XML, XML, CSV, RDF/N3, and many more formats, with practically all of them being machine readable. Machine-readable forms necessitate visualization tools. The most popular of which are browserbased visualizations because some query returns formats are text-based. IsaViz, RDF Gravity, DBpedia Mobil, and Open Link Data Explorer are some of the most popular tools. Because ontology is based on description logic, knowledge retrieval is primarily based on logic principles.

3.2 Challenges in Recommender System The most common issues or challenges associated with development of traditional recommendation system is cold start and sparse data [2, 20]. Cold-start problem can occur when system is unable to inference any information regarding users or item. This can happen when new user or the new item is added in catalog. In this situation, we cannot predict the new user taste as no information is available. Furthermore, due to insufficient and erroneous findings, users are unable to rate or purchase the item. So, to avoid the cold-start problem, numerous methods are suggested, including (a) asking users to rate some items at the start, (b) asking users to indicate their liking, and (c) recommending items based on user demographic data. There may be sleepers in some circumstances; sleepers are items that are nice but not rated. We can manage this by employing meta-data or content-based solutions, such as item entropy, item personalization, or using Linked Open Data (LOD), which eliminates the need for consumers to supply explicit input. Data sparsity can be understood as let’s say we formed a cluster of similar data and we will recommend the product based on those cluster. Now as more and more variables included in dataset or we can say with huge amount of data, noise, and uncertain data are also increased. In this situation, data will be more uniform and we will struggle and we won’t be able to do anything with those data. To overcome this issue many techniques like, multidimensional models, SVD techniques, and demographic filtering can be used.

3.3 Challenges in Knowledge Graph In knowledge graph, most common challenges we can list out is knowledge completion, harmonization of datasets, and knowledge alignment [8, 21]. Knowledge Completion: Incompleteness of knowledge graph is when there is a dashed line available in the graph or we can say that there appears to be a

246

W. Bunglawala et al.

possible relationship between two entities from the solid facts. Completing a KG is a very challenging task for researchers and that is why knowledge embedding is an active research area in this field. The symbolic compositionality of KG relations is ignored by embedding-based techniques, which limits their application in increasingly complicated reasoning tasks. And to overcome this, several other approaches like multi-hop paths are developing [22]. So basically, due to the large amount of data and relations we can say that fraction of incompleteness in KG will always be there. Harmonization Datasets: The ability to harmonize or integrate data from many sources is crucial to the creation of semantic knowledge graphs. However, different authors use different names to describe the same subject, which is a typical problem. As a result, it is possible to confuse one entity with another. We must use ontologies to tackle this difficulty since they give much more than just data harmonization. One of the functions of ontology is to provide a standard model of knowledge associated with a specific domain, as well as a common identifier that can be used to link it to other things in other ontologies. Knowledge Alignment: Knowledge graphs have become more widely available on the Web in recent decades, but their heterogeneity and multilingualism continue to limit their sharing and reuse on the Semantic Web [2]. Basically, knowledge alignment is nothing but to discover the mapping (i.e., equivalent entities, relationships and others) between two KGs. Embedding and reasoning both can be used for these types of challenge but hybrid reasoning promises more encouraging result [2].

3.4 Available Tools and Dataset As a data storage, lots of database available to store knowledge graph and graph data. And most of the NoSQL databases are used to store the KG. Some are listed below [23–25] (Table 1). Also, there are many general as well as domain specific datasets available. Some of the popular general datasets are mentioned below along with few COVID-19 datasets [23, 24] (Table 2).

4 Proposed Idea Talking about the latest pandemic COVID-19 has claimed so many lives worldwide. And it increases the need for tools that enable researchers to search vast scientific corpora to find specific information, visualize connections across the data, and discover related information in the data. Several dedicated search engines have been built due to the need of information retrieval related to scientific literature on COVID19. Search engines like Sketch Engine COVID-19, Sinequa COVID-19 Intelligent Search, Microsoft’s CORD19 Search and Amazon’s CORD19. However, this search

Recommender System Using Knowledge Graph …

247

Table 1 List of database Database name

Link

Database model

Neo4j

https://neo4j.com/

Graph

GraphDB

https://graphdb.ontotext.com/

multi model

CosmosDB:Azure https://docs.microsoft.com/en-us/azure/cosmos-db/introd Multi model uction OrientDB

https://orientdb.org/

Multi model

ArangoDB

https://www.arangodb.com/

Multi model

Janus Graph

https://janusgraph.org/

Graph

Virtuoso

https://virtuoso.openlinksw.com/

Multi model

Amazon Neptune

https://aws.amazon.com/neptune/

Multi model

Stardog

https://www.stardog.com/

Multi model

Dgraph

https://dgraph.io/

Graph

Table 2 List of datasets Dataset name

Description

Kaggle: Cord-19 [26]

It has approximately 500,000 scholarly publications on COVID-19, SARS-CoV-2, and related to coronaviruses, with over 200,000 of them having full text

Coronavirus (COVID-19) tweets dataset [27]

CSV files with the IDs and sentiment ratings of tweets about the COVID-19 pandemic are included in the collection. In real time, the Twitter stream is being monitored

AYLIEN: COVID-19 [28]

Corona virus news datasets

WordNet [29]

Princeton University offers a free extensive lexical database of English [8]

YAGO [30]

Wikipedia, WordNet, and GeoNames have all contributed to this massive semantic knowledge base

DBPedia [31]

DBpedia is a community-driven effort to extract structured content from the resources of different Wikimedia projects

Wikidata [32]

It’s a free, multilingual dataset that collects structured data to help Wikimedia Commons and Wikipedia

Google KG [3]

There are millions of items in Google’s Knowledge Graph that describe real-world entities

248

W. Bunglawala et al.

engines return thousands of search result that overlooked the inherent relationships like citation and subject topics [11]. Also, they do not provide the tool to visualize relationships, which can be beneficial for knowledge discovery. So, we need the system that can be specifically used for knowledge discovery and information retrieval. Also, we need to use every unique data we can gather such as scientific data and social media data. To build this kind of system, proposed flow diagram is mentioned below. After getting enough information about both techniques, data gathering and data extraction depends on the system so in this case if we want to build a system for COVID-19 situation some useful available datasets are mentioned in previous section along with the details on building knowledge graph using ontology (Fig. 4).

5 Conclusion and Future Work With a lot of data available, we need a fine-grained recommendation system that can help us to discover knowledge as efficiently as possible. In situation like COVID-19, it can be very helpful to discover new knowledge and knowledge retrieval. Also, to enrich the recommendation system with knowledge we need a common structure. Graph structure like knowledge graph can handle different types of data easily and efficiently. Therefore, proposed system is using KG for RS and basic approach is mentioned to achieve this task based on ontology with some techniques. Also, some popular databases and datasets that are available is mentioned along with some COVID-19 datasets that can be helpful in the situation of pandemic. For future work, we can construct a generalized KG using these techniques, which can be used for COVID-19 scientific literature and/or social media recommender system to help in situation of pandemic. Also, this paper can also be used as a reference for creating similar applications with different goals and datasets based on ontology.

Recommender System Using Knowledge Graph …

249

Fig. 4 Proposed flow diagram

References 1. C. Wise, V.N. Ioannidis, M.R. Calvo, X. Song, G. Price, N. Kulkarni, G. Karypis, COVID19 knowledge graph: accelerating information retrieval and discovery for scientific literature. arXiv preprint arXiv:2007.12731 (2020) 2. A.A. Patel, J.N. Dharwa, An integrated hybrid recommendation model using graph database, in 2016 International Conference on ICT in Business Industry & Government (ICTBIG) (IEEE, 2016)

250

W. Bunglawala et al.

3. Bluepi, Classifying different types of recommender systems, https://www.bluepiit.com/blog/ classifying-recommender-systems/ 4. H. Wang, F. Zhang, M. Zhao, W. Li, X. Xie, M. Guo (2019) Multi-task feature learning for knowledge graph enhanced recommendation, in The World Wide Web Conference, pp. 2000– 2010 5. Q. Guo et al., A survey on knowledge graph-based recommender systems. IEEE Trans. Knowl. Data Eng. (2020) 6. J. Liu, L. Duan, A survey on knowledge graph-based recommender systems, in 2021 IEEE 5th Advanced Information Technology, Electronic and Automation Control Conference (IAEAC), vol. 5 (IEEE, 2021), pp. 2450–2453 7. Ontotext, What is knowledge graph, https://www.ontotext.com/knowledgehub/fundamentals/ what-is-a-knowledge-graph/ 8. W. Li, G. Qi, Q. Ji, Hybrid reasoning in knowledge graphs: Combing symbolic reasoning and statistical reasoning. Semant. Web 11(1), 53–62 (2020) 9. Y. Wang, L. Dong, H. Zhang, X. Ma, Y. Li, M. Sun, An enhanced multi-modal recommendation based on alternate training with knowledge graph representation. IEEE Access 8, 213012– 213026 (2020) 10. H. Wang, Z. Le, X. Gong, Recommendation system based on heterogeneous feature: A survey. IEEE Access 8, 170779–170793 (2020) 11. J. Chicaiza, P. Valdiviezo-Diaz, A comprehensive survey of knowledge graph-based recommender systems: Technologies, development, and contributions. Information 12(6), 232 (2021) 12. Z. Zhao, S.-K. Han, I.-M. So, Architecture of knowledge graph construction techniques. Int. J. Pure Appl. Math. 118(19), 1869–1883 (2018) 13. W. Zeng, H. Liu, Y. Feng, Construction of scenic spot knowledge graph based on ontology, in 2019 18th International Symposium on Distributed Computing and Applications for Business Engineering and Science (DCABES) (IEEE, 2019) 14. P.S. Sajisha, V.S. Anoop, K.A. Ansal, Knowledge graph-based recommendation systems: The State-of-the-art and some future directions 15. X. Wang, D. Wang, C. Xu, X. He, Y. Cao, T.S. Chua, Explainable reasoning over knowledge graphs for recommendation, in Proceedings of the AAAI Conference on Artificial Intelligence, vol. 33, No. 01 (2019), pp. 5329–5336 16. K. Rabahallah, L. Mahdaoui, F. Azouaou, MOOCs Recommender system using ontology and memory-based collaborative filtering. ICEIS (1) (2018) 17. F. Al-Obeidat, O. Adedugbe, A.B. Hani, E. Benkhelifa, M. Majdalawieh, Cone-KG: A semantic knowledge graph with news content and social context for studying covid-19 news articles on social media, in 2020 Seventh International Conference on Social Networks Analysis, Management and Security (SNAMS) (IEEE, 2020), pp. 1–7 18. M.Y. Jaradeh et al., Open research knowledge graph: next generation infrastructure for semantic scholarly knowledge, in Proceedings of the 10th International Conference on Knowledge Capture (2019) 19. W3C, Ontology editors. https://www.w3.org/wiki/Ontology_editors 20. S. Khusro, Z. Ali, I. Ullah, Recommender systems: issues, challenges, and research opportunities. Inf. Sci. Appl. (ICISA) 2016 (Springer, Singapore, 2016). pp. 1179–1189 21. SciBite, Addressing common challenges with knowledge graphs, https://www.scibite.com/ news/addressing-common-challenges-with-knowledge-graphs/ 22. X.V. Lin, R. Socher, C. Xiong, Multi-hop knowledge graph reasoning with reward shaping. arXiv preprint arXiv:1808.10568 (2018) 23. S. Ji, S. Pan, E. Cambria, P. Marttinen, S.Y. Philip, A survey on knowledge graphs: Representation, acquisition, and applications. IEEE Trans. Neural Netw. Learn. Syst. 24. GitHub, Totogo/awsome-knowledge-graph, awesome knowledge graph—github GitHub— totogo/awesome-knowledge-graph: A curated list of Knowledge Graph related learning materials, databases, tools and other resources

Recommender System Using Knowledge Graph …

251

25. C#corner, Most popular graph database, https://www.csharpcorner.com/article/most-populargraph-databases/ 26. Kaggle, COVID-19 Open research dataset challenge (CORD-19), https://www.kaggle.com/ allen-institute-for-ai/CORD-19-research-challenge 27. IEEE Data Port, Coronavirus (COVID-19) tweets dataset, https://ieee-dataport.org/open-acc ess/coronavirus-covid-19-tweets-dataset 28. Aylien, Free coronavirus news dataset—Updated, https://aylien.com/blog/free-coronavirusnews-dataset 29. Princeton University, WordNet—A lexical database for english, “What is WordNet?” https:// wordnet.princeton.edu/ 30. Yago, YAGO: A high-quality knowledge base, https://yago-knowledge.org/ 31. DBPedia, Global and unified access to knowledge graphs, https://www.dbpedia.org/ 32. Google Search Central, Google knowledge graph search API, https://developers.google.com/ knowledge-graph

A Review of the Multistage Algorithm Velia Nayelita Kurniadi, Vincenzo, Frandico Joan Nathanael, and Harco Leslie Hendric Spits Warnars

Abstract This paper reviews multistage algorithms with various concepts, and various forms of multistage algorithms are discussed after a detailed research. Previous research searches are conducted using Google Scholar and the software Publish or Perish, and carried out in three ways: the exploration of multistage algorithm publications, classification of multistage algorithm publications, and review of multistage algorithm publications. Based on these three steps, the literature review is limited to 26 publications, and the results show that the naming of the multistage algorithm is only limited to mentioning the number of stages in the proposed algorithm, and the abbreviation of its name overlaps with terms in other algorithms. For example, abbreviation such as MSA defined as Multi Stage Algorithm is expanded in other studies as Multiple Sequence Alignment. Keywords Multistage algorithm · Multiple sequence algorithm · Literature review

1 Introduction Applying a multistage algorithm helps minimize the effect of higher demand stage requirements between selecting test models, by determining the ideal mix between the models of two different data at each calculation phase in the algorithm. The V. N. Kurniadi · Vincenzo · F. J. Nathanael Computer Science Department, School of Computer Science, Bina Nusantara University, Jakarta 11480, Indonesia e-mail: [email protected] Vincenzo e-mail: [email protected] F. J. Nathanael e-mail: [email protected] H. L. H. S. Warnars (B) Computer Science Department, Graduate Program, Doctor of Computer Science, Bina Nusantara University, Jakarta 11480, Indonesia e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 I. J. Jacob et al. (eds.), Expert Clouds and Applications, Lecture Notes in Networks and Systems 444, https://doi.org/10.1007/978-981-19-2500-9_18

253

254

V. N. Kurniadi et al.

multistage algorithm is applied to improve the Park Chen Yu (PCY) algorithm by utilizing several progressive hash tables to reduce the number of competitor sets. PCY itself was created by three researchers named Park, Chen, and Yu, where this algorithm is used to find frequent itemset mining searches for data searches in large datasets. The payoff is that multistage takes many steps to find the continuity set. This paper proposes to know more about the multistage algorithm in many ways, while this multistage model is multiscale and stochastic. This paper explores the development and implementation of multistage algorithms, including their meaning, because there are many different implementations. Exploration will be carried out in three steps such as: 1. 2. 3.

Exploration of multistage algorithm publication, Classification of multistage algorithm publication, Review of multistage algorithm publication.

2 Exploration of Multistage Algorithm Publication This multistage algorithm exploration step uses two approaches: a search for previous paper publications using the title of the multistage algorithm that is searched on Google Scholar and a search using Publish or Perish software. Exploration started by using Google Scholar to find previous publications where the title of the publication contain a multistage algorithm sentence without using citations in all years of publication, and the results displayed were 582 and the search can be seen at the following link: https://scholar.google.com/scholar?hl=en& as_sdt=0%2C5&q=allintitle%3A+multistage+algorithm&btnG=. A similar search was performed using the Publish or Perish software and that showed 582 publications with 7165 citations such as 130.27 citations per year, 12.31 citations per paper, 2.59 authors per paper, h-index 38, and g-index 74. Figure 1 shows 582 publications from 1966 to 2021, including 12 publications that do not have year information, and due to limitations in presenting, the image only shows from 1987 to 2021. Figure 1 shows that 2005 papers had a maximum number of citation of 1170, while some from 1966, 1968, 1969, 1974, 1975, and 1980 papers do not have total citations. Based on the search for titles in the search results of 582 publications, it is found that most of the paper titles do not focus on multistage algorithms but instead use the term multistage in the algorithms proposed or discussed in these publications. Therefore, the search for previous publications was expanded by searching for titles containing multistage algorithm sentences using citations. Therefore, the search was continued by using Google Scholar to search for previous publications where the title of the publication contain a multistage algorithm sentence using citations for all years of publication. The results showed 52 results, and the search could be seen at the following link: https://scholar.google.com/scholar?hl=en&as_sdt=0%2C5&q=allint itle%3A+%22multistage+algorithm%22&btnG=. A similar search was performed using the Publish or Perish software and that showed 52 publications with a total

A Review of the Multistage Algorithm

255

Fig. 1 Graph of 582 results for searching the title of the publication containing a multistage algorithm sentence without using citations in all years of publication

of 1474 citations such as, 42.11 citations per year, 28.35 citations per paper, 2.62 authors per paper, h-index 16, and g-index 38. Figure 2 shows 52 publications from 1986 to 2021, including one publication that do not have year information, and 2009 papers had a maximum number of citation of 482, while some from 1986, 2008, 2012, and 2020 papers do not have total citations. However, after a rigorous search for titles in the search results of these 52 publications, it was found that most of the papers do not have multistage algorithm titles

Fig. 2 Graph of 52 results for searching the title of the publication containing a multistage algorithm sentence using citations in all years of publication

256

V. N. Kurniadi et al.

but instead only use the term multistage in the algorithms proposed or discussed in the publication. Therefore, several publications that did not match the title multistage algorithm were omitted, and in the end, the remaining 26 papers which had the publication title as the multistage sentence algorithm were found. Figure 3 is a graphic representation for Table 1 showing 26 publications from 1989 to 2021, and the 2002 paper had a maximum number of citation of 51, while the 2003 and 2009 papers had no total citations. The limitation of 26 papers using Fig. 3 Graph of 26 results for searching the title of the publication containing a multistage algorithm sentence using citations in all years of publication

Table 1 Table of 26 results for searching title of the publication containing a multistage algorithm sentence using citations in all years of publication

Year

Sum of cites

Number of papers

1989

31

1

1994

27

1

1995

7

2

1996

3

2

1997

32

1

2002

51

1

2003

0

1

2007

37

1

2009

0

1

2010

4

1

2011

13

2

2013

1

1

2014

3

2

2016

35

3

2017

37

4

2018

18

1

2021

1

1

Total

300

26

A Review of the Multistage Algorithm Table 2 Classification of 26 selected papers

Type of publications

257 Amount

International conference

9

International journal Q1

6

International journal Q2

3

International journal Q3

2

International journal Q4

1

International journal no Q

4

Doctoral dissertation Total

1 26

Publish or Perish software shows that the 26 publications, as seen in Table 1, have 300 citations such as, 9.38 citations per year, 11.11 citations per paper, 3.00 authors per paper, h-index 8, and g-index 17. These 26 publications are focused on, and the review and classification of these publications are mentioned in the following section.

3 Classification of Multistage Algorithm Publication At this step, a classification of the multistage algorithm implementation is carried out for the 26 papers based on the type of publication, such as international conferences, ranking of international journals using scimagojr journal ratings. The table and graph show the mapping of these papers. Moreover, the provisions for ranking scimagojr international journals use journal ratings in the year the publication. In this case, three publications have a long published year, but the journal was not indexed to the scimagojr rank when the publication was published, so the three publications are categorized as no Q international journals. Table 2 shows the classification results where Fig. 4 is the graphic representation for Table 2.

4 Review of Multistage Algorithm Publication In this step, the review for 26 papers starts from the previous publication. El-Shishiny first proposed the multistage algorithm in 1989 that was used for image classification implemented on a microcomputer. This algorithm has three steps in its application, namely parallelepiped classification, ellipsoidal separation and closest distance classification using the Mahalanobis algorithm. In its implementation, this multistage algorithm has the following pseudocode: 1. 2.

training example parallelepiped classification

258

V. N. Kurniadi et al.

Fig. 4 Classification of 26 selected papers

3. 4. 5. 6. 7. 8.

10 9 8 7 6 5 4 3 2 1 0

9 6 3

4 2

1

1

while (number of clusters whose pattern includes 1) { if the number of clusters whose pattern includes > 1 then do the ellipsoidal separation else if the number of clusters whose pattern includes 1 then do Mahalanobis distance classification }

Parallelepiped classification is done by conducting classifier training to determine the smallest and largest value for each feature in the cluster. At the same time, the classification of the Mahalanobis distance between point X and cluster i as a search for the nearest cluster was introduced by Swain in 1978. Meanwhile, ellipsoidal separation separates classes with ellipsoidal domains by calculating eigenvectors and estimating elliptical parameters [1]. Moreover, the multistage algorithm required in Chen’s research in his 1994 paper was used to achieve an improvement by applying the concept of equilibrium in which the aim was to measure the staging effect of the polyethylene fraction with supercritical ethylene and 1-hexene. This multistage algorithm is combined with the SAFT equation seen in the Supercritical AntiSolvent (SAS) fractionation which is formed from a polyethylene model with ethylene and 1-hexene, where this SAFT connects macroscopic partitions. SAS itself is a process that separates polydisperse mixtures of macromolecules with varying average molecular weights from which the macromolecules get a chemical microstructure. While implementing the SAS process, paying attention to temperature and pressure is necessary to control the selectivity and solvent capacity. Besides, the solvent composition must also be considered for the polymer solubility in weight percent. In addition, this multistage algorithm is used to predict the effect of theoretical stages to increase the selectivity of the solvent, i.e., significantly to reduce the size of the light, where smaller the light, the greater the sensitivity to staging [2].

A Review of the Multistage Algorithm

259

Furthermore, the term tomography refers to cross-sectional imaging of objects that can choose between transmitting and reflecting data, and most of its progress toward electromagnetic diffraction tomography was developed based on the Fourier diffraction projection theorem. Kartal, a researcher from the Technical University of Istanbul, Turkey, developed a multistage algorithm based on matrix partitioning, run in parallel to process tomographic diffraction information as image reconstruction. This model is run by reconstructing the partitioned algorithm into sub-matrixes, and a matrix inversion process carries out each sub-matrix, and its application is almost similar to the approach carried out in nonlinear neural networks. In addition, in its application, each stage has input and output vectors where at each output at each stage, a comparison is made between the desired output vector and the actual output vector, and the results are used as output vector predictions for the next stage [3]. This algorithm has several advantages, such as the advantages of parallel calculations and more resistance to numerical errors than single-stage algorithms. Multistage algorithms have advantages in terms of time efficiency, especially in parallel execution, and, of course, reduce reconstruction errors. In addition, the use of this algorithm produces a reconstruction that is stronger than the single-stage algorithm, and as its application is in Digital Signal Processing, it will be more applicable to real-time processes where memory limitations are overcome by a multistage algorithm where each stage is applied to a chip [4]. A Master’s thesis from the University of Southwestern Louisiana, USA, proposed an improvised multistage algorithm to perform task scheduling problems in resource selection so that the execution process is faster. This master thesis done by Feng discusses many types of task scheduling, such as direct scheduling, batch scheduling, static scheduling, dynamic scheduling, preemptive scheduling, and non-preemptive scheduling. The scheduling process is the process of allocating resources in the form of resources such as network links, processors, or expansion cards where the tasks performed can be in the form of data flows, threads, or processes. In addition, scheduling activities, which are the basis of the computing process itself, are carried out by the scheduler, which is carried out by a multistage algorithm to share resources effectively to achieve the target of achieving excellent service quality [5]. Meanwhile, Pham, a researcher from Georgia Institute of Technology, Atlanta, USA, introduced an algorithm to recognize a predetermined target aimed at the image data of the Forward-Looking Infrared (FLIR) sensor, and experiments using 147 images from the TRIM2 database using targets such as helicopters, tanks, howitzer, and vehicle. The developed algorithm approach starts with scene input and preprocessing. After that, two parallel activities are carried out by performing a target edge extractor and an interior target region extractor. After that, a target detector and the target location results are obtained. The preprocessing is carried out to reduce noise using a 3 × 3 morphological filtering operator, and the target edge information is extracted using an edge detector by calculating the gradient magnitude using the Sobel operator [6]. Another paper by Pham of the Georgia Institute of Technology researched the use of Constant False Alarm Rate (CFAR) detectors and morphological principles to improve detection accuracy and reduce false alarms on Automatic Target Recognition

260

V. N. Kurniadi et al.

(ATR) Synthetic Aperture Radar (SAR) targets. The developed model focuses on efficiency in the prescreening stage, which is one of the stages of the three stages of the ATR problem, where the prescreening stage is the essential stage carried out to reduce the computational load that focuses on the potential target area of an image. The CFAR algorithm is carried out in the following steps, starting from entering the scene, and then, it performs three processes that can be carried out simultaneously namely the two-parameter CFAR detector, global thresholding, and target/clutter variance comparison. Next, a majority filter is done, then the shape/size target discriminator and the target location are obtained [7]. Dharanipragada’s paper, a researcher from Watson research center, Yorktown Heights, USA, presented an algorithm that functions to find words in a human speech whose implementation is carried out in two stages: the phone-ngram representation and the rough-to-detail search stage. The phone-ngram representation stage, or offline data preparation, provides the level of speech phonemes that can be searched efficiently using phoneme recognition through a tree of vocabulary prefixes derived from the speech in forming an index or phone-ngram table. Meanwhile, a rough-todetailed search, commonly called a run-time search, is carried out to find the order of words/phones in the speech by doing phone-ngram matching [8]. Soraluze, a researcher from Spain, proposed improving the performance of the KNN classifier by using a hierarchy or multistage classifier, where the classifier is modeled by carrying out the process of improving classifier training to form a hierarchy and using rejection techniques at all levels of hierarchy. This multistage classifier based on the KNN classifier was implemented and tested with three datasets, one dataset from the University of California Irvine (UCI) repository, and two datasets from Statlog projects such as Statlog LandSat Satellite and shuttle statlog by applying the Training Algorithm and the Incremental Training Algorithm (ITA). Meanwhile, the proposed Multistage Recognition Algorithm (MRA) is implemented as a classification process where classifier training is carried out in stages and positioned as a hierarchy according to the number of patterns. In addition, this model is also equipped with a Multistage Recognition Algorithm with Active Memory (MRAAM), which works similar to MRA, but this MRAAM still stores data at the previous stage hierarchically. This study uses a multistage classifier, either with MRA or MRAAM, making the classification process faster [9]. Furthermore, Jazayeri from the University of Calgary, Canada, used a multistage approach to model dynamic power system loads to overcome the identification problem. This model estimated model parameters by performing equations for dynamic power system loads using a zero-order containment method followed by a Nonlinear AutoRegressive Moving Average with exogenous input (NARMAX) polynomial order-2 model. After that, the slightest quadrant approach was carried out to predict the NARMAX parameters, and the values found in the early stages were used as a starting point for Levenberg–Marquardt optimization to calculate the optimal parameters [10]. Jianshuang Cui from the University of Science and Technology Beijing, China, proposed using a multistage algorithm for Job Shop (JSP) scheduling problems, starts by randomly generating initial variables and continues by connecting the stages until

A Review of the Multistage Algorithm

261

a stop condition is found. The algorithm was successfully applied to 162 instances and showed that the algorithm was robust and simple in the implementation [11]. Next is the pseudocode of the multistage algorithm 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12.

Get the initial solution randomly. Call Generalized Memory Polynomial (GMP) while Iteration counter < MaxTrials {Call New Solution Generation Procedure (NSGP) for getting a new solution if the new solution is feasible? then call Improved Critical Path Algorithm (ICPA) Getting Current Best Makespan (CBM) if CBM < OptimVal? then OptimVal = CBM else Undo the Swap operation using local search } output: OptimalVal and Current Best solution

Zamyatin from Tomsk Polytechnic University, Russia, proposed a lossless compression algorithm performed in three stages using a wavelet transform to increase the compression ratio. The three stages are 1. 2. 3.

Finding the transformation coefficient by performing a wavelet transform from the initial data. Formation of deviation data sets by considering the functional relationship of Albedo values between different image bands. By using the traditional compression algorithm to compress the data obtained.

This three-stage algorithm was implemented in Borland Developer Studio 2006 without special attention to code optimization and by using a computer that has an Intel Pentium IV processor specification with a processor speed of 2.8 GHz and tested against 10 Remote Sensing (RS) datasets such as SPOT, ADAR-5000, Airphoto, Landsat-MSS, Landsat-TM and Flightline C1 [12]. El Attar, a Ph.D. student from Dhar El Mahraz University, in his 2011 paper investigated and proposed an algorithm for automatic calibration process on a camera, where this algorithm starts with an initialization stage to get the focal length and then estimates the camera’s intrinsic parameters using a multistage algorithm [13]. In the proposed multistage algorithm model, there are five iterations carried out, namely: 1.

2. 3.

The first iteration is done by estimating the initial focal length, where the formula is run to minimize the camera with zero tilt focus and have a main point in the center of the image. The second iteration is carried out to estimate the aspect ratio based on the solution obtained from the first iteration. The third iteration is carried out to estimate the main point by estimating the coordinates of the main point of an image and the previous iteration’s output as an estimate to minimize variations with two focal distances and the main point.

262

4.

5.

V. N. Kurniadi et al.

The fourth iteration is carried out to re-estimate the focal length the same as the initial iteration to achieve automatic camera calibration and 3D reconstruction. The previous iteration’s output will be used as an initial estimate for minimization. The fifth iteration is carried out to refine the camera’s intrinsic parameters using the output of the fourth iteration to improve all parameters.

He Yuqing, a researcher from the research institute of the Hunan electric power company, Changsha, China, proposed a new algorithm and a multistage algorithm for the configuration of electric power distribution networks. System reliability was measured by three index parameters such as Average Service Availability (ASAI) and Energy Not Supplied (ENS) and then combined with a power loss index to configure the power distribution network. This fast multistage algorithm was developed for this optimization problem, and the reliability index was not calculated in the previous reconfiguration stage [14]. The Nan Ho researcher from Vietnam National University discussed the multistage algorithm, divided into three stages: preprocessing, coordinating, and postprocessing, where each stage is a search algorithm. At the initial stage this stage consists of two sub-activities, namely Coarse Optimization (CO) and Pre-optimization, whose function is to determine attractive locations. In the second stage, the transition between stages avoids overlapping, where the central coordinator acts as a transition and manages the number of instances in the algorithm. Meanwhile, there are two sub-activities in the third stage, namely looking for the global optimum and the local optimum, which function to obtain the optimal solution. The multistage algorithm is called a grid-based hybrid algorithm which is the formation of Latin Hypercube Sampling (LHS), which is a collaboration of three agents such as Coordinator Agent (CoA), Instance Agent (InA), and Evaluator Agent (EvA), where each agent can send and receive information [15]. Ruinskiy, a researcher from Tel-Hai College, Israel, presented an algorithm to find fricatives in a sound speech to record data sources where fricatives are helpful in applications for the deaf for the excessive accentuation of phonemes which can degrade the aesthetics of sound recordings. In the first stage, this multistage algorithm performs a classification process with the Linear Discriminant Analysis (LDA) algorithm to detect fricatives in speech recording sentences. In the second stage, the detection results in the first process in the form of phonemes are reclassified using the Decision Tree (DT) algorithm to eliminate false detections. This algorithm was tested on a Texas Instruments Massachusetts Institute of Technology (TIMIT) audio database corpus containing hundreds of audio sentences spoken by various speakers with different dialects, and the detection rates across the entire range of fricative phonemes were obtained. In this case, the data tested using MATLAB with the MatlabADT toolbox was on 1680 sentences by 168 different speakers where the sentences contained more than 4600 fricatives [16]. Furthermore, Wang from China’s Dalian Jiaotong University proposed a Multistage Algorithm (MA) for optimization of loading problems and capacity vehicle

A Review of the Multistage Algorithm

263

routing problems (LCVRP) and for solving capacity vehicle routing (CVRP) problems. This MA is run in 3 stages; where the first stage is loading to set all consumer demands to a minimum of vehicles. The first and second stages are intended to produce the best solution. Then in the second stage, the parallel 2-opt algorithm solves the traveling salesman problem and schedules the consumer route. Meanwhile, the best solution results are optimized using the Tabu Search algorithm (TS) [17]. Guillaume Aupy from INRIA, France, tested a multistage algorithmic model for adjoint computing developed by Stumm and Walther. The Stumm and Walther model is a development of the Griewank and Walther model, but still, the results are not optimal. The paper described four problems that are overall to solve the main problem, namely the Adjoint Computing (AC) problem, all of which are to minimize the makespan of the AC problem [18]. There are seven lemmas described in this paper, namely: 1. 2. 3. 4. 5. 6. 7.

There is an optimal solution that has a structure. An optimal solution for Problem 2 satisfies (P0) and (P1). There is an optimal solution for Problem 2 that satisfies (P0), (P1), and (P2). There is an optimal algorithm for Problem 2 that satisfies (P0), (P1), (P2), and (P3). Provided the optimal algorithm for Problem 2 that satisfies (P0–3). An optimal algorithm for Problem 2 satisfies (P0–3) and (P4). An optimal algorithm for Problem 2 satisfies (P0–4) and (P5).

Cherneta, a researcher from Tomsk Polytechnic University, Russia, developed viola-jones based on the AdaBoost algorithm, an image object detection algorithm that applies the Haar feature and tests it on vehicle license plates. The test as character recognition in the form of 5000 training data of 28 × 52 pixel image measurements presented in images with different images from camera angles, contrast, and lighting. Furthermore, the classification was tested using the Open CV library by applying 2000 vehicle images and the algorithm training time was 32.4 h, and the test accuracy reached 98.21%. The algorithm was carried on in two stages: character segmentation and recognition. This algorithm had a classification cascade architecture and adaptive operating principles, wherein the cascade process each stage uses the Haar feature as a weak classification [19]. Meanwhile, Zhang from Boehringer Ingelheim, Shanghai, China, proposed a multistage AICaps model to select the best subset that implements AICi, an “enhanced” variant of Akaike Information Criteria (AIC) and evaluated by Monte Carlo simulation. The AICaps model had several stages starting from stage 1, comparing the corrected AIC (AICc) for the Sum of Squared Error (SSE) model with a minimum order equal to 1, with AICi for all models of order greater than 1. Stage 2 compares AICc as a model of at least 2 SSE with AICi for the entire model order greater than 2. For the last stage, which is a continuation of the previous stages, the actual model is the SSE model with a minimum order of S − 1 or model full of order S, and the model with the smallest AICc is selected [20].

264

V. N. Kurniadi et al.

Furthermore, Paulus from Georgia Tech Research Institute, USA, presented an algorithm that predicts the multistage linear and nonlinear phase components of an extended well time target signal for long-lived coherent integration. This multistage algorithm uses the inner product to detect the signal components from the phase-specific dictionary and additionally is performed to predict the target phase components unknown up to the sixth order to generate an accurate extended-stay multiphase signal model. The first stage of the multistage algorithm is carried out to determine the approximation of the linear phase component of the signal, and compared to the linear phase signal model. The multistage signal model generated from the multistage algorithm is better at maximizing the output. This multistage algorithm can be applied to the target movement parameters such as lane or road environment changes and traffic acceleration conditions. The stationary limit of this multistage algorithm is at the individual target scatter points and quiescent 4.0 S is the upper limit [21]. Furthermore, Kaven and Ghobadi from the Iranian University of Science and Technology proposed a multistage model for domain decomposition and allocation to reduce the cost and time for distribution of blood components between hospitals and blood centers by applying graph partitioning or the so-called p-median methodology and metaheuristic optimization algorithms such as Enhanced Colliding Body (ECBO) algorithm. In the first stage, this multistage algorithm is carried out by forming an adjacency matrix from the graph and, in the second stage, partitioning the resulting graph from the first stage into subdomains. This multistage algorithm was tested to minimize the distance between hospitals and blood centers in Tehran, Iran, which has an area of 730 km2 with a population of 8.4 million people and more than 150 health centers [22]. A researcher named Niemi from Stockholm University, Sweden discussed using a multistage algorithm that combines atomic molecular dynamics with Landau’s mean-field theory and describes folding and protein dynamics models with atomiclevel precision. This multistage model algorithm is well suited for characterizing the conformational state of intrinsically unstructured proteins, and the investigation of isolated monomeric Myc oncoproteins that can be encountered in cancer is an example. The Landau model approach was used to investigate that since monomer Myc is unstable and to analyze the highly degenerate structural landscape. Thermal stability properties were analyzed using atomic molecular dynamics and a group of structures were observed using two helical segments of the original leucine zipper parallel to each other [23]. Algorithm aims to design a method that can accurately diagnose the rhythm for compressions generated by a piston-driven pacemaker used to deliver Cardiopulmonary Resuscitation (CPR). Meanwhile, Zambri from California State University, USA, proposed a multistage algorithm model to calibrate the hemodynamic model and predict the parameter values of the biophysiological system and the external stimulus model, respectively. The proposed multistage algorithm adopts a predictive approach to predict two sets of parameters that have different properties and scales, and a multi-step strategy is applied by applying a combination of the Tikhonov regularized Newton Method (TNM) and Cubature Kalman Filter (CKF) algorithms and

A Review of the Multistage Algorithm

265

the combination of the two is called TNM-CKF algorithm. The proposed multistage algorithm can calibrate hemodynamic models without a prior knowledge of the values of biophysiological parameters and the ability to characterize one or more events. Moreover, different activity levels of different neurons can be distinguished by this multistage algorithm, wherein the calibration of this model provides control input parameters with excellent accuracy in the obtained results [24]. Furthermore, Isasi from the University of the Basque Country UPV/EHU, Spain, proposed a multistage algorithm that used two filter artifacts, namely a rhythm analysis algorithm classified by Electrocardiogram (ECG) slope and Recursive LeastSquares (RLS). This study used data from 230 cardiac arrest patients treated with the LUCAS 2 mechanical CPR device. In addition, the data set consisted of 201 and 844 shockable and non-activated ECG segments, of which 844, 270 were asystole and 574 Rhythm Organized. Two RLSs were used to reduce CPR artifacts, followed by applying a three-stage shock and no-shock decision during mechanical compression of the piston movement based on the standard defibrillator algorithm and ECG tilt decision. The data were randomly separated into training and testing data of 60% and 40%, respectively [25]. Lastly, Harkat, a researcher from Batna University, Algeria, conducted a study by proposing a multistage algorithm that aims to remove noise in the electrocardiogram signal and is carried out in two stages where the first stage of noise variation is estimated using DONOHO, followed by baseline based wavelets. Then, an adaptive Wiener ID is used to remove noise in the second stage. A fine Savitzky–Golay (SG) filter was also applied to extend the restoration process [26].

5 Discussion and Opinion Based on the previous literature review on 26 publication content with statement multistage algorithm, some ideas, thoughts, and comments related to those literature reviews are listed. The need to effectively figure the subordinates of a capacity, emerges most of the time in numerous territories of logical registering, including scientific improvement, vulnerability evaluation, and nonlinear frameworks of conditions. When the first subsidiaries of a scalar-esteemed capacity are required, alleged adjoint calculations can process the slope at a cost equivalent to little consistent occasions of the expense of the work itself, regardless of the number of free factors. This adjoint calculation can emerge from discretizing the constant adjoint of a halfway differential condition or applying the alleged invert or adjoint method of algorithmic (likewise called programmed) separation to a program for registering the capacity. Multistage coordinates the Active Distribution Network (ADN) planning process, where the planning is worried about the portion of employments to rare assets (machines). In the income board, each activity has a specific weight (or income), and the objective is to yield a practical subset of the occupations with the most extreme complete weight. Moreover, the scheduler must choose which machine,

266

V. N. Kurniadi et al.

each acknowledged activity ought to be appointed and when it ought to be executed, inside the time interim, between its discharge date rj and cutoff time dj. The goal is to expand the all-out weight of acknowledged employments. The proposed method optimized multiple planning options, that is, in a structured way. Meanwhile, huge size multistage stochastic 0–1 optimization issues are exceptionally hard to understand, primarily because of the number of imperatives and the 0–1 factors. To improve the exhibition of calculation for taking care of substantial estimated issues, a few considerations dependent on the presentation of the equal adaptation are introduced. Multistage stochastic programs typically exhibit time incoherence if the target is neither in expectation nor in full functionality. The development and widespread use of stochastic programming is closely linked to the growing capacity of computing made available from the field’s inception. A multistage linear stochastic model was designed to maximize electricity generation, storage, and transmission investment over a long planning period. Moreover, multistage stochastic issues emerged in a wide assortment of real-world applications in the vitality, account, and transportation fields. Right now, multistage stochastic direct projects that fulfill the accompanying conditions: 1. 2. 3.

The time skyline length T is limited yet conceivably huge (there might be several periods and stages); For each timeframe, the set of tests acknowledge of the exogenous data process is limited (and moderately little); The stage cost is a direct capacity of the choice for each stage.

Random Matching (RM): RM is a straightforward and effective strategy to process maximal coordination and limits the coarsening level in heuristic strategies like covetous calculation. In adjoint computation, memory and circle are conventional terms for a two-level stockpiling framework, demonstrating any stage with a double memory framework. The architecture of rhythm analysis in Electrocardiogram is beat examination. Sifting ought to uncover the basic heart mood of the understanding. Subsequently, sˆecg(n) was utilized to analyze the musicality as shockable or nonshockable. Fuzzy numbers propose scientific methods for determining specific solutions. They developed a far-reaching secluded system to determine different setups of fluffy numbers for fluffy positioning. Moreover, the Data Filtering-Based Recursive Least-Squares Algorithm in multistage Algorithm is the Recursive Least-Squares calculation which has high proficiency in utilizing every single estimated datum at each progression, so it has excellent boundary, estimation precision and high union rate. A small percentage of particular nodes are called anchors or beacons. Anchors know their areas since they may have GPS equipment or are conveyed to known areas. The rest of the nodes are deployed and sent indiscriminately to areas with area mindfulness. All atomic molecular dynamics intends to stimulate the time development of every molecule in a given protein, counting solvents. It delivers every iota’s discrete and piecewise direct time direction to answer a discretized (semi-)old-style Newton’s condition. Algorithm for the multistage stochastic problem through more straightforward issues to construct the pursuer’s instinct, initially, uses a combining lower-bound

A Review of the Multistage Algorithm

267

and upper-bound. In the generation of double limits, a Progressive Hedging (PH) Bounding Approach was used. The first methodology for producing double-limits linear programming relaxations of an adjusted variant of the unique MSSP plan utilize the PH. Multistage stochastic programming has discovered applications in an assortment of parts. In finance, MSP has been applied to portfolio advancement to amplify the average return while controlling the hazard, just as in resource obligation board. In the vitality segment, an old-style achievement story is aqueous age planning for Brazil, including the month-to-month arranging of intensity age of an arrangement of hydro and warm plants to fulfill vitality need even with stochastic water inflows into the hydro-supplies. Various diverse arrangement ways to deal with dynamic streamlining issues for frameworks depicted by normal differential conditions ODES or DAEs have been proposed. The critical quality of the whole discretization approach plot above is how streamlining is carried out across the discretized factor space and discretized imperatives. An extraordinary calculation depends on recently characterized vitality and unpredictability boundaries (EVP’s) as the essential progressive estimation (SA) factors has been demonstrated to be astoundingly steady and effective. Use of the instability boundaries as SA factors, typically leads to the decision of the recently characterized family member S-boundaries as the factors of an iterative technique, in which the semi Newton strategy for Broyden is proficiently utilized, to quicken union. Utilization of the S-boundaries, which are extraordinary mixes of the fluid and fume stage rates and the stage temperatures, maintains a strategic distance from the challenges related to connections between these factors. Furthermore, it is conceivable to comprehend the entirety of the model conditions without any of the interior irregularities that emerge with different strategies. Meanwhile, using calculation to an issue in stages, can decrease the measure of time taken to find that arrangement (comparative with the time taken to apply a calculation to the whole issue) and impressively improve the nature of that arrangement. The technique offered is a mixture of heuristic sequencing and transformative strategies, which can outperform either strategy alone. To comprehend the massive case of these issues, a k-stage requirement system that yields the worldwide ideal specifically cases and is helpful for issues where endogenous vulnerability is uncovered during the first few time frames of the arranging skyline is proposed. To take care of the more broad issues of enormous size, a NAC unwinding technique dependent on loosening up the NA requirements and including them if abused is also proposed. At long last, a Lagrangian deterioration calculation that can foresee the specific lower limits for the arrangement gotten is portrayed. Large-scale problems are beyond the range that, the new heuristics generate in less than one percent of the computing resources required by the optimal procedures solution. Moreover, it is conceivable to mix some heuristics and select the best arrangement as the great arrangement. Further, enlargement of heuristic calculations to correct arrangement procedures may help in improving the computational effectiveness and enhance the scope of the materialism of the specific arrangement strategies, since few advancing calculations require a preliminary answer to start the inquiry for an ideal

268

V. N. Kurniadi et al.

arrangement. Any end-point limitations from balance way imperatives, will experience the ill effects of their angles, as for the improvement boundaries being zero at the arrangement. In any case, it is intriguing to take note that these issues can be eased by receiving a half-breed approach like that of imbalance imperatives. Subsequently, notwithstanding the end-point limitations, every correspondence imperative is upheld as a point limitation at the stage limits. The current methodologies for tackling the Multistage Stochastic issue, depend on a model that abuse the moderate stream law and might cause an absurd outcome. The most popular calculation actualized a GA utilizing a Prufer-coded form that is a helpless decision in transformative calculations to tackle the multistage stochastic issue. The principal reason was introducing an increasingly productive and successful calculation to discover an approximated multistage stochastic arrangement. The settled decay calculation for illuminating multistage stochastic direct projects can adequately be executed equally without significant alteration. The equal execution exploits the freedom of subtrees to disperse the arrangement work among various laborer errands. Every specialist task requires correspondence with a chief errand to take care of the whole issue. Therefore, the settled disintegration is perfect for systems with similarly moderate correspondence times.

6 Conclusion In conclusion, the multistage algorithm has various concepts and forms. This multistage algorithm has many functions in each form, one of which is the stochastic model of the multistage algorithm, seen from various papers implemented by the multistage algorithm itself in the algorithm. This stochastic model has optimizations that are quite difficult to understand, and usually, this model is a typical time incoherence. The results of the literature review of 26 published papers containing the multistage sentence algorithm can help understand the meaning of the term multistage algorithm itself, which is the use of multistage intended for the development of an algorithm carried out with more than one stage in its application. The use of multistage algorithms in various fields of human life is undeniable that this multistage algorithm, apart from various concepts and implementations, does not reduce its function as part of how to think like a human, which is represented in an algorithm is called a multistage algorithm. The combination of multistage algorithms with other algorithms show the greatness of this algorithm and any algorithm, which is of course inseparable. Moreover, each algorithm can support each other to certainly help and increase the efficiency in human life, which is undoubtedly assisted by this algorithm as an implementation of how to think like a human.

A Review of the Multistage Algorithm

269

References 1. H. El-Shishiny, M.S. Abdel-Mottaleb, M. El-Raey, A. Shoukry, A multistage algorithm for fast classification of patterns. Pattern Recogn. Lett. 10(4), 211–215 (1989) 2. C.K. Chen, M.A. Duran, M. Radosz, Supercritical antisolvent fractionation of polyethylene simulated with multistage algorithm and SAFT equation of state: staging leads to high selectivity enhancements for light fractions. Ind. Eng. Chem. Res. 33(2), 306–310 (1994) 3. Z. Li, Rho-multistage algorithm on elliptic curves. Beijing Ligong Daxue Xuebao/Trans. Beijing Inst. Technol. 15(3), 261–264 (1995) 4. M. Kartal, B. Yazgan, O.K. Ersoy, Multistage parallel algorithm for diffraction tomography. Appl. Opt. 34(8), 1426 (1995) 5. Z. Feng, Improved multistage algorithm for task scheduling. Master thesis, University of Southwestern Louisiana, USA, 1996 6. Q.H. Pham, M.J. Smith, Morphological multistage algorithm for recognition of targets in FLIR data. Proc. SPIE Int. Soc. Opt. Eng. 2756, 14–25 (1996) 7. Q.H. Pham, T.M. Brosnan, M.J. Smith, Multistage algorithm for detection of targets in SAR image data, in Algorithms for Synthetic Aperture Radar Imagery IV, July 1997, vol. 3070 (International Society for Optics and Photonics, 1997), pp. 66–75 8. S. Dharanipragada, S. Roukos, A multistage algorithm for spotting new words in speech. IEEE Trans. Speech Audio Process. 10(8), 542–550 (2002) 9. I. Soraluze, C. Rodriguez, F. Boto, A. Cortes, Fast multistage algorithm for K-NN classifiers, in Lecture Notes in Computer Science (Including Subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 2905 (2003), pp. 448–455 10. P. Jazayeri, W. Rosehart, D.T. Westwick, A multistage algorithm for identification of nonlinear aggregate power system loads. IEEE Trans. Power Syst. 22(3), 1072–1079 (2007) 11. J. Cui, L. Cheng, T. Li, A multistage algorithm for the job shop scheduling problem, in IEEM 2009—IEEE International Conference on Industrial Engineering and Engineering Management (2009), pp. 808–812 12. A. Zamyatin, Multistage algorithm for lossless compression of multispectral remote sensing images, in ISPRS TC VII Symposium—100 Years ISPRS, Vienna, 5–7 July 2010, vol. XXXVIII, Part 7A (IAPRS, 2010) 13. A. El-Attar, M. Karim, H. Tairi, S. Ionita, A robust multistage algorithm for camera selfcalibration dealing with varying intrinsic parameters. J. Theor. Appl. Inf. Technol. 32(1), 46–54 (2011) 14. Y. He, D. Liu, C. Zeng, M. Wen, Novel model and multistage algorithm for distribution network reconfiguration considering system reliability. Dianli Xitong Zidonghua/Autom. Electr. Power Syst. 35(17), 56–60 (2011) 15. T.N. Ho, N. Marilleau, L. Philippe, H.Q. Nguyen, J.D. Zucker, A grid-based multistage algorithm for parameter simulation-optimization of complex system, in Proceedings—2013 RIVF International Conference on Computing and Communication Technologies: Research, Innovation, and Vision for Future, RIVF 2013 (2013), pp. 221–226 16. D. Ruinskiy, Y. Lavner, A multistage algorithm for fricative spotting, in Proceedings—2014 22nd Annual Pacific Voice Conference—Voice Technology: Software, Hardware Applications, Bioengineering, Health and Performance, PVC 2014 (2014) 17. C. Wang, C. Jin, J. Han, A multistage algorithm for multi-objective joint optimization of loading problem and capacitated vehicle routing problem. ICIC Express Lett. Part B Appl. 5(5), 1453–1459 (2014) 18. G. Aupy, J. Herrmann, P. Hovland, Y. Robert, Optimal multistage algorithm for adjoint computation. SIAM J. Sci. Comput. 38(3), C232–C255 (2016) 19. D.S. Cherneta, A.A. Druki, V.G. Spitsyn, Development of multistage algorithm for text objects recognition in images, in 2016 International Siberian Conference on Control and Communications, SIBCON 2016—Proceedings (2016) 20. T. Zhang, J.E. Cavanaugh, A multistage algorithm for best-subset model selection based on the Kullback-Leibler discrepancy. Comput. Stat. 31(2), 643–669 (2016)

270

V. N. Kurniadi et al.

21. A.S. Paulus, W.L. Melvin, D.B. Williams, Multistage algorithm for single-channel extendeddwell signal integration. IEEE Trans. Aerosp. Electron. Syst. 53(6), 2998–3007 (2017) 22. A. Kaveh, M. Ghobadi, A multistage algorithm for blood banking supply chain allocation problem. Int. J. Civ. Eng. 15(1), 103–112 (2017) 23. A. Niemi, J. Liu, J. Dai, J. He, N. Ilieva,. Towards multistage algorithm to model intrinsically unstructured proteins, in APS March Meeting Abstracts, Mar 2017, vol. 2017, pp. M1–248 24. B. Zambri, R. Djellouli, T.M. Laleg-Kirati, An efficient multistage algorithm for full calibration of the hemodynamic model from BOLD signal responses. Int. J. Numer. Methods Biomed. Eng. 33(11) (2017) 25. I. Isasi, U. Irusta, E. Aramendi, U. Ayala, E. Alonso, J. Kramer-Johansen, T. Eftestol, A multistage algorithm for ECG rhythm analysis during piston-driven mechanical chest compressions. IEEE Trans. Biomed. Eng. 66(1), 263–272 (2019) 26. A. Harkat, R. Benzid, N. Athamena, A multistage algorithm design for electrocardiogram signal denoising. J. Circuits Syst. Comput. 30(4) (2021)

Technology for Disabled with Smartphone Apps for Blind People Hartato, Riandy Juan Albert Yoshua, Husein, Agelius Garetta, and Harco Leslie Hendric Spits Warnars

Abstract Nowadays, technology is developing rapidly. The middle class has experienced technology such as smartphones, but unfortunately, there are still many applications that are not friendly for people with disabilities. The closest example is in Indonesia, where people with disabilities still receive less attention; all development support facilities do not pay attention to the comfort aspect for disabled users. This paper presents five features to help users, especially those who have blindness. These features are chat to speak, chat using voice, motion detect for emergency needs, detect object, voice to search engine, and weather information. In addition, a use case diagram was used to describe the application process and a class diagram to describe the database design. Keywords Mobile application · Application for disabilities · Vision assistant apps · Blind users application · Information systems

Hartato · R. J. A. Yoshua · Husein · A. Garetta Information Systems Department, School of Information Systems, Bina Nusantara University, Jakarta 11480, Indonesia e-mail: [email protected] R. J. A. Yoshua e-mail: [email protected] Husein e-mail: [email protected] A. Garetta e-mail: [email protected] H. L. H. S. Warnars (B) Computer Science Department, Graduate Program, Doctor of Computer Science, Bina Nusantara University, Jakarta 11480, Indonesia e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 I. J. Jacob et al. (eds.), Expert Clouds and Applications, Lecture Notes in Networks and Systems 444, https://doi.org/10.1007/978-981-19-2500-9_19

271

272

Hartato et al.

1 Introduction Persons with disabilities, who in daily conversation are referred to as people who are lacking or disabled, are often considered and ridiculed as unproductive secondclass citizens, unable to carry out their duties and obligations so that their rights as humans and citizens are neglected. Meanwhile, many technological advances have also been made in disabilities, both in software and hardware, especially for physical disabilities, individuals with visual or hearing impairments [1]. However, seeing the conditions of people with disabilities around us, it can be seen that applications for people with disabilities are still needed to support all the problems faced by persons with disabilities. Therefore, an application that is suitable and supports people with disabilities is still needed to help people with disabilities live their lives better. Moreover, discriminatory treatment for children with disabilities can have long-term and traumatic effects that can affect future employment opportunities and participation in civilian life, especially for persons with disabilities who have experienced it since birth. This will impact their social life, intelligence, mentality, and self-confidence in society. Refer to data released by UNICEF and WHO, according to a 2004 study, 93 million children under the age of 14 have disabilities, and for those aged 18 years and over, the number exceeds 150 million. In general, people with disabilities are divided into four parts, namely the first is physical disabilities, such as movement disorders that cause they cannot walk, and the second is sensory disabilities, such as hearing or vision problems. The third is intellectual disability, such as memory loss, and the fourth is a mental disability, such as phobia, depression, schizophrenia, or anxiety disorders. Vice President of the Republic of Indonesia K. H. Ma’ruf Amin at the Inclusion Indonesia Dialogue through a video conference at the Vice President’s Official Residence on Thursday January 14, 2021, stated that there were 209,604 Indonesians with disabilities based on data compiled by the Ministry of Social Affairs through the Disability Management Information System or 0.0007724492 of the total population of Indonesia. 2021 with 271,349,889 inhabitants [2]. However, data on persons with disabilities in Indonesia is certainly different from one agency to another due to the scattered data. This paper will only be limited to blind people with disabilities whose data is also unclear, considering that data on persons with disabilities are not integrated. Blind people in Indonesia still do not get enough care from other people. For example, many children are hidden by their families because they cannot do much and are afraid of being hurt. Many public services for people with disabilities do not support standards and become a big problem. Many adolescents do not continue their education due to a lack of a support system because they are blind people. Blindness causes sufferers to experience cognitive, motor, emotional, and social problems. Disadvantages of sufferers can be reduced with the help of adults around them. Therefore, the role of adults around them becomes critical. Older people are their trustworthy source of strength to live the rest of their lives. These problems will become significant if not addressed as soon as they determine their future. So, this paper will focus on one

Technology for Disabled with Smartphone Apps for Blind People

273

aspect of the many disabilities, blindness, and discuss how to help the problem by using technology to help their lives in the future. Understanding what persons with disabilities are, especially persons with disabilities who are blind, it is necessary to interact with them. From the interactions made, it can be seen whether disabilities can impact increasing their mentality, IQ, and EQ. Seeing the problems faced by people with visual disabilities, an application was designed and built which aimed at helping persons with disabilities, especially blind people, carry out their daily activities. So that they can do their work more comfortably or do activities that were not possible before. Many mobile applications for people with disabilities, such as Optacon, BrailleType, Aipoly, and Evasion AI devices. It could be an example that many developers are aware of this problem.

2 Existing Works Increase self-knowledge in directing careers for blind people; they also need to be educated in special schools for people with disabilities. Not to categorize them, but to ensure they are getting exemplary service. Hope for the future that they will get the right job for them and not wholly lean on someone for the rest of their lives. There are a small number of persons with disabilities to find the decent work they need because the competition is too fierce. Fixing their shortcomings, there are many ways to help them get better, one of which is a technology [3]. Elmar Krajnc from FH Joanneum Kapfenberg, Austria. According to Elmar’s paper and his research from seven users, they hired a test person to test which user interface could be the best option for blind people. There were two blind, two visual disorders, and three ordinary people. His research could be successful because the navigation and position of each feature were relatively simple and easy to access. These methods were used to make a proper application for blind people. The result is that the interface collaborating with the talking touch feature has a success rate of 88%, and the gesture application had an 89% success rate. From this research, some data and information can be learned to lead us to fix some deficiencies in the proposed mobile application or even as research to help other applications. Approach them also needs to be done with care, especially with children. From the explanation of Webster and Roe, there are two approaches to understanding it more deeply. First, it needs to treat them like children without disabilities because they do not feel different from other people with that treatment. Second, the prediction of the possibility of their different developmental styles. This approach can better understand what they need and how to make excellent mobile applications for them. The application design has been made easy to use. Even users who have never had experience with this application before can learn to use it quickly, and it can be implemented as soon as possible to help their daily life [4]. Mobile applications are IT software artifacts developed explicitly for mobile operating systems installed on handheld devices. Application design sometimes does not pay attention to the aspects

274

Hartato et al.

of making a functional application design. Effectiveness and efficiency are the two main aspects of making a good application design [5]. There are so many mobile application designs that can help blind people, to help overcome this problem; it is present in various ways, and it can be through cellphones, google glasses, virtual reality, and even combining humans with technology [6]. In addition, the technology applied to humans has been carried out by the Optacon device. This technology, which still uses infrared rather than ultrasonic, is still a dilemma because ultrasound is needed to have a proper vision to see the obstacles ahead. The main deal is that the technology is quite expensive; it is around 1500 GBP. So, an intermediary is still the best choice as an intermediary to help people with disabilities [7]. Currently, mobile applications for people with disabilities are overgrowing because cellphones have turned into touch screens. Interaction with touch screens is even more demanding from a visual point of view. With the evolution of this gadget, application developers will be more comfortable making applications that are easy to interact with [8]. In one of the researches, the difficulty for visually impaired people to use the application is only how to familiarize themselves with this application; however, to use the application, they need to remember where the buttons are. The use of this mobile application is how this application can manage and care for people with disabilities [9]. They also want to have a support system that helps them and provides the right direction for raising their children. Good qualities are needed to meet the needs of persons with disabilities and sufficiently increase their happiness in life. Several studies related to services that should be aimed at persons with disabilities who have cognitive problems. So, this is where the use of mobile applications helps people with disabilities do things that are difficult for persons with disabilities to do themselves [10]. A study published recently has conducted a literature review of 235 papers that discuss the development of research on the application and use of mobile applications designed to provide solutions to problems faced by people with blindness and include existing problems, challenges, and opportunities for development in front of him [11]. A study in Malaysia conducted a literature review of 136 published papers between 2013 and 2019 which took from Google Scholar, IEEE Xplore, and Science Direct, where these papers discuss a speech recognition project that is used for people with blindness by using a mobile application that is equipped with object and distance detection technology for navigation purposes for people with blindness [12]. Including, a study in India conducted a literature review of published papers discussing smartphone applications used by people with visual impairments to help with all the daily problems faced by people with blindness, as well as looking at the opportunities, challenges ahead, and evaluating the usability of the smartphone application [13]. The authors carried out the literature survey from the UK in their paper, which compiles a published paper discussing the implementation of smart cities in the UK that support visually impaired persons in terms of their ease of mobility and quality of life. Several assistive technologies were developed to assist visually impaired

Technology for Disabled with Smartphone Apps for Blind People

275

persons by applying vision, speech, and augmented reality (AR) technologies, such as Buzzclip, Horus, Sunu Band, eSight, EVA, MAPTIC, 3D soundscape, AIRA, and so on [14]. There are several application implementations to support the blind, including Google with an android application called Lookout, and a smart system consisting of a microcontroller board, sensors using solar power, as well as a mobile application that utilizes the global positioning system (GPS) called Seeing GPS eyes or blind square. Its implementation can use several technologies such as small cells or beacons or radio frequency identification (RFID), plus the application of voiceover screen readers, smart furniture, tactile floors/paving, or so-called blister paving [15]. Meanwhile, another survey was conducted by researchers from Romania, where they searched for previously published papers on smart systems applicable to the blind implemented either by the internet of mobility (IoM) or internet of mobile things (IoMT) that implements sensors such as vibration, microphone, headphones, ultrasonic, camera, and photoelectric. Researchers from Norway conducted a study on the use of smartphones for assistive devices for blind teenagers in Nepal where a survey was conducted on 21 blind students, both male and female, aged between 15 and 17 years by looking at the level of benefits of using smartphones for reading aids for them in carrying out their duties as a teacher. Students always read the subject matter provided, especially in doing homework [16]. Also, a qualitative research study was conducted in Hong Kong, The Republic of China, where a mobile application that can be used by visually impaired persons in the tourism sector is proposed, where they can easily access information about tourism in the city of Hong Kong despite their limitations as blind people [17]. A study in New Zealand conducted a study for blind students in New Zealand where the use of digital tools in helping students with visual impairments is incomplete if there is no support for other factors such as the involvement of the closest person or parents. A researcher from Malaysia carried on the research, where created an interactive learning mobile application for learning mathematics with braille using Nemeth code to help blind students in Bangladesh where with this application, the student can quickly learn how to do easy calculations, including self-learning facility, and interactive features such as hearing and touching to physical things [18]. In addition, the group learning model, both with others with disabilities and with students who are non-disabled, can also help them understand the lessons taken and do group assignments and interact digitally as communication in the work group [19]. Meanwhile, Shakya, a Professor from Tribhuvan University, Nepal, conducted an in-depth analysis of the application of Big Data in the banking sector by prioritizing the use of tools, applications, and technologies that are appropriate for the banking sector, especially banking in India. This study conducted a survey on banking in India which also involved its customers, and several survey results showed the improvements needed in implementing big data in the banking sector in India [20]. Patel from RCOE, MU, Mumbai, India, built an application architecture to capture complaints from citizens in terms of monitoring and experiencing the services of government officials who are less severe in serving the people. The application built for this smartphone is more concerned with problems related to road infrastructure, which sometimes breaks down quickly, especially during the rainy season. In addition,

276

Hartato et al.

careless maintenance of road infrastructure will cost more funds and may even cause accidents for residents who use the damaged highway [21]. Moreover, Prasanna from Shri SSS Jain College for Women, India, developed a mobile application for blind and visually impaired people where communication with other people using this smartphone uses voice to text technology where the voice generated will be converted into text. In this case, persons with disabilities can make letters or writings through their voices. In addition, text to voice technology is also used where blind and visually impaired people can listen to written text, whether it is text obtained from another person. This is very useful for people with disabilities where they can channel their talents who like to read books or write-through sound. So they do not need to use their eyes to read but use their ears to read. In addition, the accelerometer sensor is used to detect the physical shock or shake of the smartphone so that it can be used as a key to open the smartphone. Besides that, it is used if people with disabilities experience a fall due to the high accelerometer value [22].

3 Proposed Smartphone Application The proposed system is a friendly use to a blind user. So, to fulfill their needs, it should know and make a simple yet effective process in this smartphone application. As simple as it could be for the process, it would make the blind user understand the process more and no need much time to learn using this application. The system flow in this application can be seen in Fig. 1. The flow of this application process is made as simple as it could be for more natural use. From starting, go through login but without using an ID or password because it will be an obstacle for a blind user. Next, it needs to access the proposed system using a phone camera as a procedure for its application to run. After that, there are so many features to which section of objects. However, the limitation is that to use full features, which need to pay some amount per month for money, could proceed by scanning the fingerprint for IOS. The processes are based on this figure. Based on Fig. 2a, it can be seen that this application’s main page is simple so that the user can use it easily. The first feature is that the user could chat with a friend of theirs by speaking through a microphone, and when their friend replies, the application will make the chat could speak by detecting the words. Moreover, based on Fig. 2b, this page has a feature to help the user when no one is around the user at home when the user suddenly collapses because of something wrong. It could detect when a user is holding their phone, and suddenly the user collapse and the phone drop to the floor. The gesture motion will detect it and send an emergency signal to other family and caretakers to take immediate action. If the signal is false, the user could shake it three times. This feature helps the user so much, decreases the chance of someone passing away, and helps to give much more service to people with special needs. So, in the future, all the people who need special needs and affection will feel cared for by someone even more. So, all people get what they deserve because at this moment,

Technology for Disabled with Smartphone Apps for Blind People

277

Fig. 1 Use case diagram of a smartphone application for blind people

many incidents get a late response from the caretaker and give a high risk of someone passing away. It must be able to minimize all the bad possibilities for someone. This feature can also be used by people who are not disabled to help them get treatment as soon as possible. The emergency signal sent will go directly to an agency’s emergency handling server without another intermediary, so it does not take much time. Moreover, the sending of this signal is very strong so that it is not disturbed by signal problems to respond to emergencies. Based on Fig. 3a, the sensor implanted in the phone and camera could detect objects around the user; this can be done because it makes an AI that gives the best experiences. Meanwhile, based on Fig. 3b, this feature could direct what users want to search through a pre-existing search engine. So, blind people also could surf through the internet like other people without limitations because all the commands and responses are using voice over from smartphones. Based on Fig. 4, the application could give information about outside local weather. Even though the user who has blindness could not see it, at least they could know the current weather status by hearing from voice notifications. So, they

278

Hartato et al.

Fig. 2 a Chat to speak menu, b motion detect menu

could prepare the best outfit to go outside. Moreover, the application could also tell the weather forecast if it would have a chance of raining or not, so the user could prepare to bring the umbrella. Furthermore, most of all, the interaction of this application by motion gestures like shaking phone and voice command makes more natural interaction between the user and the application. In order to make a proper smartphone application, many technologies could pair with smartphone features. One of them is the sensor in the application; in this way, it could let the application system make a feature that could collaborate with a sensor on the smartphone. The sensor is the best thing that it could insert in this application, and it is because the sensor is sufficient to give up-to-date data to the user about their health and could interact with surrounding to improve the use of this application. RFID-based assistive devices. The example was like when a new user downloaded the application, the application needs to ask permission to allow it to use a user camera and other features in a smartphone. It would be done by using a voice and gesture sensor. The application asks using voice, and the user interacting with shake their

Technology for Disabled with Smartphone Apps for Blind People

279

Fig. 3 a Detect object menu, b voice to search engine menu

phone to give permission and do not give permission by not shake their phone for 5 s. With sensors, could make an obstacle detection to prevent any incident that does not want to happen. However, to make this thing work, should spread the receiver of RFID in an area so the RFID implemented in a smartphone could send signals to another receiver around that area. The second component is a sensor proxy that acts as an online representative for the mobile sensors. By what objects are in front of him by notifying by a voice from the user’s mobile, including notification of updates about the weather around the user’s area. This proxy could work by sending those site addresses through a proxy. The proxy is always available and provides the entire sensor data still up to date. For people who are blind and recognize a new environment, this application will tremendously help users recognize the new environment quickly and precisely through the software’s sensors. Activity recognition is the application monitoring the activity by using signal processing in an environment. When users hold their phones, they could do the interaction by activity such as shaking their phones to have an interaction. In this way, the use of sensors could be optimized. Every ongoing activity, there will be

280

Hartato et al.

Fig. 4 Weather information menu

a confirmation from the system using voice notification. Moreover, when the user interacts with the application, the application will remind how many times has been spent by the user, the level of smartphone battery, and other things. The fall detection system is The feature that could help the user get emergency help when the user is alone and other family members are outside. When the user holds the smartphone and suddenly collapses, the smartphone will detect the phone’s gesture and send the alert to a family member and caretaker about the user that collapses. So, they could do something immediately. However, there is a confirmation if the phone is dropped accidentally by shaking the phone for 5 s. The interactive system’s feature gives the user that the voiceover from the application could help many things too. Such as reading the incoming chat from friends from different social media, helping them reply to messages from friends, helping the user play music from a playlist, or integrating with Spotify or Joox. A reminder of what time at that moment automatically reads news for the user when they say, “read the news today.”

Technology for Disabled with Smartphone Apps for Blind People

281

4 Conclusion Disabled people that all this time are suffering to live their life slowly could feel life again because time by time, there are so many assisting things and devices to help them. The improvement in technology also fulfilled what they need and want. Furthermore, there will be more advanced technologies to help them. By these changes into a good thing, many people and sides feel the extraordinary impact of those technologies. Hopefully, people who do not have disabilities need to encourage and help them make them enjoy their lives. It could come in many ways and many chances because everyone has the right to live a good life. Things can be better, and the mobile application that supports disabled people needs to be free of charge to decrease their life problems. Moreover, designing an excellent interface like a simple placement of features, an excellent interactive application, a clear voice for blind people to navigate by the application, and some touch gesture features to be implemented. Also, improved existing features to be more interactive, fast respond, and clearly could help more for the user. Embed a new and fresh feature with future technologies will do it. Collaboration with the internet of things method tries to integrate this application with another thing that could implement with Artificial Intelligence. Also, try to add a technology that uses things and combines humans, which is the user, with the technology itself. The example is like implanting the technology to the user’s hands, ear, legs, eyes, and even place near the brain to send and receive a signal between the technology and its user. The technology is implanted near the eyes and interacts with the user’s retina, and if it is possible with the user, the user’s retina still has a chance of working. So, the benefit that the user gets could directly he feel. In this way, make a breakthrough in technology for disabled people, especially those who have blindness. Moreover, every data gathered from the user is generated and collected to the server as data for further research purposes and improvement for technology and health care. Acknowledgements This work is supported by the Research and Technology Transfer Office, Bina Nusantara University, as a part of Bina Nusantara University’s International Research Grant contract number: No.017/VR.RTT/III/2021 contract date: 22 March 2021.

References 1. J. Oliveira, T. Guerreiro, H. Nicolau, J. Jorge, D. Goncalves, BrailleType: unleashing braille over touch screen mobile phones, in INTERACT 2011, Part I, LNCS, vol. 6946 (2011), pp. 100– 107 2. A. Csapo, G. Wersenyi, H. Nagy, T. Stockman, A survey of assistive technologies and applications for blind users on mobile platforms: a review and foundation for research. J. Multimodal User Interfaces (2015). https://doi.org/10.1007/s12193-015-0182-7 3. M. Bhuiyan, A. Zaman, M.H. Miraz, Usability evaluation of a mobile application in extraordinary environment for extraordinary people. J. Inst. Inf. Technol. (2014)

282

Hartato et al.

4. E. Krajnc, M. Knoll, J. Feiner, M. Traar, A touch sensitive user interface approach on smartphones for visually impaired and blind persons. J. Inf. Qual. e-Health 7058 (2011). https://doi. org/10.1007/978-3-642-25364-5_41 5. T. Guerreiro, P. Lagoa, P. Santana, D. Goncalves, J.A.P. Jorge, NavTap and BrailleTap: nonvisual texting interfaces. J. MIS Q. 39(2), 435–472 (2015) 6. G. Fortino, R. Giannantonio, R. Gravina, P. Kuryloski, R. Jafari, Enabling effective programming and flexible management of efficient body sensor network applications. J. IEEE Trans. Hum.-Mach. Syst. 43(1) (2013) 7. E. Terradas, An overview of the internet of things for people with disabilities. J. Netw. Comput. Appl. 35(2012), 584–596 (2011). https://doi.org/10.1016/j.jnca.2011.10.015 8. N. Pannurat, S. Thiemjarus, E. Nantajeewarawat, Automatic fall monitoring; a review. Sensors (2014) 9. J. Cecilio, K. Duarte, P. Furtado, BlindeDroid: an information tracking system for real-time guiding of blind people, in The 6th International Conference on Ambient Systems, Networks, and Technologies (ANT 2015) (2015). https://doi.org/10.1016/j.procs.2015.05.039 10. T. Loukopoulos, M. Koziri, N. Panagou, P.K. Papadopoulos, D.K. Iakovidis, Cloud video guidance as “Deus ex Machina” for the visually impaired, in Technological Trends in Improved Mobility of the Visual Impaired, ed. by S. Paiva (Springer, 2020), pp. 127–143 11. A. Khan, S. Khusro, An insight into smartphone-based assistive solutions for visually impaired and blind people: issues, challenges, and opportunities. Univ. Access Inf. Soc. 1–34 (2020) 12. A.M. Norkhalid, M.A. Faudzi, A.A. Ghapar, F.A. Rahim, Mobile application: mobile assistance for visually impaired people-speech interface system (SIS), in 2020 8th International Conference on Information Technology and Multimedia (ICIMU), Aug 2020 (IEEE, 2020), pp. 329–333 13. S.S. Senjam, Smartphones for vision rehabilitation: accessible features and apps, opportunity, challenges, and usability evaluation, in Software Usability, ed. by L.A.A. Castro, D. Cabrero (IntechOpen, 2021), pp. 1–18 14. R.M. Aileni, G. Suciu, V. Suciu, S. Pasca, J. Ciurea, Smart systems to improve the mobility of people with visual impairment through IoM and IoMT, in Technological Trends in Improved Mobility of the Visual Impaired, ed. by S. Paiva (Springer, 2020), pp. 65–84 15. D. Sobnath, I.U. Rehman, M.M. Nasralla, Smart cities to improve mobility and quality of life of the visual impairment, in Technological Trends in Improved Mobility of the Visual Impaired, ed. by S. Paiva (Springer, 2020), pp. 3–28 16. P. Sankhi, F.E. Sandnes, A glimpse into smartphone screen reader use among blind teenagers in rural Nepal. Disabil. Rehabil. Assist. Technol. 1–7 (2020) 17. K.L. Lam, C.S. Chan, M. Peters, Understanding technological contributions to accessible tourism from the perspective of destination design for visually impaired visitors in Hong Kong. J. Destin. Mark. Manag. 17, 100434 (2020) 18. L. Nahar, R. Sulaiman, A. Jaafar, An interactive math braille learning application to assist blind students in Bangladesh. Assist. Technol. 1–13 (2020) 19. E. Pacheco, P. Yoong, M. Lips, Transition issues in higher education and digital technologies: the experiences of students with disabilities in New Zealand. Disabil. Soc. 1–23 (2020) 20. S. Shakya, S. Smys, Big data analytics for improved risk management and customer segregation in banking applications. J. ISMAC 3(03), 235–249 (2021) 21. V. Patel, D. Kapadia, D. Ghevariya, S. Pappu, All India grievance redressal app. J. Inf. Technol. Digital World 2(2), 91–99 (2020) 22. S. Prasanna, K. Vani, V. Venkateswari, Design and development of a mobile-based application for automated SMS system using voice conversion technique for visually impaired people, in Innovative Data Communication Technologies and Application (Springer, Singapore, 2021), pp. 307–317

Mobile Apps for Musician Community in Indonesia Amadeus Darren Leander, Jeconiah Yohanes Jayani, and Harco Leslie Hendric Spits Warnars

Abstract Art and music are human needs in terms of socializing with each other, and every human civilization is always equipped with music. In Indonesia, every community activity, whether related to family, religious, or cultural events, is inseparable from the presence of music as a compliment and means of socializing society. People interested in music are not well informed about the choice of offers about the variety of music that can be held while music organizers and players have difficulty marketing their musical expertise. Seeing these existing problems, building an online platform where session musicians can interact and share is necessary to build a strong relationship between them was proposed. This paper outlines the design of a mobile app that provides an online forum for discussion of all things session musicians and helps find musical activity opportunities. The design was developed using Unified Modeling Language (UML) diagrams, mainly using use case diagrams, class diagrams, and activity diagrams, including user interface displays as a communication display between users and the applications they use. The application was built using Arduino Studio and MySQL to save the database. Keywords Community mobile apps · Information systems · Musician community mobile application

A. D. Leander · J. Y. Jayani Computer Science Department, School of Computer Science, Bina Nusantara University, Jakarta 11480, Indonesia e-mail: [email protected] J. Y. Jayani e-mail: [email protected] H. L. H. S. Warnars (B) Computer Science Department, Graduate Program, Doctor of Computer Science, Bina Nusantara University, Jakarta 11480, Indonesia e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 I. J. Jacob et al. (eds.), Expert Clouds and Applications, Lecture Notes in Networks and Systems 444, https://doi.org/10.1007/978-981-19-2500-9_20

283

284

A. D. Leander et al.

1 Introduction In the past decades, the changes in social and cultural practices in Indonesia have successfully popularized the emergence of different music genres, notably pop, rock, hip-hop, orchestra, and many more [1]. Moreover, well-established businesses are attempting to incorporate musicians in their work for purposes such as promotion or marketing [2]. Many of these large businesses use the services of a particular type of musician to aid in the development of musical pieces and recordings, called session musicians. Session musicians also called studio musicians or backing musicians, play an essential role in the music industry. They are considered by many people as the “best-kept secrets” in that particular industry [3]. By definition, session musicians are contracted to aid in live performances or recording sessions. They primarily play standard instruments such as the guitar, keyboard, drums, and bass, but few outliers can specialize in strings, brass, and woodwinds. Usually, session musicians are not permanently affiliated to a specific solo artist, band, or ensemble but instead are only temporarily hired as required. Great session musicians are remarkably creative, excellent sight-readers, and highly skilled in their respective instruments [4]. Despite everything mentioned above, the term session musician is still very foreign to the public’s ears. This is because session musicians are not always visible and, more often than not, do not receive the proper spotlight they deserve. Most of the popular music associated with celebrities and recognizable characters can be traced back to hired collaborators who work in the background at recording studios, away from the public eye [5]. A famous example of the unnoticed work of session musicians is Motown Records, a renowned American record label composed of professional session musicians who could improvise complex arrangements easily. They released multiple hit singles, yet none of the musicians who worked on the masterpieces, particularly in the 1960s, were individually credited and remains a mystery [6]. There is a dilemma in the method of information transport between session musicians. There is difficulty in ensuring that everyone gets a reasonable opportunity to propel, or even kickstart, their careers as session musicians [7]. Moreover, current and prospective session musicians do not have a robust platform where they can actively share information and experiences among each other. Therefore, session musicians cannot establish broader connections and relationships, aside from the small group they frequently work with. This negatively affects several aspects, one of them being job sharing. Without generalized communication, sharing information regarding jobs will not be equally spread out. This paper discusses a proposed solution to this problem. It outlines methods to implement a mobile-based forum application that enables session musicians to connect. With this app, session musicians will be able to share their knowledge and experiences and promote open jobs among each other through posts on a forum thread. It is a new effort to unite session musicians of all demographics. The following

Mobile Apps for Musician Community in Indonesia

285

sections will describe similar past research, methodologies used to achieve the objectives, and present the results and conclusion.

2 Existing Works Since there is only a limited amount of research regarding the development of mobile applications for musicians, this section will mainly focus on discussing topics that are related and relevant to this research paper, namely the benefits of music, common approaches to mobile application development, and user interface design, as well as research regarding the use of online forums for education and learning purposes. Music is associated with cultural matters such as rituals, magic, or healing purposes in the prehistoric ages. As time goes by, the purpose and use of music have revolutionized to become a form of personal enjoyment, artistic expressions, and entertainment [8]. In this current age, music and technology are very closely tied. The ability to share musical artworks and information on the internet is one possible way to enhance the engagement of Indonesian people in the musical culture itself. This is beneficial as music is one of the best remedies for the heart, mind, and soul [9]. Generally, musicians are at a significantly higher level of subjective well-being than non-musicians, meaning they are more satisfied with their lives and have a more positive outlook and emotions [10]. Mobile applications are currently on the rise and are expanding rapidly in the current industry. However, the development of mobile applications is still considered complex, as it is difficult to mirror the complete behavior of desktop applications [11]. Moreover, there is no single process to develop solid mobile applications as each app differs from the other [12]. Due to this reason, a dedicated framework lifecycle is required in order to produce high-quality applications. The phases in this life cycle resemble those of typical application development lifecycles: Identification, Design, Development, Prototyping, Testing, and Maintenance [13]. Aside from a concrete development process, various standard practices must also be abided by to improve app performance. Making sure that the mobile application is compatible and working correctly on multiple platforms is mandatory in this era. In contrast with desktop applications, mobile applications must be able to incorporate more instinctive and gesture-focused capabilities to provide user convenience. Lastly, an agile development approach is most suitable for mobile apps due to its everimproving nature [14]. A good user interface is a crucial element toward satisfaction in the eyes of both end-users and developers. The user interface connects the system with the software users to allow efficient interaction [15]. The user interface will reflect the whole system, as users often think of the user interface as the system itself. Therefore, the usability of the user interface is essential in the overall quality of the software [16]. The enhancement of applications has dramatically impacted how education and information access are provided to people [17]. An effective online application-based tool for learning is online discussion forums. Online discussion forums demonstrate

286

A. D. Leander et al.

promise for online learning, both individually and as a collaborative group [18]. They allow a simple yet effective method to communicate between online discussion forum participants, as they do not have to be in the same place or at the same time to be interacting with each other [19]. Although there are no face-to-face interactions and discussions between participants, they seem more able to build upon each other’s ideas and be more active in expressing their thoughts if they disagree on something [20]. Participants are also more willing to touch on sensitive issues and share honestly. The positive results are due to online discussion forums’ social interaction and collaborative nature [21]. Meanwhile, Vivekanandam, from Lincoln University College, Malaysia, developed a hybrid algorithm for the classification process in machine learning and model identification. Besides that, this approach is carried out by functional selection by securing the selection operator from the genetic algorithm. The proposed framework is run in six steps, starting with preparing population data, and the second step is calculating the fitness value. The third step is to make a selection in the genetic algorithm section, and the fourth step is to change the mutation. Meanwhile, the fifth step is to choose between the second and last step, the sixth step, the results are processed using the Support Vector Machine (SVM) [22]. Also, Mugunthan, a researcher from Sri Indu College of Engineering and Technology, India, designed the Extreme Learning Machine (ELM) method using the sigmoidal bias function in the classification process. In addition, handling stochastic matrices provides low performance for learning rate and robustness of determination, and this ELM version is modified in six steps to get better accuracy and minimize errors in the classification process [23]. Moreover, Chowdhury, in his master’s thesis at the Icahn School of Medicine at Mount Sinai, USA, conducted a systematic assessment of the community in New York City on the use of Covid-19 applications, especially those assessed were applications such as the mhealth application, which can be downloaded on the Appstore or Google Play Store. The study results show that most of the digital health applications designed to overcome the Covid-19 problem are not designed properly and do not adapt to the community’s literacy needs. This means that software engineering rules in capturing data using the user requirement tool are not correctly and adequately [24].

3 Proposed Idea The use case diagram for the proposed mobile application is illustrated in Fig. 1. In total, there are eight use cases to be developed within the application, which are “Register,” “Login,” “Create performance event,” “Manage performance event,” “Register event,” “Forum,” “Information” and “Rate and comment event.” Each use case will be further explained in the following sections. There will be two main types of users in the proposed mobile application: musicians and event organizers. Musicians who want to use the app to gain information

Mobile Apps for Musician Community in Indonesia

287

Fig. 1 Use case diagram of mobile apps for session musicians community in Indonesia

surrounding the world of session musicians and register for music events can register as a musician and enjoy the provided services. If users want to use the mobile app to look for session musicians to perform at their events, they can register themselves as event organizers. Figure 2a represents the initial landing page of the proposed mobile application and the logo app name, “JamSession,” lies on the center of the screen. This page contains two buttons, “Login” and “Register,” redirecting users to the individual login and register pages. The main menu for musicians is displayed in Fig. 2b. Users can view and edit their profiles when logged in as musicians using the “View Profile” feature. Users also have other feature options: “Create Forum Thread” allows musicians to create new forum threads to serve as discussion platforms, “Search for Forum” provides a method to go through and find forum threads that interest the user, “Register for Event” which allows musicians to search, discover and register for events, and “Information” aims to serve users with news and articles regarding session musicians. The “Logout” feature logs the user out of his/her account. The main menu for event organizers is shown in Fig. 2c. Unlike musicians, event organizers can access fewer features and are mostly only related to events. Event organizers can make and list new events using the “Create an Event” feature and manage their current active events using the “View Events” feature. Like musicians, event organizers also have “View Profile,” and “Logout” features that function the same way. The “Register” use case is a process that is performed when a new user wants to create an account to be able to start using the app, in the case where the user has not previously created an account. The “Register” button on the landing page

288

A. D. Leander et al.

Fig. 2 a Landing page; b main menu for musicians; c main menu for event organizers

will redirect to the register option, as seen in Fig. 3a. Users are prompted to choose which type of user they are registering as, either a Musician or an Event Organizer. Depending on which option the user selects, the user will be redirected to either the register page for musicians, Fig. 3b, or the registration page for Event Organizers, Fig. 3c. Both register pages will be asking the user to enter their full name, email, and password. The only difference between these two pages is that an extra field is provided for the user to fill in the musical instrument that he/she plays on the registration page for musicians. Once the user has input all required personal information into the provided on-screen text boxes, the system will validate whether all of the personal information entered is already in the correct format and abide by the previously specified constraints. According to the activity diagram in Fig. 4a, once the user has input all required personal information into the provided on-screen text boxes, the system will validate whether all of the personal information entered is already in the correct format abide by the previously specified constraints. If the validation fails, the user will again have to input the appropriate personal information. If the validation result is correct, the user clicks on the “Create your account” button on the screen, and the system will then create the user’s account, store it in the database, and redirect the user to the login page. At the same time, the system will send a confirmation email that is used to activate the account to the linked email address. As seen on the activity diagram shown in Fig. 4b, the “Login” use case represents the user’s process of logging into his/her account. Initially, the user will need to input his/her email and password. The

Mobile Apps for Musician Community in Indonesia

289

Fig. 3 a Register option; b register as musician; c register as event organizer

input data will then be passed to the system and validated appropriately. If the data is valid, the user is logged in and is redirected to the appropriate main menu page, depending on whether the user is a Musician or Event Organizer. The “Login” use case represents the user’s process of logging into his/her account. Initially, the user will need to input his/her email and password. The input data will then be passed to the system and validated appropriately. If the data is valid, the user is logged in and is redirected to the appropriate main menu page, depending on whether the user is a Musician or Event Organizer. Figure 5a represents the page the user will be redirected to when the “Login” button is tapped. The login page prompts users to enter the email address associated with their account and the corresponding password. If the user forgets his/her password, a “Forgot password” feature is also available to reset their password. There is no option to specify whether the user is a Musician or an Event Organizer because the user’s role is stored in the database. The first one is the “Create performance event” use case, which is intended to be used by event organizers. Since it is building an app dedicated to session musicians, specifically to help them discover and share job opportunities and provide use cases regarding event performances. Event organizers will be able to use the app to create and add event listings for the musicians to apply and, hopefully, perform at. Firstly, the event organizer will enter all relevant information about the event according to the required fields on the app. The system will then validate whether all information follows all predefined restrictions. Next, users have the option to upload a file to give more information about the event, such as through posters and

290

A. D. Leander et al.

Fig. 4 a Activity diagram for register use case; b activity diagram for login use case

brochures. If the user does choose to upload a file, the file extension and file size will be validated further to ensure that it is acceptable. The user interface of this feature can be seen in Fig. 5b. The event name, number of performers needed, the role of each performer, performance fee, and details fields are provided for filling in. The details field can enter other relevant details such as date, time, and venue. Event organizers can upload images through the “Upload file” button to provide more information or raise more interest. The “Create event” button will save and create the event listing. The activity diagram for the use case “create performance event” is described in Fig. 6a. Firstly, the event organizer will enter all relevant information about the event according to the required fields on the app. The system will then validate whether all information follows all predefined restrictions. Next, users have the option to upload a file to give more information about the event, such as through posters and

Mobile Apps for Musician Community in Indonesia

291

Fig. 5 a Login; b create event; c manage events

brochures. If the user does choose to upload a file, the file extension and file size will be validated further to ensure that it is acceptable. Once all of the previously mentioned steps are completed, the user can choose to save or discard the event, and the system will handle it accordingly. If the user saves the event, it will be saved to the database. Meanwhile, the “Manage performance event” use case aims to allow each event organizer to manage the events they have created. The event organizers will only be able to view the events that they have made. They can edit each event’s information and view and approve the list of musicians who have applied. The illustration for the manage event page on Fig. 5c shows how the events are arranged in a list with previews of their details, combined with the “Confirm Musicians” button and the “Edit” button. The system will show all the manageable events, and the user selects one. The user can then edit the event information, which will be validated before being saved to the database. The activity diagram in Fig. 6b further describes the “Manage performance event” use case, showing the use case flow. The system will show all the manageable events, and the user selects one. The user can then edit the event information, which will be validated before being saved to the database. The next event-related use case is the “Register event” use case. In the “Register event” use case, session musicians can search through the available listing of event performances and apply to the ones they are interested in. In this use case, session musicians and event organizers can discuss the job, potentially regarding the details, fee, number of musicians required, and many more.

292

A. D. Leander et al.

Fig. 6 a Activity diagram for the create performance event use case, b activity diagram for the manage performance event use case

Mobile Apps for Musician Community in Indonesia

293

Figure 7a demonstrates the activity diagram of the “Register Event” use case. The app will display all available events that session musicians can still register upon starting this feature. Available events are events that have not fulfilled their quotas

Fig. 7 a Activity diagram for the register event use case, b activity diagram for the forum use case

294

A. D. Leander et al.

Fig. 8 a Registered event; b create forum thread; c search for forum

yet. All of the events will be displayed in a list form, as seen in Fig. 8a so that users can search through and select any of them. When a user sees an event that he/she is interested in, all the user has to do is tap on it, and more detailed information surrounding the event will be displayed. If the user decides that he/she would like to perform at this event, the user can easily tap on the “Join” button, and his/her name will be registered to the event. The event organizer will then be able to manually decide whether the session musician is eligible to perform for this event or not. The subsequent use case is the “Forum” use case. As the name suggests, the functionality of this use case is to provide session musicians with a forum where they can discuss different topics and share experiences. Users will be able to create forum threads regarding a specific title and comment on the numerous existing forum threads. Overall, the “Forum” use case performs two main activities. The activity diagram for the “Forum” use case is illustrated in Fig. 7b. Overall, the “Forum” use case performs two main activities. The user can either “Create forum thread” or “Reply to forum thread” to start this activity diagram. The first option, “Create forum thread,” means that the user would like to create a new forum thread that did not exist before. The user will initially be prompted to choose a forum category, and the system will respond with a page for creating new forum threads. The user is then asked to input the thread title and the initial forum message. Users also have the option to upload an image. As usual, the image’s file extension and size are validated. After this is done, if the user chooses to save the forum thread, it will be saved to the database.

Mobile Apps for Musician Community in Indonesia

295

The second option is the “Reply to forum thread” path. The system will display a message box for the user to fill in this activity. The user can set the title of the reply and write whatever message the user wants to write. Again, he/she can also upload images to support their reply. Once the user is satisfied with the reply, he/she clicks on the “Reply” button, and the reply is saved to the database. The page will automatically refresh to display the newly created thread reply. Figure 8b illustrates the “Create Forum Thread” feature. The user must fill in the thread title and the original comment message. Once all of the details are as desired, the users can tap on the “Create Thread” button to generate the forum thread. Users also can upload images to include in their forum threads. Users will be able to search for specific forum threads using the “Search for Forum” option, as seen in Fig. 8c. Initially, the search page will display all forum threads available to read. A search bar is present for users to type in keywords of the forums they are looking for. The system will then search for those particular keywords and display the appropriate results. Users can just tap on their desired forum threads to read more about them. It also proposes an “Information” use case representing the functionality that allows users to search and view various information regarding the session musician’s community. Users will be able to explore knowledge surrounding session musicians, such as guides on how to become better musicians, performance tips, instruments tips and tricks, etc. This feature aims to educate users more about the field they are dealing with and broaden their knowledge about the music industry. In Fig. 9a, the “Information” use case is elaborated in the form of an activity diagram. First and foremost, the system will fetch all the information articles from the database and display them in a grid-like format. Users can then view all of the article previews and select any of them they want to read. When the user selects an article, the app will redirect the user to a page that shows the complete article and its images or external links. Figure 10a represents the “Information” feature. Users will be able to read articles and guides about the session musician community by simply tapping on any one of the articles, which will redirect to the article’s page. Meanwhile, the “Rate and comment event” use case, as seen in Fig. 10b, is the musician’s process of rating and commenting on events that they have performed at. This way, it can receive feedback about which events and event organizers are favored by musicians and which event organizers should be blacklisted and rejected in the future. The flow of this use case can be represented using the activity diagram in Fig. 9b, and the user interface in Fig. 10b. The user will be able to rate the event on a scale of 1–5 and also has the option to write additional comments. The comments will be validated before it is saved into the database. For the class diagram, there are eight classes, as seen in Fig. 11, and the first class is the “Musician” class which represents the session musicians that will use this app. This class provides the session musicians’ information such as their name, email, gender, musical instrument, and private information such as their account password. Each musician’s account also has its own unique “userId,” which is the primary key of this class. Musicians can also provide several ratings and comments. Any musician can post zero to many comments, register for zero to many events, and create zero to many forum threads.

296

A. D. Leander et al.

Fig. 9 a Activity diagram for the information use case, b activity diagram for the rate and comment event use case

The “ForumThread” class is the representation of the forum. There will be three attributes in this class, which are “forumId,” “category,” and “DateTime.” The attribute “category” is used to categorize the forum to ease any musician’s process to find a forum thread of a specific topic. It found that categorized forum threads are more convenient than uncategorized forum threads. A single forum thread must have

Mobile Apps for Musician Community in Indonesia

297

Fig. 10 a User interface for information; b user interface for rate and comment event

at least one comment posted, but there is no maximum limit to how many comments can be posted. A single user only creates each forum thread. The next class is “Comment,” representing each comment in a thread. In this “Comment” class, it stores the “userId” and “forumId,” which represent any user that posts a comment in a forum thread. Other than those two attributes, two attributes will store the comment title and comment description, and “DateTime” to store the date and time that the comment was posted. A single musician can only write each comment, and each comment can only be written inside one forum thread. There is another type of user in the app, other than the session musicians, and the event organizer. The “EventOrganizer” class is the representation of this type of user. This class has the same attributes as the musicians, except removing the “musical instrument” attribute and the app’s event organizers’ role is to create events for the session musicians to participate in and perform in. An event organizer can create many events where the bare minimum is at least one event. This class also acts as the connector of session musicians and event organizers. Thus, it has another class called “Event” to store the data of the event’s name, event’s ID, organizer’s ID, and several performers or musicians. This number will represent the quota of musicians needed for the specific event. An event can only be organized by a single event organizer but can be joined by one or more musicians. Each event

298

A. D. Leander et al.

Fig. 11 Class diagram of mobile apps for session musicians community in Indonesia

can also only have one set of event details. Each event can also have multiple ratings and comments by multiple musicians. The class “EventDetails” is only associated with the “Event” class. The class “EventDetails” stores the event’s details, for example, the description of the event itself, the role of the performers or musicians, and the fee for the musicians. Each event detail can only be specified to a single event, as every event has different details. The “RegisterEvent” class is used to model the transactions between musicians and events, which are performance agreements since no actual money-related transactions can be made using the app. It will store the associated “userId,” “eventId,” and “DateTime.” One musician can only make each event registration for one event. Lastly, the “RatingComment” class represents musicians’ ratings and comments on certain events. The class contains the associated “userId” and “eventId” and supporting attributes to store ratings and comments, which are named the same. Each rating and comment is associated with a single event and made by a single musician.

Mobile Apps for Musician Community in Indonesia

299

4 Conclusion It is proposing a mobile app for session musicians that hopefully can be a platform to connect session musicians and event organizers. Session musicians can browse available events to look for job opportunities, while event organizers can create events in the app and get session musicians to perform for their events. An explorative was provided yet straightforward user interface that will attract session musicians. Several additional features can aid session musicians in fields other than job searching, such as a forum thread where they can discuss and share their ideas and an information section where there will be much valuable information such as tips and tricks, articles, and many more. More features will be added to enhance the app to be more supportive toward session musicians for the plans. These features include chat or direct messages between session musicians, a social network where they can connect and showcase their music and skills, in-app payment methods for event performances, and many more. In conclusion, hopefully, this app can answer session musicians’ needs and help them provide for their families.

References 1. M.J. Khadavi, Dekonstruksi Musik Pop Indonesia dalam Perspektif Industri Budaya. J. Humanit. Univ. Muhammadiyah Malang 9(2), 47–56 (2014) 2. A. Bagaskara, Menegosiasi Otentitas: Kancah Musik Independen Indonesia dalam Konteks Komodifikasi oleh Perusahaan Rokok. MASYARAKAT J. Sosiol. 22(2), 235–255 (2017) 3. J. Herbst, T. Albrecht, The skillset of professional studio musicians in the German popular music recording industry. J. Ethnomusicol. 30, 121–153 (2018) 4. I. Campelo, That extra thing—the role of session musicians in the recording industry. J. Art Rec. Prod. 10 (2015) 5. J. Herbst, T. Albrecht, The work realities of professional studio musicians in the German popular music recording industry: careers, practices, and economic situations. J. Int. Assoc. Study Pop. Music 8(2), 18–37 (2018) 6. B. Wright, Reconstructing the history of Motown session musicians: the Carol Kaye/James Jamerson controversy. J. Soc. Am. Music 13(1), 78–109 (2019) 7. W. Wiflihani, Fungsi Seni Musik dalam Kehidupan Manusia. ANTHROPOS J. Antropol. Sos. Budaya 2(1), 101–107 (2016) 8. A. Kusumawardhani, Membangun Musik Indonesia Melalui Budaya Berbagi. J. Ilmu Komun. 11(2), 121–134 (2014) 9. A. Roffiq, I. Qiram, G. Rubiono, Media Musik dan Lagu Pada Proses Pembelajaran. J. Pendidik. Dasar Indones. 2(2), 35–40 (2017) 10. C. Aryanto, S. Hartono, Perbandingan Subjective Well-Being Musisi dan Non-Musisi. J. Ilmiah Psikol MIND SET 6(1), 1–13 (2014) 11. H.K. Flora, X. Wang, S.V. Chande, An investigation into mobile application development processes: challenges and best practices. Int. J. Mod. Educ. Comput. Sci. 1–9 (2014) 12. L. Chandi, C. Silva, D. Martinez, T. Gualotuna, Mobile application development process: a practical experience, in 12th Iberian Conference on Information Systems and Technologies (CISTI), Lisbon, Portugal, June 2017 13. T. Vithani, A. Kumar, Modeling the mobile application development lifecycle, in International MultiConference of Engineers and Computer Scientists (IMECS), Hong Kong, Mar 2014

300

A. D. Leander et al.

14. N. Kumar, K. Krishna, R. Manjula, Challenges and best practices in mobile application development. Imp. J. Interdiscip. Res. 2(2), 1607–1611 (2016) 15. N. Shamat, S. Sulaiman, J. Sinpang, A systematic literature review on user interface design for web applications. J. Telecommun. Electr. Comput. Eng. 9(3–4), 57–61 (2016) 16. D. Saha, A. Mandal, User interface design issues for easy and efficient human computer interaction: an exploratory approach. Int. J. Comput. Sci. Eng. 3(1), 127–135 (2015) 17. S. Kocakoyun, Developing of Android mobile application using Java and Eclipse: an application. Int. J. Econ. Mech. Mechatron. Eng. 7(1), 1335–1354 (2017) 18. K. Amano, S. Tsuzuku, K. Suzuki, N. Hiraoka, Learning together for mastery by using a discussion forum, in 2019 International Symposium on Educational Technology (ISET), Hong Kong, July 2019 19. A. Ezen-Can, S. Kellog, K.E. Boyer, S. Booth, Unsupervised modeling for understanding MOOC discussion forums: a learning analytics approach, in 5th International Conference on Learning Analytics and Knowledge, Poughkeepsie, New York, USA, Mar 2015, pp. 146–150 20. J. McDougall, The quest of authenticity: a study of an online discussion forum and the needs of adult learners. Aust. J. Adult Learn. 55(1), 94–113 (2015) 21. M.G. Alzahrani, The effect of using online discussion forums on students’ learning. Turk. Online J. Educ. Technol. 16(1), 164–176 (2017) 22. B. Vivekanandam, Design an adaptive hybrid approach for genetic algorithm to detect effective malware detection in Android division. J. Ubiquitous Comput. Commun. Technol. 3(2), 135– 149 (2021) 23. S.R. Mugunthan, T. Vijayakumar, Design of improved version of sigmoidal function with biases for classification task in ELM domain. J. Soft Comput. Paradigm (JSCP) 3(02), 70–82 (2021) 24. S. Chowdhury, Exclusion by design: a systematic assessment of community engagement in COVID-19 mobile apps. Doctoral dissertation, Icahn School of Medicine, Mount Sinai, 2022

A Genetic-Based Virtual Machine Placement Algorithm for Cloud Datacenter C. Pandiselvi and S. Sivakumar

Abstract Cloud computing is a process of renting hardware, software, and platforms over the internet. Users are charged on a pay-per-use model for these services. A cloud service provider provides resources on the basis of user’s request. To efficiently allocate resources, server consolidation techniques are used. Virtual machine placement (VMP) is a server consolidation technique in which virtual machines are created and mapped to physical servers for allocation of resources. This research work explores the first fit, best fit, and genetic algorithm for mapping physical servers to virtual machines. For an optimum mapping of VMP, a hybridization of best-fit algorithm and genetic algorithm suggested the improved genetic algorithm (IGA). The performance of these VMP algorithms is evaluated with datasets obtained from Microsoft Azure, Google Cloud, and Amazon EC2. The simulation results showed that the proposed IGA algorithm improves resource utilization and decreases the execution time. Keywords Virtual machine placement · Genetic algorithm · Best fit · First fit

1 Introduction Cloud computing allows users to share a pool of resources based on their needs. Virtualization technologies enable live migration of virtual machines (VM), which allows VM resources such as CPU and memory to be freely moved between physical servers (PS) [1]. Server consolidation technique is used to place several VM on a single PS, allowing the PS to operate at maximum resource efficiency. Server consolidation techniques requires four major steps namely (i) PS overload detection,

C. Pandiselvi (B) Department of Computer Science, Cardamom Planters’ Association College, Bodinayakanur, India e-mail: [email protected] S. Sivakumar Cardamom Planters’ Association College, Bodinayakanur, India © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 I. J. Jacob et al. (eds.), Expert Clouds and Applications, Lecture Notes in Networks and Systems 444, https://doi.org/10.1007/978-981-19-2500-9_21

301

302

C. Pandiselvi and S. Sivakumar

(ii) PS underload detection, (iii) VM selection and migration, (iv) VM placement [2]. Virtual machine placement (VMP) is the most important aspect of server consolidation. The VMP is the process of mapping, which VM should be mapped to each PS. The problem of VMP is to place VM resources reasonably according to the processing capacity of each PS. An improper VMP mapping can cause resources to be performed on unsuitable PS, resulting in unintended consequences. This might damage the cloud service provider’s reputation. To avoid this, an efficient technique should be utilized to accurately allocate resources to the PS that will allow their efficient performance. A genetic algorithm (GA) is a form of random searching with improved optimization and internal implicit parallelism. It can obtain and instruct the optimized seeking space, as well as automatically alter the seeking direction [3]. The GA approach computes the VMP in advance that it will have on the system after the new VM resource is deployed in the system based on historical data and current conditions of the system. It then selects the solution with the least amount of impact on the system. This results in better server consolidation and a reduction in the number of dynamic virtual machines. With the advantages of GA, this paper presents a genetic algorithm and an improved genetic algorithm (IGA) for VMP in cloud computing environment. IGA is a used to determine an optimized solution for the problem based on crossover, mutation, and selection. The IGA starts its search by looking at an arbitrary selection of solutions. Every solution is given by the fitness function that is evaluated by best-fit algorithm. Thereafter, three operators similar to natural genetic operators such as crossover, mutation, and selection are used to change the population of solutions to a new population. It works iteratively, by applying these three operators in sequence in each generation until a termination requirement is satisfied. The effectiveness of the IGA algorithm performance is evaluated using Microsoft Azure, Google Cloud, and Amazon EC2 datasets. Microsoft Azure features an instance kind with one or more instance sizes that can be scaled to meet the needs of a specific workload. Google Cloud has a Google’s infrastructure, which is used to generate and run-on virtual machines. Amazon EC2 provides a wide selection of instance types optimized to fit different use cases. Premature convergence and a long execution time are the limitations of genetic algorithms [4]. Premature convergence causes many genetic algorithms to converge on sub-optimal solutions. The long execution time is another important major problem. The majority of genetic methods take a long time to process multiple generations before getting the best result. As a result, IGA is a hybrid of best fit and genetic algorithm, with the goal of overcoming the constraints of IGA to the placement problem. The following are some of the work’s major contributions: (i)

The performance of first fit and best fit techniques is evaluated to find the fitness value of VMP problem with heterogeneous datasets.

A Genetic-Based Virtual Machine Placement Algorithm …

(ii)

303

A novel placement idea, an improved genetic algorithm (IGA) is proposed, and compared their performance.

2 Related Work Traditional VM placement algorithms in cloud data centers rely solely on the present state of the overall system. They fail to overlook system variability and historical behavioral data, resulting in system overloading and load unbalancing. However, the GA computes the impact on both system overloading and unbalancing. Kumar and Smys [5], formulate the ubiquitous service for medical care utilizing the cloud. The blockchain technology is used to safeguard the reliability of the balanced data. Andi [6], presented the serverless architecture enhances another layer to the cloud computing model while the management of servers is abstracted from the developers. This provides two services such as Backend as a Service and Function as a Service. Shakya [7], introduced a data security analysis and solution for privacy protection framework during data migration. The approach provides a strict separation between sensitive and non-sensitive data and provides encryption for the sensitive data. Bhalaji [8], proposed accurate prediction of the workload and sequence of resources along with time series. Goncalves and Resende [9], proposed a biased random key genetic algorithm for 2d and 3d bin-packing problems and virtualization management in datacenters. Sonklin et al. [10], presented an improved genetic algorithm for the virtual machine placement and the VM policy that has been employed in VM consolidation. Lu et al. [11], explored a genetic algorithm and some of its modified versions are discussed for optimal machine placement based on improved genetic algorithm in cloud. Belgacem et al. [12], investigates a virtual machine placement approach based on the micro genetic algorithm in cloud computing. Pandiselvi and Sivakumar [13], proposed a particle swarm optimization bin-packing algorithm to utilize resources and energy consumption. The proposed algorithm explained heuristic method to address virtual machine placement problem. Moreover, the sizes of working set are considered while placing applications on physical machines. Rashida et al. [14], proposed a genetic algorithm based on memetic grouping has the objective of cost efficiency and energy-efficient VM placement in multi-cloud environment. The fitness function determines the competence of the GA. In GA, the fitness function is discovered to be a problem-dependent and critical criterion for getting optimal results. When defining the fitness function for a problem is difficult, a simulation can be utilized to determine the fitness function value of a genetic algorithm. Kour et al. proposed an application of some fitness functions in various fields is investigated. The fitness function is critical in applications because the success of a genetic algorithm is determined by how effective the fitness function is. If the fitness function is poorly designed, the genetic algorithm’s later operations will provide an optimized output [15]. Theja and Babu presented a fitness-based adaptive

304

C. Pandiselvi and S. Sivakumar

evolutionary method for VM policy that has been employed in VM consolidation [16]. Xu et al. explored to maximize resource utilization, every PM on the cloud computing platform focused on multidimensional resource load balancing. The approach customizes ant colony optimization in the framework of virtual machine sharing to avoid early convergence or slipping into limited optima [17]. Li et al. investigates a genetic algorithm with a novel heuristic packing technique that translates a genetic material box filling and container loading sequences into a compressed filling solution [18]. Pandiselvi and Sivakumar used to find the best virtual machine placement, a bin-packing technique analyzed with four different fitness strategies [19]. To better understand the behavior of the fitness function, a review of numerous research articles was conducted, and it was discovered that various fitness values based on genetic algorithms have been examined for diverse applications. The GA algorithm is compared with best fit, first fit, worst fit, best fit decreasing, and first-fit decreasing placement algorithms are tabulated as shown in Table 1. Table 1 Related works based on genetic algorithm Author

Algorithm

Based on

Resources considered

Objective

Performance better than

Kour et al. [15]

Fitness functions with different domains

Genetic

CPU

Resource utilization, minimizing cost and time

Best fit, first-fit algorithm

Theja and Babu [16]

Adaptive genetic algorithm (A-GA)

Genetic

CPU

VM reuse strategy

Best-fit decreasing algorithm

Xu et al. [17]

Ant colony optimization

Genetic

CPU, memory, and storage

SLA violation First fit decreasing, worst fit algorithm

Li et al. [18]

Packing heuristic procedure

Genetic

CPU, memory

High consumption time

Greedy algorithm, first-fit algorithm

Pandiselvi and Sivakumar [19]

Particle swarm optimization bin packing

Genetic

CPU

Energy consumption, resource utilization

Best fit, first fit

A Genetic-Based Virtual Machine Placement Algorithm …

305

3 Placement Algorithms Placement algorithms are one of the key mechanisms in data centers for designing an efficient server consolidation in cloud. The principle is based on the process of selecting the most suitable PS among the VM. So, the placement algorithm aims at determining the most optimal VM to PS. The mapping can achieve both the increase in the resource utilization and decrease in the PS overloading. The placement algorithms are namely first fit, best fit and standard genetic algorithms are used to achieve the minimized execution time and maximize the resource utilization. Resource utilization problem: Consider a cloud data center that provides the limited number of PS with resources like CPU capacity, Memory in GB since there are a greater number of user’s requests for resources, the cloud service provider must decide on how many numbers of PS can be allocated to VM based on cloud user’s request. By virtualization technology, the cloud data centers have its unlimited number of VM that is used to deploy PS for maximizing resource utilization. To formulate a resource utilization in VMP problem, a set of terminus servers, S, with the capacity of resources, i.e., CPU or memory and set of user requests to be allocated, P, are given. The destination servers are numbered as 1, 2, 3, … The number of users requested services is m, which are numbered as 1, 2, 3, … Then to find the minimum PS can be modulated are follows, min

n 

PSi

(1)

i=1

Subject to: n 

X i j = 1 j = 1, 2 . . . m

(2)

i=1

PSi ∗ PScpui ≥

m 

∗Scpu j ∗ X i j

(3)

∗Smem j ∗ X i j

(4)

j=1

PSi ∗ PSmemi ≥

m  j=1

where, • In Eq. (1) PSi specifies whether the ith server is being used or not, i.e., PSi = 1 specifies the ith server is being used, otherwise, it is not being used. • In Eq. (2) X i j specifies the allocation of resources to servers, i.e., X i j = 1 specifies the resource j is allocated to the ith server, otherwise, the resource j is not allocated to the ith server.

306

C. Pandiselvi and S. Sivakumar

• In Eq. (3) PScpui shows the CPU resource provided by the ith server, while Scpu j shows the CPU resources needed by the resource j. • In Eq. (4) PSmemi shows the memory resource provided by the ith server, while Smem j shows the memory resource needed by the resource j. First fit (FF): The FF algorithm considers the first user request for resources in an available PS with capacity more than or equal to its size. Each resource is allocated to the lowest initialized PS into which it fits. FF method does not search for appropriate size, but just allocates the resources in the nearest capacity available with sufficient size. The execution time for allocating all resources is added. Best-fit (BF): The BF algorithm looks for free PS over the full list. It looks for a PS that is close to the real PS size that is required. The free list of PS is kept in this approach in order of least to largest size. The execution time for placing all PS on that machine is then updated using the algorithm’s computation of execution time. Table 2 lists 12 various VM instances of dataset prepared for general, compute, memory, and storage purposes in this research. For each situation, there are four different types of virtual machines. Amazon EC2, Microsoft Azure, and Google Cloud are some of the solutions. Amazon EC2 [20], provides a wide selection of instance types optimized to fit different use cases. Instance types comprise varying combinations of CPU, memory, storage, and networking capacity and offer the flexibility to choose the appropriate mix of resources for any application. Each instance type includes one or more instance sizes, allowing the applications to scale your resources to the requirements of your target workload. Microsoft Azure [21] Improve the accuracy of cloud infrastructure models with publicly available datasets. Save time on data discovery and preparation by using curated datasets that are ready to use in machine learning workflows also and easy to access from Azure services. Google Cloud instances [22] are ideal for compute-bound applications that benefit Table 2 VM instance dataset

Cloud service provider Microsoft Azure

Google Cloud

Amazon EC2

VM instances

CPU capacity

Memory in GB

D1v2/1

1

3.5

D2V2/2

2

7

D3v2/3

4

14

D4v2/4

8

28

G1-STD-1

1

3.75

G1-STD-2

2

7

G1-STD-4

4

15

G1-STD-8

8

30

EC2.small.1

1

2

EC2.large.2

2

8

EC2.xlarge.3

4

16

EC2.2xlarge.4

8

32

A Genetic-Based Virtual Machine Placement Algorithm …

307

from high-performance processors. Instances belonging to this family are well suited for batch processing workloads, media transcoding, high-performance web servers, high-performance computing (HPC), scientific modeling, dedicated gaming servers, ad server engines, machine learning inference, and other compute-intensive applications. Genetic Algorithm (GA) The GA could be seen as an intelligent probabilistic search in the space of solutions for a hard problem. Starting from the name itself, the terminology of the GA is derived from the evolutionary biology, where individuals from a population by a recombination of genetic characteristics of their parents, plus a small probability of some random genetic mutation. Population initialization, fitness function computation, selection, crossover, and mutation are the steps for solving GA. Population Initialization The initial population is the first step in the GA algorithm. In algorithms, population initialization is critical since it can affect the speed of convergence as well as the quality of the final solution. To make it appropriate for genetic operations, each solution in the population is referred to as an individual, and each individual is represented as a chromosome. Individuals are chosen from the initial population and some operations are performed on them to create the following generation. Fitness Calculation A fitness function is used to measure the quality of the chromosome in the population according to the given optimization objective. The fitness value is defined by getting the chromosome. Consider the goal function given by Eq. (2) to minimize the number of active PS. Calculating fitness values for each chromosome is a difficult challenge as well, hence a novel method called GA is utilized to get the fittest value. Selection In GA, the next phase is selection. Following the calculation of the fitness value for each chromosome, selection procedures are used to choose from the population. Roulette Wheel Selection, tournament selection, and so on are examples of common selection processes. To avoid the population falling into local convergence and degradation, avoid tournament selection, which may result in a local solution. Rather, a chosen technique based on fitness value order is utilized. Each chromosome’s fitness value can be determined first, then sequenced from high to low, with the high fitness values being handed over to the next generation. It must ensure that no duplicate chromosomes are found throughout this operation. Simultaneously, maintaining a tiny part of the poorest solution, which may help avoid local convergence. Crossover The goal of a crossover, also known as recombination, is to produce offspring from two parents P with as much relevant information from both parents as possible. A

308

C. Pandiselvi and S. Sivakumar

crossover operator is used to combine the genetic information of two parents and create new offspring. Crossover operators include single-point crossover, double point crossover, partial crossover, sequential crossover, and so on. Mutation Mutation is a process in which a portion of a gene is arbitrarily transformed. The mutation operator can be used to replicate the great chromosomes that are deleted during the selection or crossover stages. The mutation operator also ensures that the probability of finding any point in the problem space is never zero, regardless of the dispersion of the initial population. Assume that the mutation operator includes moving a chromosome to another chromosome with a chance of picking two chromosomes at random and swapping genes. To avoid chromosomal visits, genes are swapped in the same chromosome to avoid recurrence. To carry out this mutation, two chromosomes from the array containing the permutation are chosen at random and their genes are swapped. Algorithm 1: Procedure of GA 1. Begin 2. Initialize Population 3. Evaluate fitness value by objective function 4. Repeat till (termination condition occurs) 5. Perform a. Chromosome selection b. Parent crossover c. Offspring mutation d. new chromosome evaluation e. Select chromosome for future generation 6. End Algorithm 1: GA procedure is used to analyze the efficient placement of VM, at step 2: the procedure starts with initial population, i.e., the number of PS and the number of VM with CPU capacity. In step 3: a fitness value is used to measure, according to the given optimization objective. In step 5 a. the two-fitness value is selected, in step 5 b. the crossover operator is used to swap between two-fitness values, and in step 5 c. the new offspring is produced. In steps 5 d. and 5 e. the produced offspring are interchanged to create a new fitness value and the PS is placed in that fitness value. Improved Genetic Algorithm (IGA) The genetic algorithm is a metaheuristic algorithm based on theory of evolution. It is a randomized algorithm in which random changes are applied to the current solution to find the new solution. The IGA is used to analyze the fitness value as a best-fit value using the best-fit algorithm, the procedure of IGA is used to analyze the efficient resource utilization. The fitness value is one of the pivotal parts of the algorithm is trying the optimize in IGA algorithm. IGA begins generating problem

A Genetic-Based Virtual Machine Placement Algorithm …

309

Fig. 1 Flowchart of IGA

solutions as a population. This population then undergoes an evaluation of fitness value using the best-fit algorithm. Each iteration includes the following processes: (i) selection, (ii) crossover and (iii) mutation, in which evolution data such as the fitness value are updated shown in Fig. 1. Algorithm 2: Procedure of IGA 1. Begin 2. Initialize Population 3. Evaluate fitness value by best-fit algorithm 4. Repeat till (termination condition occurs) 5. Perform a. Chromosome selection b. Parent crossover c. Offspring mutation d. new chromosome evaluation e. Select chromosome for future generation 6. End

Algorithm 2: PGA procedure starts with initializing a number of PS and number of VM with CPU capacity. In step 3, the fitness value is calculated using best-fit algorithm, in which the VM searches for small sufficient PS among the free available PS and starts by picking the VM and find the minimum PS that can be assigned to

310

C. Pandiselvi and S. Sivakumar

current VM, then the best-fit value is selected at step 5 a., the selection process is used to select two best values at random. In step 5 b. the crossover operator is used to swap between two-fitness values and then offspring is produced. In step 5 c., the produced offspring is interchanged to create a new best-fit value and PS is placed in that best-fit value more efficiently.

4 Performance Evaluation 4.1 Experimental Setup NetBeans IDE is an open-source (https://en.wikipedia.org/wiki/Open-source_sof tware) integrated development environment. The language-aware NetBeans IDE editor detects errors and assists the documentation popups and smart code completion with the speed and simplicity of text editor. Of course, the Java editor in NetBeans is much more than a text editor—it indents lines, matches words and brackets, and highlights source code syntactically and semantically. Therefore, the NetBeans IDE is used to implement IGA. The performance of the IGA is evaluated in terms of execution time and resource utilization. Because access to real data centers is challenging, a simulation-based evaluation is used to evaluate the proposed algorithm’s performance to existing works that are currently used by the majority of cloud service providers. A variety of PS are included in the simulated cloud environment, as well as a number of randomly created VM resource demands, such as CPU capacity and memory.

4.2 Evaluation of FF and BF Algorithm The performance of FF and BF algorithms is evaluated to find the fitness value. The computation of execution time of the algorithm is used as a fitness criterion for checking the fitness value of the algorithms. The simulation processing time to place all VM is referred to as execution time. In this situation, the better strategy will be stated as the average minimal execution time, because it completes all of the VM placement processes in the data center. The experiments are done to identify the effect of applying the fitness value to the IGA algorithm so that the two placement algorithms FF and BF are examined by using three VM instances datasets shown in Table 2. In this scenario, from Table 3 the computation of execution time of the FF and BF algorithms with physical servers namely PS1, PS2, PS3, and PS4 is compared and shown as the BF algorithm gives better results compared with FF algorithm. The BF algorithm is more suitable for handling VM requests efficiently in the IGA algorithm.

A Genetic-Based Virtual Machine Placement Algorithm …

311

Table 3 Execution time for BF and FF algorithm for different datasets Execution time (in s) Azure

Google

BF

FF

Amazon EC2

BF

FF

BF

FF

PS1

3.059

4.090

3.005

4.015

3.000

4.003

PS2

3.012

4.016

3.015

4.015

3.003

4.015

PS3

3.015

4.015

3.003

4.015

3.014

4.015

PS4

3.014

4.014

3.001

4.001

3.016

Total

12.10

16.10

3.025

Average

4.025

12.02

16.04

3.005

12.03

4.010

3.000

4.014 16.04 4.010

Execuon me (Sec)

Figure 2 explores the average computation execution time of BF algorithm as for Azure dataset 3.025 s, Google dataset 3.005 s, and Amazon dataset 3 s and then Fig. 3 explores the average computation execution time of FF algorithm as Azure dataset 4.025 s, Google dataset 4.01 s and Amazon EC2 dataset 4.01 s. The comparison of average calculation execution time clearly shows that the BF algorithm using Amazon EC2 dataset performs significantly better than the FF approach. To obtain an efficient fitness value, the BF algorithm is applied in Algorithm 2: IGA method. 3.03 3.02 3.01 3.025

3

3.005

2.99

3

2.98 Azure

Google

Amazon EC2

VM instance types

Execuon me (Sec)

Fig. 2 Average execution time for best-fit algorithm

4.03 4.025 4.02 4.015 4.01 4.005 4

4.025 4.01 Azure Google VM instance types

Fig. 3 Average execution time for first-fit algorithm

4.01 Amazon EC2

312

C. Pandiselvi and S. Sivakumar

Number of VM Placed

12 10 8 6 PS in Idle

4

VM Placed

2 0 1

2

3

4

5

6

7

8

9

10

11

12

Number of PS Used

Fig. 4 Placement of VM by IGA

4.3 Evaluation of IGA Algorithm The parameter configuration is critical for evaluating the effectiveness of the IGA algorithm. As a result, the algorithms were run five times in the same instance to achieve relevant results. The parameters used in the effective IGA are determined empirically in order to produce a satisfying solution in an acceptable length of time. As a result, 100 iterations is the maximum number of iterations. To validate the IGA’s efficiency, it is tested utilizing the three cloud providers listed in Table 2: Amazon EC2, Microsoft Azure, and Google Cloud. The performance of IGA algorithm is evaluated by the resource utilization and overall execution time of the algorithm. The resource utilization and VM placement for both the IGA and GA algorithm are separately analyzed. The algorithms take 12 sets of physical servers from VM instances dataset taken from Table 2. The IGA algorithm has placed the CPU capacity of VM in PS and other PS are put in idle are shown in Fig. 4. Therefore, the GA algorithm has placed the CPU capacity of VM in PS and other PS are put in idle are shown in Fig. 5. In the IGA algorithm, the number of VM placed in PS is minimized when compared to GA shows the efficient placement and resource utilization.

Number of VM Placed

12 10 8 6

PS in Idle

4

VM Placed

2 0

1

2

3

4

5

6

7

Number of PS Used

Fig. 5 Placement of VM by GA

8

9

10

11

12

A Genetic-Based Virtual Machine Placement Algorithm …

313

Table 4 Simulation result of IGA and GA Algorithm type

Execution status

No of VM instances used

No of VM placed in PS

PS in idle

Execution time in s

IGA

Success

12

03

09

5.005

GA

Success

12

04

08

5.012

Execution Time (ms)

6000 5000 4000

IGA

3000

GA

2000 1000 20

40

60

80

100

Number of Iterations Fig. 6 Comparison of execution time for IGA and GA algorithm

Table 4 displays the overall execution time of two algorithms: the IGA and GA algorithms. The table shows that the IGA algorithm has three number of PS placed and nine number of PS are put in idle where the resources are utilized. The execution time of IGA algorithm is also minimized. As a result, when compared to the existing GA, the IGA uses fewer PS, has a faster response time, has a higher resource utilization rate, and consumes less power. In Fig. 6: the simulation execution time to place the VM as execution time of the algorithm is compared and shown that IGA algorithm has minimum execution time compared with GA algorithm. Therefore, the IGA algorithm performs better than GA algorithm.

5 Conclusion One of the most important difficulties in cloud computing is virtual machine placement. A suitable virtual machine placement is necessary for good system performance. Three placement methods are discussed in this paper: best fit, first fit, and genetic algorithm, as well as a new placement algorithm. The experimental results reveal that the proposed an improved genetic algorithm performs better than the genetic algorithm in terms of execution time and resource utilization. As a result, the proposed genetic algorithm outperforms the traditional genetic algorithm. This strategy can be implemented in existing cloud computing systems

314

C. Pandiselvi and S. Sivakumar

to improve resource efficiency while reducing physical server overloading and algorithm execution time.

References 1. P. Srivastava, R. Khan, A review paper on cloud computing. Int. J. Adv. Res. Comput. Sci. Softw. Eng. 8(6) (2018). ISSN: 2277-128X 2. C. Pandiselvi, S. Sivakumar, A review of virtual machine algorithm in cloud data centre for server consolidation. IJERCSE 5(3), 182–188 (2018) 3. M. Chen, M. Li, F. Cai, A model of scheduling optimizing for cloud computing resource services based on buffer-pool agent, in 2010 IEEE International Conference on Granular Computing (GrC), Aug 2010 4. N. Avinash Kumar Sharma, A multi objective genetic algorithm for virtual machine placement in cloud computing. Int. J. Innov. Technol. Explor. Eng. (IJITEE) 8(8) (2019). ISSN: 2278-3075 5. D. Kumar, S. Smys, Enhancing security mechanisms for healthcare informatics using ubiquitous cloud. J. Ubiquitous Comput. Commun. Technol. 2(1), 19–28 6. H.K. Andi, Analysis of serverless computing techniques in cloud software framework. J. IoT Soc. Mob. Anal. Cloud 3(3), 221–234 (2021) 7. S. Shakya, An efficient security framework for data migration in a cloud computing environment. J. Artif. Intell. 1(01), 45–53 (2019) 8. N. Bhalaji, Cloud load estimation with deep logarithmic network for workload and time series optimization. J. Soft Comput. Paradigm 3(3), 234–248 (2021) 9. J.F. Goncalves, M.G. Resende, A biased random key genetic algorithm for 2D and 3D bin packing problems. Int. J. Prod. Econ. 145(2), 500–510 (2013) 10. C. Sonklin et al., An improved genetic algorithm for the virtual machine placement problem. Aust. J. Intell. Inf. Process. Syst. 16(1), 73–80 (2019) 11. J. Lu et al., Optimal machine placement based on improved genetic algorithm in cloud computing. J. Supercomput. (2021). https://doi.org/10.1007/s11227-021-03953 12. A. Belgacem et al., New virtual machine placement approach based on the micro genetic algorithm in cloud computing, pp. 66–72 (2021) 13. C. Pandiselvi, S. Sivakumar, Constraint programming approach based virtual machine placement algorithm for server consolidation in cloud data center. IJCSE 6(8) (2018). E-ISSN: 2347-2693 14. S.Y. Rashida et al., A memetic grouping genetic algorithm for cost efficient VM placement in multi-cloud environment. Clust. Comput. https://doi.org/10.1007/s10586-019-02956-8,2019. 1234567 15. H. Kour, P. Sharma, P. Abrol, Analysis of fitness function in genetic algorithms. J. Sci. Tech. Adv. 1(3), 87–89 (2015) 16. P.R. Theja, S.K.K. Babu, An adaptive genetic algorithm based robust QoS oriented green computing scheme for VM consolidation in large scale cloud infrastructures. J. Sci. Technol. (2014). https://doi.org/10.17485/ijst/2015/v8i27/79175 17. P. Xu, G. He, Z. Li, Z. Zhang, An efficient load balancing algorithm for virtual machine allocation based on ant colony optimization. Int. J. Distrib. Sens. Netw. 14(12) (2018). https:// doi.org/10.1177/1550147718793799 18. X. Li, Z. Zhao, K. Zhang, A genetic algorithm for the three-dimensional bin packing problem with heterogeneous bins, in Proceedings of the Industrial and System Engineering Research Conference (2014) 19. C. Pandiselvi, S. Sivakumar, Performance of particle swarm optimization bin packing algorithm for dynamic virtual machine placement for the consolidation of cloud server. IOP Conf. Ser. Mater. Sci. Eng. 1110 (2021). https://doi.org/10.1088/1757X/1110/1/012007

A Genetic-Based Virtual Machine Placement Algorithm … 20. Amazon EC2, https://aws.amazon.com/ec2/instance-types 21. Microsoft Azure, https://docs.microsoft.com/en-in/azure/open-datasets/dataset-catalog 22. Google Cloud, https://console.cloud.google.com/marketplace

315

Visual Attention-Based Optic Disc Detection System Using Machine Learning Algorithms A. Geetha Devi, N. Krishnamoorthy, Karim Ishtiaque Ahmed, Syed Imran Patel, Imran Khan, and Rabinarayan Satpathy

Abstract Computer-aided diagnosis relies heavily on the accurate localization of the optic disc (OD) prior to OD segmentation. A medical diagnostic system that uses deep learning is being developed, however, it typically necessitates a huge amount of computation due to the nature of medical imaging. OD pre-processing uses an algorithm that mimics human visual attention to locate the OD automatically. As humans, we use our visual perceptions in order to make sense of our surroundings. If you look at a picture, there are certain parts of it that catch your eye. Human visual perception can be predicted using computational visual attention (CVA) models. Based on human visual perception notions, these models were developed. When it comes to finding OD in fundus retinal images, the bottom-up (BU) saliency paradigm is tested. When paired with mathematical morphology, the IK saliency model and Otsu’s technique are effective tools for detecting OD in retinal images. Keywords CVA · Fundus · Human visual perception · Optic disc · Retinal images

1 Introduction Today, glaucoma is one of the most pressing health and therapeutic issues of our time because of its rapid progression. The number of persons who are infected with the disease is on the rise. Glaucoma affects 1–2% of the population, and over half A. Geetha Devi (B) Department of ECE, PVP Siddhartha Institute of Technology, Vijayawada, Andhra Pradesh, India e-mail: [email protected] N. Krishnamoorthy MCA Department, SRM Institute of Science and Technology, 89 Bharathi Salai, Ramapuram Campus, Chennai, India K. I. Ahmed · S. I. Patel · I. Khan Computer Science, Bahrain Training Institute, Higher Education Council, Ministry of Education, Manama, Bahrain R. Satpathy CSE (FET), Sri Sri University, Cuttack, Odisha, India © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 I. J. Jacob et al. (eds.), Expert Clouds and Applications, Lecture Notes in Networks and Systems 444, https://doi.org/10.1007/978-981-19-2500-9_22

317

318

A. Geetha Devi et al.

of those people are unaware that they have it. Medical images, as we know, play a vital function in conveying information from different sections of the body, detecting disorders, medical research, and education [1]. As a result of these studies, glaucoma may now be diagnosed automatically using image processing algorithms. Automated image processing systems can process massive volumes of photos in a short period of time and at a low cost while avoiding human errors and other flaws. Clear area in the retina that is almost round in shape and has a diameter of 1.5 mm and boundaries at the periphery is the optic disc. The optic nerve can be found in a yellowish-white region at the center of the optic disc. Donut-like appearance of the optic nerve head is due to optic cup’s depth being greater than surrounding nerve tissue [2]. On display in Fig. 1, you’ll see an optical disc and cup in action. When axons from the ganglion cells depart the eye, they form the optic nerve at the optic disc. Extracting a visual representation from an optical disc is meant to make it more meaningful and easier to understand. Extraction of objects and boundaries from images can be done using Retinal Image Extraction [3, 4]. For some characteristic or computed attribute, every pixel in a region is the same. The first step in any image analysis method is commonly called “image extraction.” Next, feature extraction and object recognition are significantly dependent on the extraction’s performance. An object may never be recognized unless a good extraction algorithm is used [5]. The goal of image extraction is to divide an image into useful sections for a specific application. It is possible to extract useful information about the scene’s surfaces from Fig. 1 Optic disc and optic cup in the retinal image

Visual Attention-Based Optic Disc Detection System …

319

simple gray-level images. A sequence of operations aiming at gaining a comprehensive knowledge of an image usually begins with image extraction as an important first step [6, 7].

2 Related Work Retinopathy diagnosis and fundus image analysis rely on the optic disc’s position as a starting point for subsequent processes, such as optic disc segmentation. An enhanced Harris corner location algorithm is presented in this study by Deng in 2021. When examining the retinal fundus image, the optic disc appears to have the most corners, due to the presence of densely packed vessels and apparent gray shading in the image. Image augmentation, vessel extraction, matching filters, and other approaches are all utilized to pinpoint the precise location of the subject matter [8, 9]. Computer-aided diagnosis relies heavily on the accurate localization of the optic disc (OD) prior to OD segmentation. Deep learning research is encouraging the development of sophisticated medical diagnostic systems, yet medical images frequently require a considerable amount of computation. Automated detection of the location of the OD is performed using an algorithm that mimics human visual attention [10]. These are the four steps that Liang and his colleagues took, in the order listed [11]. Both healthy and diseased retinal pictures were included in the two datasets that were utilized to evaluate the suggested approach. The MESSIDOR dataset properly identifies the OD in 1195 of the 1200 photos (99.58%). On top of that, DRIVE has a detection accuracy of 100%, which exceeds current models. Two datasets of the prospective OD region are presented [12].

3 Proposed Visual Attention-Based Optic Disc Detection System Figure 2 depicts the VAODD system’s main phases. The STructured Analysis of the REtina (STARE) project dataset was employed as the primary data source for this study [13–15]. 52 images are disjunctive type and 27 are conjunctive type among 79 fundus retinal images in the chosen subset. Saliency maps are computed using the Itti–Koch (IK) (2001) computational BU saliency model, which is closely aligned with the Feature Integration Theory (FIT). According to Fig. 2, each stage is broken down into the following categories: pre-processing; processing; and post-processing (Fig. 2).

320

A. Geetha Devi et al.

Fig. 2 Major steps in the visual attention-based optic disc detection (VAODD) system

Visual Attention-Based Optic Disc Detection System …

321

Fig. 3 a Actual image, b presented location by the IK model

3.1 Pre-processing Fundus retinal picture foreground from background can vary greatly in color, intensity, and orientation, as seen in Fig. 3a. Figure 3 shows the places detected using the IK model (b) [16]. After inspecting all four corners, the OD region can be distinguished due to the varying color, intensity, and direction of the retinal image and its background. Before employing the IK model for OD detection, a pre-processing step is necessary. Pre-processing involves using the morphological opening operator (◯) on the original image (Im). The following is the definition of the operation of opening by means of structuring element B: Impre = ◦(B)(Im )

(1)

In this example, the disc-like morphological organizing element is B. Using the input image Impre as a starting point for processing; an image is removed from it before being used as input.

3.2 Processing When calculating the saliency map, the IK model is used. The input image Impre is deconstructed using linear filters customized to specific stimulus parameters such as brightness, red, green, blue, and yellow colors, or various local orientations. At a variety of spatial scales, this breakdown is carried out. These channels can be subdivided to show smaller and larger things. Gaussian pyramids are used to create a variety of spatial sizes. The following equations are used to produce color channels: Red = red − (green + blue)/2

(2)

322

A. Geetha Devi et al.

Green = green − (red + blue)/2

(3)

Blue = blue − (red + green)/2

(4)

Yellow = red + green − 2(|red − green| + blue)

(5)

Int = (red + green + blue)/3

(6)

In this example, the color channels are used to construct four different Gaussian pyramids: red, green, blue and yellow. For the Gaussian pyramid Int(σ ), Int is used. Gabor O(σ, θ ), is used to extract local orientation information from Int. As you move outward, you’re subtracting (green–red) from your immediate surroundings, and then multiplying that result by six to get the six different red/green feature maps. An example of this is that IK model uses the RG(c, s) map to account for red/green and green/red double opposition at once. RG(c, s) = |(Red(c) − Green(c))(Green(s) − Red(s)|

(7)

BY(c, s) = |(Blue(c) − Yellow(c))(Yellow(s) − Blue(s)|

(8)

Int(c, s) = |Int(c)Int(s)|

(9)

The saliency map is created by combining the normalized maps together. Model outputs a saliency map with the locations that are most important to users.        Saliency Map = 1/3 N Int + N Color + N Or1

(10)

Binary segmentation is used to remove the undesirable sections. It is necessary to employ a threshold in thresholding, and the Otsu algorithm is used. A picture can be transformed into a two-color image via thresholding, which separates light areas from dark ones. It is applied to a binary image and then subtracted from the threshold binary picture using morphological open operations. Following the thresholding of the saliency map, an ithresh image is generated. It is used in post-production.

3.3 Post-processing It is possible to eliminate little objects from an ithresh image while keeping the shape and size of larger ones using the morphological open method. A large enough structuring element is possible, but it isn’t necessary. It is also possible to keep the real edge of the OD intact by keeping the structuring element as small as possible.

Visual Attention-Based Optic Disc Detection System …

323

Table 1 VAODD result analysis Disjunctive OD target of 52 images Pass

Fail

45

7

Accuracy: 86.5%

Conjunctive OD target of 27 images

Total 79 images

Pass

Fail

Pass

Fail

14

13

59

20

Accuracy: 51.8%

Iout = (B)(Ithresh )

Accuracy: 74.1%

(11)

4 Results An optical disc detection (VAODD) technique based on visual attention is tested using 81 STARE dataset fundus images. This figure shows the intermediate outcomes of the VAODD system, including pre-processing, saliency map computation, postprocessing image, and OD detection. Figure 5 shows instances of fundus retinal images where the VAODD technology fails to detect the optic disc (OD). Table 1 shows the analysis of the results. Of 52 disjunctive and 27 conjunctive types of pictures, it can identify OD in 45 and 14, respectively. For disjunctive images, the VAODD system takes an average of 6.36 s and for conjunctive images, an average of 6.45 s, as shown in Figs. 4 and 5. Understanding how the BU technique is used in IK model works for OD detection is examined further. The most salient visual location can be selected using the maximum of the saliency map, and this is where the focus of attention (FOA) should be. As demonstrated in Fig. 6, the FOA for disjunctive images is one, but the FOA for conjunctive images varies when utilizing the IK model. FOA one and FOA four are displayed in Fig. 7 of the fundus retinal picture.

5 Conclusion In this study, the bottom-up (BU) saliency model is examined for its ability to locate OD in fundus retinal pictures. When paired with mathematical morphology, the IK saliency model and Otsu’s technique are effective tools for detecting OD in retinal images. Images with OD as a disjunctive type of target had an image success rate of 86.79%, whereas images with OD as a conjunctive type of target had an image success rate of 50%. It also demonstrates that the BU technique alone is not sufficient for detecting a target. EGODD (eye gaze-based optic disc detection) is presented to solve the limitations of this work. This system uses a combination of bottom-up and top-down methods to detect the optic disc.

324

A. Geetha Devi et al.

Fig. 4 a Retinal image as input, b pre-processing results, c saliency map, d post-processing result, and e detected OD

Fig. 5 For both disjunctive and conjunctive instances, the suggested algorithm takes a long time

Fig. 6 Saliency map showing the first FOA (i.e., disjunctive case) and the subsequent FOA (i.e., conjunctive case) as OD pop-ups (left) (right)

Visual Attention-Based Optic Disc Detection System …

325

Fig. 7 OD appears as a pop-up on the fundus image’s first FOA (on the left) and the image’s fourth FOA (on the right)

References 1. L. Deng, Y. Wang, J. Han, Optical disc location based on similarity to improved Harris algorithm, in 2021 40th Chinese Control Conference (CCC) (2021), pp. 2185–2189. https://doi. org/10.23919/CCC52363.2021.9550565 2. S. Athab, N.H. Salman, Localization of the optic disc in retinal fundus image using appearance based method and vasculature convergence. Iraqi J. Sci. 61(1), 164–170 (2020) 3. M. Liang, Y. Zhang, H. Wang, J. Li, Location of optic disk in the fundus image based on visual attention, in 2020 International Conference on Computer Information and Big Data Applications (CIBDA) (2020), pp. 446–449. https://doi.org/10.1109/CIBDA50819.2020.00106 4. D.V. Gunasekeran, D.S.W. Ting et al., Artificial intelligence for diabetic retinopathy screening prediction and management. Curr. Opin. Ophthalmol. (2020) 5. Z. Wang, N. Dong, S.D. Rosario, M. Xu, P. Xie, E.P. Xing, Ellipse detection of optic disc-andcup boundary in fundus images, in 2019 IEEE 16th International Symposium on Biomedical Imaging (ISBI 2019) (2019), pp. 601–604. https://doi.org/10.1109/ISBI.2019.8759173 6. H. Fu, J. Cheng, Y. Xu et al., Joint OD and cup segmentation based on multi-label deep network and polar transformation. IEEE Trans. Med. Imaging 37(7), 1597–1605 (2018) 7. Z. Gui-Ying, Z. Xian-Jie, Deep learning based on optic disk automatic detection. J. Guizhou Educ. Univ. 33(3), 27–32 (2017) 8. K.-K. Maninis, J. Pont-Tuset, P. Arbeláez, Deep retinal image understanding, in International Conference on Medical Image Computing and Computer-Assisted Intervention (2016), pp. 140– 148 9. B. Harangi, A. Hajdu, Detection of the OD in fundus images by combining probability models. Comput. Biol. Med. 65, 10–24 (2015) 10. S.H. Zheng, J. Chen, L. Pan et al., OD detection on retinal images based on directional local contrast. Chin. J. Biomed. Eng 33(3), 289–296 (2014) 11. A. Borji, M.-M. Cheng, Q. Hou, Salient object detection: a survey. Eprint Arxiv 16(7), 3118 (2014) 12. E. Erdem, A. Erdem, Visual saliency estimation by nonlinearly integrating features using region covariances. J. Vis. 13(4), 11–11 (2013) 13. R.K. Gupta, S.-Y. Cho, Window-based approach for fast stereo correspondence. IET Comput. Vis. 7(2), 123–134 (2013) 14. N. Sinha, R.V. Babu, Optic disk localization using L1 minimization, in 2012 19th IEEE International Conference on Image Processing (2012), pp. 2829–2832

326

A. Geetha Devi et al.

15. Sivaswamy, S.R. Krishnadas, Optic disk and cup segmentation from monocular color retinal images for glaucoma assessment. IEEE Trans. Med. Imaging 30(6) (2011) 16. A.K. Mishra, Y. Aloimonos, L.-F. Cheong, A.A. Kassim, Active visual segmentation. IEEE Trans. Pattern Anal. Mach. Intell. 34(4) (2012)

An Overview of Blue Eye Constitution and Applications Jadapalli Sreedhar, T. Anuradha, N. Mahesha, P. Bindu, M. Kathiravan, and Ibrahim Patel

Abstract Every day, the Blue Eyes Technology grows in terms of new developments, allowing it to produce something new and valuable for human beings every day. Technology that can sense human emotions and sensations is being developed in order to make work easier for humans. A computer being able to recognize and respond to human emotions is crucial for this technology to succeed. In order to identify both physical and psychological actions, Blue Eyes technology employs a range of sensors and then pulls essential information from these processes. It is, therefore, possible to calculate the user’s mental, emotional, physical, or informational state. Keywords Blue eye · Sensors · Human emotions · Blue eye technology · Blue eye constitution

1 Introduction The ultimate goal of this research is to create computational computers with humanlike sensory capacities. Using camera and microphone, it produces a computational J. Sreedhar (B) EEE Department, Vignana Bharathi Institute of Technology, Hyderabad, India e-mail: [email protected] T. Anuradha Department of Electrical and Electronics Engineering, KCG College of Technology, Chennai, India N. Mahesha Department of Civil Engineering, New Horizon College of Engineering, Bangalore, India P. Bindu Department of Mathematics, Koneru Lakshmaiah Education Foundation, Vaddeswaram, AP, India M. Kathiravan Department of Computer Science and Engineering, Hindustan Institute of Technology and Science, Padur, Kelambaakkam, Chengalpattu, India I. Patel Department of ECE, B V Raju Institute of Technology, Narsapur, Medak, India © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 I. J. Jacob et al. (eds.), Expert Clouds and Applications, Lecture Notes in Networks and Systems 444, https://doi.org/10.1007/978-981-19-2500-9_23

327

328

J. Sreedhar et al.

computer that also feels like a human and can identify the actions and feelings of a user. Aside than that, it sounds like a human voice. The term blue in the blue eye technology refers to Bluetooth, which allows for wireless communication, and the word eye refers to eye movement, which provides us with a wealth of useful and intriguing information. This technology’s primary goal is to give computers the ability to think like humans [1]. The ability to see the world from another person’s perspective is a skill that we all possess, and it’s this ability that allows computers to achieve human intelligence and power. The goal of the blue eye technology is to create computing machines with human-like per spectral and sensory abilities. It employs our non-intrusive sensing method, which includes the use of the most up-todate video cameras and microphones, to identify the user’s actions. Users’ physical and emotional states can be discerned by the machine based on what they are looking at and what they are trying to accomplish [2, 3]. For example, weariness and mental disease can be avoided by using blue eyes. As a result, sitting in front of a computer for a lengthy period of time might lead to exhaustion or mental disease in the form of fatigue. The computer equipped with Blue Eyes Technology can help us overcome these constraints. If you have a mental illness, you can employ Blue Eyes Technology to help construct a machine that understands your emotions and interacts with you [4]. A Blue Eyes Technology can be utilized to overcome these constraints.

2 Technologies Used in Blue Eye Devices that can detect human activity make blue eyes a popular topic in computer literature. It’s the primary goal of blue eyes to create computers that feel. Affective Computing is a term used to describe the process of creating computers that are able to recognize human emotions. The devices must be able to detect even the slightest shifts in our state of mind or behavior [5]. When a person is happy or angry, they could click their mouse quickly. The user’s voice and interest are primarily understood through the use of artificial intelligence speech recognition and a simple user interest tracker. People’s emotional feelings must now be taken into account by the machines. With the use of technical input devices such as the Emotion Mouse (which tracks the user’s emotions), these inputs are taken into consideration [6]. The following are some of the different Affective Computing implementation strategies that have been discussed.

2.1 Manual and Gaze Input Cascaded (MAGIC) Pointing It’s a new strategy for dealing with the “eye gaze” for the human–computer interface. An excellent pointing method for enhanced computer input has been gaze tracking.

An Overview of Blue Eye Constitution and Applications

329

However, there are several disadvantages to using eye-tracking technology. An alternate method, known as MAGIC (Manual and Gaze Input Cascaded) is considered in order to overcome these obstacles [7]. MAGIC Pointing is a gaze-tracking technology that is used in conjunction with manual controls to select and aim the cursor. By warping the cursor back to its old home position, MAGIC pointing reduces the amount of cursor motion required while selecting a target.

2.2 Artificial Intelligence Speech Recognition The spoken words are scanned and compared to words that have been stored in the brain. Internal storage of the user’s voice will be made available to the computer. Because of the wide range of pitch frequency and time gap, pattern matching is developed to discover the best fit.

2.3 Simple User Interest Tracker (SUITOR) This is a simple tool for keeping tabs on user preferences. Desktop computers benefit from it since it provides a wider range of information. Using this information, a computer screen’s scrolling ticker is filled with relevant information about the user’s current job.

2.4 Emotion Mouse Even the tiniest changes in a person’s emotional state can be detected by Blue Eyes Technology’s equipment. Depending on the person’s mood, he or she may strike the keyboard fiercely or softly, depending on the type of keyboard [8]. Simply by moving or touching the computer’s mouse or keyboard, the Blue Eyes Technology detects human emotional behavior, and the system begins to react accordingly. Emotion Mouse and other sophisticated devices are used to do this. Simply by touching the mouse, our Emotion Mouse is able to detect emotions. When a user interacts with a computer, the Emotion Mouse analyses and identifies their emotions, such as sadness, happiness, rage, excitement, and so on. The image (Fig. 1) displays a real mouse with the emotion mouse installed on it. Different infrared detectors and temperature-sensitive chips are included in the mouse’s sensors.

330

J. Sreedhar et al.

Fig. 1 Emotional mouse

3 Construction of Blue Eye 3.1 Software Required Connection manager, data analysis module, and visualization module are all part of the software system. • Connection Manager: Wireless connectivity between mobile data acquisition units and the central computer system unit will now be handled by a software connection manager. In order to create Bluetooth connections, authenticate users, buffer inbound data and provide alerts to the CSU Hardware’s connection manager [9]. • Data Analysis Module: Analyzing the raw sensor data, it provides information on the operator’s physiological status. Each working operator is under the watchful eye of a separate data analysis module. Many smaller analyzers are used to gather different kinds of data for the module. Analyses of eye movements, pulse rate,

An Overview of Blue Eye Constitution and Applications

331

and custom analyzers are the most important for determining an operator’s level of visual attention, as well as determining the pulse rate of the operator. • Visualization Module: Using the visualization module, they can see each of the working operators’ geological conditions, and they can see a preview of the second video source and record an audio file. The supervisor receives immediate notification of any and all incoming alarms [10]. When the visualization module is in offline mode, all of the recorded physiological parameters such as alert videos and audio data can be retrieved from the database and viewed.

3.2 Hardware Required Both the data gathering and central system units are hardware components. A.

Data Acquisition Unit

The Blue Eyes Technology uses a mobile component called the DAU. the physiological data from sensors is collected by DAU and sent to the CSU for processing and verification. The sensors and the Central System Unit (CSU) are connected via Bluetooth, which acts as a wireless interface. The operator is given a unique PIN and ID for authentication. Communication between the devices is accomplished through the use of a keyboard, beeper, and LCD display. A micro jack plug is used to transport the user’s data [11]. B.

Jazz Multi-sensor

Eye movement sensors, such as the Jazz multi-sensor, are used to collect critical psychological data in data collecting systems. Data on eye position, blood oxygenation, horizontal and vertical axis accelerations as well as ambient light intensity can also be retrieved from the device’s raw digital data [12]. Direct infrared holographic transducers can be used to monitor eye movement in that multi-sensor. C.

Central System Unit

The CSU is the second most significant component of the Blue Eyes Technology when it comes to connecting to a wireless network. The main components of the CSU are a wireless Bluetooth device and a speech information transmission system. USB, parallel, and serial cable wires are used to connect this CSU to a computer. The tiny jack plug is used to access audio data [10]. The serial and power ports of the personal computer are being used to communicate with a program that includes the operator’s personal ID. There are sensors in this technology that can assess a person’s emotional state, which can be utilized by the supervisor to track the progress of the selected operator’s work (Fig. 2).

332

J. Sreedhar et al.

Fig. 2 Blue Eye structure

4 Applications of Blue Eye • Allows users to work on other tasks at the same time as they utilize a speech recognition system. Using voice instructions, a user can remain focused on observation and manual tasks while still managing the machine. • It has been reported that some large stores are using Alma den’s Blue Eyes software to construct surveillance systems that monitor and interpret client movements, according to IBM engineers in San Jose, California. Blue Eyes collects video data on eye movement and facial expression to explore ways for computers to predict consumers’ wishes [7]. Your computer might, for example, identify comparable links and open them in a new window if your attention lingers on a web page’s title. However, snooping on customers turns out to be the first practical use of this study. • A human operator is necessary at all times in the sector of security and control, thus this can be used. • The vehicle sector could also benefit from this technology. An emotional state of an individual can be determined just by touching their mouse, as the computer is designed to be capable of doing so. Military operations are another key use for speech processing. • A good example is the ability to fire a weapon using only your voice. Pilots don’t have to use their hands to communicate with computers if they have high-quality speech recognition technology. • Other examples include radiologists analyzing hundreds of X-rays, ultrasonograms, and CT scan results while concurrently speaking their findings to a speech recognition system linked to word processors. Instead of writing the words, the radiologist can concentrate on the photographs [13]. • Computers could potentially be used to make airline and hotel reservations using voice recognition. To make a reservation, cancel a reservation, or inquire about the schedule, a user merely needs to state his demands.

An Overview of Blue Eye Constitution and Applications

333

5 Emotion Computing Using Blue Eye Technology The facial expression study of Paul Ekman has revealed a link between emotional state and physiological data. Ekman’s Facial Action Coding System is described in a selection of works by Ekman and others on tracking facial behavior. Devices that monitor numerous parameters, such as pulse, galvanic skin response (GSR), temperature, and somatic movement are fitted to subjects in one of Ekman’s investigations Facial expressions corresponding to the six most common emotions were assigned to each participant in the study. Emotions such as joy and surprise are also included in his list of six. By analyzing physiological data, Dryer (1993) was able to identify several emotional states. These include GSR, heart rate, skin temperature, and general somatic activity (GSA). There are two main types of data analysis: descriptive and predictive. The first step in determining the data’s dimensionality is to run it through an MDS process.

6 Results Six different physiological evaluations are used to represent the six different emotions that are shown in GSA, GSR, pulse, and skin temperature during the five-minute baseline and test sessions. Approximately three to four times a second, GSA data was sampled, and a pulse was discovered as the result of a beat. Individual physiological variance was taken into account while calculating the difference between the baseline and test results. When scores deviated by more than 1.5 standard deviations from the mean, they were deemed missing. A total of twelve scores were omitted as a result of these criteria. The Emotion mouse’s notion is based on solid evidence. correlation models are used to link the physiological data. Calibration is used to create the correlation model. Calibration signals generated by users with known or measured emotions at the time of calibration are evaluated using statistical analysis of the association between attributes and emotions (Fig. 3; Table 1).

7 Conclusion Today’s world is expanding at an incredibly rapid rate since it comprises mostly of real-time systems. The Blue Eyes Technology provides a more powerful and userfriendly computing environment for users. We can communicate wirelessly thanks to Bluetooth, and we can also learn new things thanks to our eyes’ movements. As technology continues to advance, it is only a matter of time before everyone is familiar with and uses this technology in their daily lives. Even our cell phones may be affected. In any case, this is merely a projection based on current technical trends. Unlimited the market for blue eye technologies in the future is expected to increase.

334

J. Sreedhar et al.

Fig. 3 Graph showing emotion scores wrt baseline index

Table 1 Emotion scores Anger

100%

50%

25%

6.5

6.5

6.5

Disgust

5.8

5.7

5.8

Fear

5

5

5

Happiness

6.5

6.2

5.8

Sorrow

5

5.2

5.3

Surprise

4.8

4.9

4.8

References 1. K. Dhinakaran, M. Nivetha, N. Duraimurugan, D.C.J.W. Wise, Cloud based smart healthcare management system using blue eyes technology, in 2020 International Conference on Electronics and Sustainable Communication Systems (ICESC) (2020), pp. 409–414. https://doi.org/ 10.1109/ICESC48915.2020.9155878 2. M. Kumawat, G. Mathur, N.S. Saju, Blue eye technology. IRE J. 1(10) (2018). ISSN 2456-8880 3. B. Oyebola, O. Toluwani, Blue eyes technology in modern engineering: an artificial intelligence. Int. J. High. Educ. 45–65 (2018) 4. H.A. Patil, S.A. Laddha, N.M. Patwardhan, A study on blue eyes technology. Int. J. Innov. Res. Comput. Commun. Eng. 5(3) (2017) 5. M.R. Mizna, M. Bachani, S. Memon, Blue eyes technology, in Eighth International Conference on Digital Information Management (ICDIM 2013) (2013), pp. 294–298. https://doi.org/10. 1109/ICDIM.2013.6693995 6. H. Sharma, G. Rathee, Blue eyes technology. Int. J. Comput. Sci. Manag. Res. (2013) 7. Psychologist World, Eye Reading Language (Body Language), July 2013, www.psychologist world.com/bodylanguage/eyes.php 8. S. Madhumitha, Slide Share, Blue Eyes Technology, Mar 2013, www.slideshare.net/Colloq uium/blue-eyes-technology 9. S. Chatterjee, H. Shi, A novel neuro fuzzy approach to human emotion determination, in 2010 International Conference on Digital Image Computing Techniques and Application (DICIA) 10. A. Aly, A. Tapus, Towards an online fuzzy modeling for human internal states detection, in 2012 12th Internal Conference on Control, Automation Robotics and Vision (ICARCV 2012), Guangzhou, China, 5–7 Dec 2012

An Overview of Blue Eye Constitution and Applications

335

11. D. McDuff, R. Kaliouby, T. Seneschals, M. Arm, J. Cohn, R.W. Picard, Affective-MIT facial expression dataset (AMFED): naturalistic and spontaneous facial expressions collected in-thewild, in 2013 IEEE Computer Society Conference on Computer Vision and Pattern Recognition Workshops (CVPRW’10), Portland, OR, USA, June 2013 12. R. Nagpal, P. Nagpal, S. Kaur, Hybrid technique for human face emotion detection. Int. J. Adv. Comput. Sci. Appl. 1(6) (2010) 13. F. Zhizhong, L. Lingqiao, X. Haiying, X. Jin, Human computer international research and realization based on leg movement analysis, in 2010 International Conference on Apperceiving Computing and Intelligence Analysis (ICACIA)

Enhancement of Smart Contact on Blockchain Security by Integrating Advanced Hashing Mechanism Bharat Kumar Aggarwal, Ankur Gupta, Deepak Goyal, Pankaj Gupta, Bijender Bansal, and Dheer Dhwaj Barak

Abstract The demand for blockchain is growing every day. Smart contracts, on the other hand, are commonly employed in business applications. The hashing mechanism is a security method that ensures the blockchain’s dependability. The purpose of the proposed article is to look at the function of hashing mechanisms in blockchain security. The current study took into account a number of previous studies in the field of hashing mechanisms and blockchain. Previous research has encountered challenges such as restricted scope, security, and performance. The suggested work is expected to create a more secure and high-performing solution. Simulation work has been done to confirm the execution time of smart contracts considering several hashing mechanisms. Comparative analysis of collision is also considered to present the security of the proposed work. Keywords Hashing mechanism · Smart contracts · Secure hash algorithm (SHA)-256 · Message digest algorithm · Blockchain · Rivest-shamir-adleman (RSA) encryption

1 Introduction The need for blockchain technology is developing. There is a considerable application of smart contracts in corporate world. Use of improved hashing techniques for smart contracts on the blockchain may be beneficial. Prior research in the realm of blockchain and hashing mechanisms has been included in the present study. Lacks of security and efficacy have impeded earlier studies in the past. Security and performance are both anticipated to be better with the proposed improvement. Proposed research focused on presenting advanced hashing mechanism for blockchain technology. The present research has also considered hashing techniques such as SHA256 and MD5 that are frequently used. The objective of the work is to simulate the execution time comparison among previous hashing and proposed RSA integrated B. K. Aggarwal · A. Gupta (B) · D. Goyal · P. Gupta · B. Bansal · D. D. Barak Deperment of Computer Science and Engineering, Vaish College of Engineering, Rohtak 124001, Haryana, India e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 I. J. Jacob et al. (eds.), Expert Clouds and Applications, Lecture Notes in Networks and Systems 444, https://doi.org/10.1007/978-981-19-2500-9_24

337

338

B. K. Aggarwal et al.

MD5 hashing mechanisms. Moreover proposed mechanism is supposed to provide better transaction success rate and lower failure rate as compared to previous hashing mechanisms [1].

1.1 Blockchain New shared databases have been created by storing data in blocks, which are connected via cryptography. A new block is created each time fresh data is received. Chaining each block together once it has been filled with data and linked to the next helps maintain track of the sequence in which they were formed. The core use case for blockchain is transactional data; however other forms of data may be kept as well. No one person or organization has any influence over blockchain technology since it is employed in a decentralized manner. Data may be saved and transmitted using blockchain technology, but it can never be tampered with. To create immutable ledgers, which are records that cannot be changed or destroyed, a blockchain must be used. DLT (distributed ledger technology) is a common term for blockchain because of this [2]. Working of Blockchain If new transaction has to be added to the chain, it will be verified by all of network’s members. The definition of “valid” in the blockchain system may differ from system to system. After then, a majority of the parties must agree to the transaction. After then, all nodes in the network get a block containing a collection of permitted transactions. They then check to see whether the new block is legitimate. Hash of previous block serves as a unique fingerprint for each subsequent block. Security and fault tolerance are built into the design of blockchains. As a consequence, a decentralized consensus was established using a blockchain. Since blockchains may be used to management activities, store events, and medical data, they can also be utilized to track the origin of food and the traceability of votes [3, 4].

1.2 Smart Contracts It is possible to activate a smart contract when a set of predetermined conditions are met. So that there is no third party or the long process involved in the execution of a contract, both parties will know precisely what is going to happen. The “if/when/then” instructions in the blockchain may be used to build a smart contract [5]. Role of Smart Contracts in Various Sectors Blockchain is now being used in areas other than digital money, such as health, the Internet of Things, and education. Research is proposing a comprehensive mapping analysis in this research to gather and evaluate important blockchain technology

Enhancement of Smart Contact on Blockchain Security …

339 Check External Data

Preset Trigger condition

Preset response rule

….

Block

State

Block

Condition 1: Response 1 Condition 2: Response 2 Condition 3: Response 3 ……. Condition n: Response n

Smart Contracts

Value

Block

Block

Block

….

Fig. 1 Mechanism of smart contract [7]

research in the higher education sector. The research aims to assess the current status of blockchain technology. Research gaps and obstacles are also highlighted in this paper. It is critical to offer and satisfy the changing needs of contemporary society while also providing for the individual. Using technology, it seems that the conventional system, which has been in existence for a long time, has improved. How Smart Contracts Work Smart contracts are composed of a series of “if/when…then…” expressions written in code and stored on a distributed ledger. Computers in a network carry out the activities when certain criteria are fulfilled and confirmed by other computers in the network. Actions like distributing money to the right people, registering an automobile, providing alerts, or issuing a ticket may fall under this category. This transaction’s blockchain will be updated as soon as it has been completed. Any number of criteria may be included in a smart contract to guarantee that the work is completed effectively for everyone concerned. If/when/then rules must be agreed upon, all possible exceptions examined, and a framework for addressing disputes must be developed before the terms may be defined. Smart contracts may then be developed by developer, but organizations that use blockchain for business are increasingly web interfaces, giving templates, and other online tools to make the production of smart contracts more convenient [6] (Fig. 1).

1.3 Blockchain Hash Function A function for calculating bitcoin’s value using a hash function, you may create a new string with a predetermined length from any input string. In computing, a hash is a kind of identifier. This hash is a consequence of a hash algorithm. A one-way hashing method is used to create message digests from an input file or text string. A key is unnecessary. In order to interpret an encrypted message, the receiver must be the intended recipient. Those who should not have access to a file’s data can’t read it thanks to this tool.

340

B. K. Aggarwal et al.

The “hash” or “hash value” is a numerical representation of a particular input value that is generated by a hash function. A processing unit that accepts input of any length and returns a fixed-length output—the hash value—as its result. Blockchain relies heavily on hashing. Hashing, a cryptographic method, may turn data into a string of characters. Hashing, on the other hand, is more efficient since the length of the hash is fixed [8].

1.4 Hashing Techniques Hashing is the process of converting any length of input into a cryptographic fixed output using a mathematical approach (Bitcoin uses SHA-256, for example). A block’s hash value may be computed using the following algorithms: MD5 To create a 128-bit (16-byte) hash value, the MD5 algorithm (Message Digest) utilizes a hexadecimal integer representation to produce a 32-digit hexadecimal integer in text format. MD5 has been utilized in cryptography applications and is commonly used to verify the integrity of data. SHA1 The SHA1 cryptographic hash function was developed by NSA. One of the most often used SHA1 hash functions; SHA1 provides a 160-bit (20 bytes) hash result. Hash functions like SHA1 are used in a wide range of applications and protocols, making them the most popular of the current generation. In the long run, the SHA1 algorithm may not be secure enough. SHA1 is not recommended for use. SHA224 The SHA224 cryptographic hash algorithm was developed by the National Security Agency (NSA). SHA224 generates a 224-bit hash value, which is usually expressed as a 56-digit hexadecimal number. SHA256 SHA256 cryptographic hash function was developed by NSA. Hexadecimal numbers are used to represent SHA256’s 256-bit (32-byte) hash value. People who make Bit-coins use the hash function and mining algorithm called (Table 1).

1.5 Encryption Algorithm Based on RSA The most widely used asymmetric encryption algorithm, RSA, was developed in 1977 by Ron Rivest, Adi Shamir, and Leonard Adleman (known collectively as “RSA”). Its strength comes from the “prime factorization” process on which it depends. Simply

Enhancement of Smart Contact on Blockchain Security …

341

Table 1 Different types of hashing algorithms [9] Keys for comparison

MD5

SHA

Security

SHA 128-bit 2128-bit operation is less safe

MD5 is not as secure as MD5

Length of the message digest

Operation in 264-bits

A 2160-bit operation using 160 bits

To unearth the original message, attacks are required

It is also quicker, requiring just Operation with 280 bits 64 iterations

The search for two mails with the same MD

There have been occasional reports of attacks

80 iterations were necessary since MD5 was so much slower

Speed

As compared to SHA, it has less resilience to collisions. In the case of SHA 256, it is 126 bits, while in the case of SHA 512, it is 256 bits

There have been no reports of a similar incident at this time

So far, all assaults have been successful

SHA 128-bit 2128-bit operation is less safe

MD5 has a lower rate of collisions compared to

Resistance to Collisions

Operation in 264-bits

MD5 is not as secure as MD5

multiplying two astronomically enormous random prime numbers yields an even greater one with this approach. We have to figure out the original prime numbers from this astronomically large number. It’s impossible for today’s supercomputers or even humans to solve this problem, even with the right key length and entropy. RSA-768bit key breaking required more than 1500 years of computing time (distributed over hundreds of workstations) in 2010—significantly longer than the normal 2048-bit RSA key in use at the time.

2 Literature Review In our investigation into how hashing techniques may be used to build smart contracts on the blockchain, we reviewed a variety of academic research publications on blockchain, Smart Contracts, and hashing algorithms. To assist us to keep on track with our research goal, we’ll shortly be providing a brief introduction to the papers that serve as the foundation for our work. Blockchain technology has been called a disruptive technology by Nofer et al. [1]. They believe that the financial sector is the primary application of blockchain technology. Cryptography, distributed technology, and consensus accounting mechanisms were all discussed by Chen et al. [2]. Credit risk is one of the issues brought on by the rapid growth of Internet technology, according to Zhang et al. [3]. In their paper, Hassan et al. [5] proposed insurance contract architecture based on smart contracts. There is evidence that this technology and its qualities may encourage

342

B. K. Aggarwal et al.

open research, according to a paper released in 2019 by Leible [4, 6]. Described how the banking sector might be disrupted by international money transmission, automated bank ledgers, digital assets, and smart contracts using blockchain technology in 2015. Throughout 2016, Jesse [10] conducted a detailed mapping analysis to gather all relevant blockchain studies. In 2017, Igor [11] investigated the use of blockchain technology to store, retrieve, and disseminate data over a decentralized network. An in-depth look at blockchain technology was provided by Zibin [12] in 2017. New solutions for NIST-specified secure hash algorithms were released by Gueron [13] in 2012 for the 2nd Generation Intel® CoreTM CPU. In 2013, Putri [8] employed the MD5 + salt technique to safeguard the database user’s login credentials. To make data more secure, Raju [14] sought to replace the MD5 method with SHA-2, which hashing techniques provide 256 and 384 hash values in 2018. To better understand how Man in the Middle (MITM) attacks affect patient privacy, Purwanti [15] carried out the study in 2018.

3 Problem Statements There have been several researches that did excellent work in field of hashing algorithm, blockchain, and smart contracts. But the scope of those researches is limited. Limited work has been made for smart contract security. Thus, there is a need to inject advance hashing mechanism into blockchain. It’s essential that the transactions be authenticated. In the case of an invalid transaction request, the consensus algorithm acts in line with and cancels the claim. The use of blockchain and smart contracts allows this solution to overcome all of the trust and security difficulties that depend on a traditional insurance policy.

4 Proposed Work In the realm of hashing mechanism, blockchain, and smart contracts, there have been a number of notable researchers. These studies, on the other hand, have a restricted scope. Smart contract performance has received just a small amount of attention. Working of Tradition Smart Contract Traditional smart contract took a lot of time because previous experience and probability and success rate of previous transaction is not considered. If certain requirements are satisfied, then a smart contract will be activated. Contracts may be automated to ensure that both sides know exactly what will happen without the need for a third party or a lengthy procedure. It is possible to create a smart contract using the "if/when/then" instructions contained in the blockchain.

Enhancement of Smart Contact on Blockchain Security …

343

Advance MD5 New MD5 contains 64-bits, as recommended by this paper. An evaluation of MD5’s performance has been carried out here. Because of this, extra bits were connected to boost collision resistance. A 32-bit additive random number was used to get this result. Smart contract running would make use of an advance hashing mechanism that is an integration of RSA and MD5. This system would provide more reliability and better performance. Working on Proposed Smart Contract In the proposed effort, a random key was established to reduce likelihood of collisions in the present MD5. Aside from that, RSA has also been utilized to enhance the security of MD5. It is not possible to compare the size of the SHA386 and the SHA512 hashes to the SHA256 and MD5. As the clock’s size grows, so does the number of clock cycles. CPU clock cycles, storage space, and collision probability are all used to evaluate the hashing algorithm’s performance. Performance of hashing algorithm = f (CPU clock cycle consumed, storage space, collision probability) (Fig. 2).

MD5 (32 BIT)

Random Number (32 Bit)

RSA Advance MD5 (64 BIT)

Smart Contract

Bob Wants to Sell His House

Land Deed is digitized

Matching of buyer and seller Smart contract receives and distributes assets

Automation of Clearing and Settlement Digitization of Currency

Fig. 2 RSA integrated MD5 used in smart contract

John wants to buy House

Undisputed Ownership

344

B. K. Aggarwal et al.

Party A wants to Execute Contract

Add Advanced MD5 Mechanism in Block-Chain

Party B is interested in Contract

Smart Contract Assure the Reliability Smart Contract Deed is digitized

Automation of Clearing Settlement

Undisputed Contract

Fig. 3 Working on AI-based proposed smart contract

A smart contract will be triggered if specific conditions are met after considering the previous experience. Contracts would be automated so that both parties know precisely what will happen without the need for a third party or a long process (Fig. 3). The above figure is presenting how the proposed work is considering the validity and reliability of the contract before execution on the bases of previous experience. If the buy or seller did default in the previous transaction then such parties would be filtered during smart contract execution.

5 Results and Discussion While creating a block, blockchain technology is used to perform hashing. For hashing algorithms, there are only four of them: SHA128, SHA256, and MD5. SHA256 has emerged as a popular hashing algorithm in the world of blockchains. Many factors are taken into account while choosing a hashing algorithm, such as storage capacity, time complexity, and collision ratio. “SHA256 is the best approach for preventing collisions, and it also consumes the least amount of storage space”. The suggested study on smart contracts uses an enhanced MD5 algorithm. It’s less secure than SHA256, but it’s also more efficient in terms of storage space. Create an improved MD5 algorithm that is quicker and more resistant to collisions than the standard MD5 algorithm. SHA 256, Traditional MD5, and suggested MD5 Smart contract execution times are compared (Table 2). Considering the above table, Fig. 4 has been plotted to present comparison of different hashing mechanisms. Table 3 is presenting a comparison of transaction failure after detection of treat in case different hashing techniques such as SHA256, MD5, and total time by RSA integrated MD5. Figure 5 is showing plotting of failure rate considering Table 3.

Enhancement of Smart Contact on Blockchain Security … Table 2 Time comparison of various hashing mechanisms

Smart contract

SHA 256

345 MD5

RSA integrated MD5

1

575

535

154.64

2

540

545

120.71

3

552.5

545

98.39

4

557.5

550

92.86

5

545

542.5

90

6

547.36

548.76

98.29

7

550.33

550.62

103.56

8

557.60

555.25

108.65

9

567.02

564.62

112.99

10

570.76

568.23

115.14

11

575.09

576.25

125.09

12

577.86

583.71

130.39

13

581.15

593.38

130.66

14

586.66

598.30

134.26

15

591.89

601.09

134.74

Fig. 4 Time comparison of smart contract execution

Table 4 presents comparison of success rate of transaction considering different hashing techniques such as SHA256, MD5, and total time by RSA integrated MD5. Figure 6 is showing plotting of failure rate considering Table 4. Table 5 is presenting the time consumption during the execution of different hashing techniques such as SHA256, MD5, and total time by RSA integrated MD5. Figure 7 is showing plotting of the failure rate considering Table 5.

346 Table 3 Comparison of transaction failure

B. K. Aggarwal et al. Smart Contract

SHA 256 (%)

MD5 (%)

RSA integrated MD5 (%)

1

92

99

87

2

50

57

49

3

41

43

33

4

76

81

70

5

24

32

22

6

98

100

98

7

42

45

34

8

16

22

10

9

62

65

59

10

82

90

74

11

92

96

89

12

32

37

24

13

41

46

36

14

27

37

19

15

58

59

57

Fig. 5 Comparison of transaction failure after detection of treat

6 Conclusions It has been concluded that the proposed work is a more secure and reliable approach as compared to tradition work. Results conclude that RSA integrated MD5 is providing a better success rate and lower failure rate as compared to standard MD5 and SHA 256. Moreover, the performance is also high. A smart contract’s security has been

Enhancement of Smart Contact on Blockchain Security …

347

Table 4 Comparison of success rate of transaction Smart contract

SHA 256 (%)

MD5 (%)

Total time by RSA integrated MD5 (%)

1

8

1

13

2

50

43

51

3

59

57

67

4

24

19

30

5

76

68

78

6

2

0

2

7

58

55

66

8

84

78

90

9

38

35

41

10

18

10

26

11

8

4

11

12

68

63

76

13

59

54

64

14

73

63

81

15

42

41

43

Fig. 6 Comparison of success rate of transaction considering different hashing techniques

increased by making use of an advanced hashing mechanism. Excessive security has been used to remove some invalid smart contracts. As a result of the suggested work, security and reliability both have been enhanced. The hashing is done by the blockchain technology during block creation. Hashing algorithms such as SHA256 and MD5 are only a few examples. It has been found that SHA256 is employed extensively in blockchains. When selecting a hashing method, a number of criteria, including storage space, time complexity, and collision ratio, are taken into account.

348 Table 5 Comparison of time consumption during execution of different hashing techniques

B. K. Aggarwal et al. Smart contract

SHA 256

MD5

RSA Integrated MD5

1

196.43939

205.95019

55.826479

2

201.54868

216.38711

59.271071

3

220.96657

263.40521

38.623334

4

220.80778

234.26493

41.236319

5

204.00119

245.58745

36.394729

6

223.24887

273.09077

41.498342

7

270.72091

251.10353

34.836661

8

208.30354

224.91529

54.050097

9

249.03755

257.04228

41.905234

10

271.40838

239.73005

42.803357

11

239.51108

228.7882

59.45131

12

236.38767

291.17711

45.556797

13

210.26131

286.17963

47.653667

14

231.33028

266.7778

57.426786

15

270.4546

227.27734

51.215905

Fig. 7 Comparison of time consumption during execution of different hashing techniques

The SHA256 algorithm is the most collision-resistant and consumes the least amount of space. A modified MD5 algorithm is used in the proposed study. Aside from the fact that MD5 is less secure than SHA256, it takes up less storage space. The goal of this study is to create a faster and more collision-resistant version of the MD5 algorithm. Therefore, a system that can be both secure and efficient is required. In this study, the revised MD5 was simulated for storage capacity, collision probability,

Enhancement of Smart Contact on Blockchain Security …

349

and CPU clock cycles. As a consequence, the improved MD5 results surpass MD5 and SHA256 in terms of collision resistance while using less storage and time.

7 Future Scopes Considering demand and need of blockchain, present work is need of the day. However, the scope of artificial intelligence is wide but present research laid significant foundation to put intelligence in execution of smart contracts. Such smart contracts could be used for identity protection and health care systems. In cryptocurrency, blockchain plays a vital role. There have been a number of different uses for blockchain, such as healthcare. Furthermore, blockchain is commonly used in the cloud for distributed computing. In the future, the security method presented in the study might be advantageous for such infrastructures.

References 1. M. Nofer, P. Gomber, O. Hinz, D. Schiereck, Blockchain. Bus. Inf. Syst. Eng. 59(3), 183–187 (2017). https://doi.org/10.1007/s12599-017-0467-3 2. Y. Chen, Y. Zhang, B. Zhou, Research on the risk of block chain technology in Internet finance supported by wireless network. EURASIP J. Wirel. Commun. Netw. 2020:71 (2020), https:// doi.org/10.1186/s13638-020-01685-6 3. Q. Zhang, X. Zhang, Research on the Application of Block Chain Technology in Internet Finance, vol. 885. Springer International Publishing (2019) 4. L. Stephan, S. Steffen, S. Moritz, G. Bela, A review on blockchain technology and blockchain projects fostering open science. Front. Blockchain (2019) 5. I.A. Hassan, R. Ahammed, M.M. Khan, N. Alsufyani, A. Alsufyani, Secured insurance framework using blockchain and smart contract. 2021 (2021) 6. G.W. Petersz, E. Panayiy, Understanding modern banking ledgers through blockchain technologies: Future of transaction processing and smart contracts on internet of money (2015) 7. P. Eze, Eziokwu, A triplicate smart contract model using blockchain technology. Circ. Comput. Sci. DC CPS 1–10 (2017). https://doi.org/10.22632/ccs-2017-cps-01 8. A.P. Ratna, P.D. Purnamasari, A. Shaugi, M. Salman, Analysis and comparison of MD5 and SHA-1 algorithm implementation in Simple-O authentication based security system, in 2013 International Conference Qualitative Research QiR 2013—Conjunction with ICCS 2013 2nd International Conferences Civil Sp., 2013, pp. 99–104 9. S, Prabhav, Trusted execution environment and linux a survey. Int. J. Comput. Trends Technol. (2017) 10. J. Yli-Huumo, D. Ko, Where is current research on blockchain technology? A Syst. Rev. (2016) 11. I. Zikratov, A. Kuzmin, V. Akimenko, V. Niculichev, L. Yalansky, Ensuring data integrity using blockchain technology, in Proceeding of 20th Conference of Fruct Association (2017) 12. Z. Zheng, S. Xie, H. Dai, X. Chen, H. Wang, An overview of blockchain technology: architecture, consensus, and future trends. IEEE, 2017 13. S. Gueron, Speeding up SHA-1, SHA-256 and SHA-512 on the 2nd generation Intel® CoreTM processors, in Proceeding of the 9th International Conference Information Technolonogy ITNG 2012 (2012), pp. 824–826

350

B. K. Aggarwal et al.

14. R. Raju, S. Aravind Kumar, R. Manikandan, Avoiding data replication in cloud using SHA2, in 7th IEEE International Conference on Computation of Power, Energy, Information and Communication ICCPEIC 2018 (2018), pp. 210–214 15. S. Purwanti, B. Nugraha, M. Alaydrus, Enhancing security on E-health private data using SHA512, in 2017 International Conference on Broadband and Wireless Sensors Powering, BCWSP 2017, vol. 2018-January (2018), pp. 1–4

Evaluation of Covid-19 Ontologies Through OntoMetrics and OOPS! Tools Narayan C. Debnath, Archana Patel, Debarshi Mazumder, Phuc Nguyen Manh, and Ngoc Ha Minh

Abstract Ontology provides a way to encode human intelligence so that machines can understand and make decisions by referring to this intelligence. For this reason, ontologies are used in every domain, specifically in the domains that relate to the emergency situation. Covid-19 became a serious concern and emerged as the most significant emergency for the world. Many Covid-19 ontologies are available on the Web to analyse the Covid-19 data semantically. However, a few questions arise from work done so far: How many Covid-19 ontologies are available? Which is a good Covid-19 ontology in terms of richness? What is the pitfall rate of the available Covid19 ontologies? This paper focuses on these questions by providing a comprehensive survey on available Covid-19 ontologies. By this paper, analysts and researchers find a road map, an overview of research work that exists in terms of Covid-19 ontologies. Keywords Ontology · Covid-19 · Ontology evaluation · Ontology pitfall scanner! · OntoMetrics

1 Introduction Ontology is a semantic model that represents the reality of a domain in a machineunderstandable manner. The basic building block of an ontology is classes, relationships, axioms, and instances [1]. The axioms (a statement is taken to be true, to act as N. C. Debnath · A. Patel (B) · D. Mazumder · P. N. Manh · N. H. Minh Department of Software Engineering, Eastern International University, Binh Duong, Vietnam e-mail: [email protected] N. C. Debnath e-mail: [email protected] D. Mazumder e-mail: [email protected] P. N. Manh e-mail: [email protected] N. H. Minh e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 I. J. Jacob et al. (eds.), Expert Clouds and Applications, Lecture Notes in Networks and Systems 444, https://doi.org/10.1007/978-981-19-2500-9_25

351

352

N. C. Debnath et al.

a premise for further reasoning) impose constraints or restrictions on sets of classes and relationships allowed among the classes. These axioms offer semantics because by using these axioms, machines can extract additional or hidden information based on data explicitly provided. Nowadays, ontologies are used everywhere due to the following reasons: (i) they simplify the knowledge sharing among the entities of the system; (ii) it is easier to reuse domain knowledge and (iii) they provide a convenient way to manage and manipulate domain entities and their interrelationships. W3C has standardized web ontology language (OWL) for the description of the facts in resource description framework (RDF) which can be formatted with syntaxes like N- triple, TURTLE, RDF/XML, RDF/OWL, etc. Web ontology language (OWL) is designed to encode rich knowledge about the entities and their relationships. OWL classes are built on top of and add additional semantics to RDFS classes. Whereas classes in most other languages only have heuristic definitions, OWL classes have a rigorous formal definition. Relations between individuals are described by OWL properties. There are three types of OWL properties: object properties, data properties, and annotation properties. One of the most powerful features of OWL is the ability to provide formal definitions of classes using the description logic (DL) language. These axioms are typically asserted on property values for the class. There are three kinds of axioms that can be asserted about OWL classes, namely quantifier restrictions, cardinality restrictions, and value restrictions. Ontologies are either developed by handcraft or by using ontology development tools. The protégé editor provides an environment for creating and maintaining ontologies [2]. It is open source and has many plugins. Several ontology repositories exist, namely OBO Foundry (contains biological science-related ontologies), bio portal (comprehensive repository of biomedical ontologies), agro portal (vocabulary and ontology repository for agronomy and related domains), and OLS (provides single-point access of the latest version of biomedical ontologies). These all repositories contain more than a thousand ontologies about a domain [3]. Users use these repositories to determine the appropriate ontology as per need. However, sometimes users get more than one ontology for the same domain; simultaneously, it is crucial to examine which ontology best meets the user requirement. Ontology evaluation is a way that determines the relevance and importance of the ontology in a specified domain. Ontology evaluation is an important process for the development and maintenance of an ontology. Many Covid-19 ontologies have been developed to analyse the Covid-19 data semantically. Now, the question is how to assess quality ontology among the available Covid-19 ontologies. The aim of this paper is to evaluate the available Covid-19 ontologies to find out the suitable ontology as per the need of the user. The major contributions include evaluation of richness, anomalies along with determining pitfall rate of Covid-19 ontologies. The rest of the paper is organized as follows: Sect. 2 shows the literature about the Covid-19. Section 3 looks at the available ontology development methodologies and Covid-19 ontologies. Section 4 focuses on the evaluation of the Covid-19 ontologies to check the richness, anomalies, and pitfalls rate. The last section concludes the paper.

Evaluation of Covid-19 Ontologies Through …

353

2 Literature To deal with complex highly connected big data it is essential that the data is understandable to humans as well as machines. This is where the web ontology language (OWL) is utilized. OWL defines the semantics of data in terms of ontologies which combine many different logical formalisms to represent data in an intuitive way. Many authors have proposed ontologies for analysing the Covid-19 data semantically. The authors González-Eras et al. [4] have proposed an ontology, namely COVID-19 pandemic ontology, by integrating of existing Covid-19 ontologies. They have also used ontologies from other domains to cover the broader aspects of the pandemic. For the development of COVID-19 Pandemic ontology, they have followed ontological mining approaches, which include ontology alignment, ontology linking or mapping, and ontology merging or fusion or mixing. They have evaluated the developed ontology by competency questions, ontological metrics, and protégé reasoner. The authors Kachaoui et al. [5] have developed a methodology for the acquisition of knowledge from data lake for the development of intelligent systems. The proposed methodology has four phases, namely Acquisition layer (loading of data from different sources), Exploration layer (analyse and pre-process the data to make it clear), Semantic layer (prepare new dataset in the form of ontology), and Insight layer (create any type of insights). They have also addressed three questions: the detection of the contaminated person of Covid-19, to avoid and reduce disease propagation, Can the government use big data to prevent Covid-19. The authors FonouDombeu et al. [6] have developed an ontology called COVID-19 ontology (COVIDonto) that extends the existing ontologies and provides more information about the origin, symptoms, treatment, and spread of Covid-19. The NeOn methodology are used for the development of this ontology and data is collected from different sources to reflect the different aspects of Covid-19. They have reused the resources from biomedical ontologies therefore, the developed ontology can be easily integrated with other ontologies. The authors Sayeb et al. [7] have proposed an actor centred approach to strengthen the ability of health care information system (HIS) by providing a precise definition of desired acts, identifying the component, and measuring the performance of actors. They have developed C3HIS ontology by using the protégé tool and contain two aspects—crisis and actors. They have shown the application of the C3HIS ontology to manage healthcare services in HIS. The authors Kouamé and Mcheick [8] have developed a COVID-19 ontology model called SuspectedCOPDcoviDOlogy along with an alert system that identifies COPD patients with Covid-19. The SuspectedCOPDcoviDOlogy ontology contains five ontologies, namely evaluation vital sign ontology (it identifies the key terms of the domain), questionnaire ontology (it contains questions that need to be answered by the person), symptom COVID-19 ontology (the answer of the questionnaire form is extracted by this ontology), alert ontology (it contains all alerts message), and service ontology (it provides service like sending the alert messages or emails). They have utilized the SWRL reasoning engine and OWL/DL tool for generating alerts. When Covid-19 results are positive, the system automatically generates a questionnaire

354

N. C. Debnath et al.

form and sends it to the staff or patient. The authors Ahmad et al. [9] have provided a survey on the ontologies and tools that support the analytics of Covid-19. They have followed the systematic mapping studies (SMS) to collect and analyse the documents for the review. Ontology can be developed either from scratch or by modifying an existing ontology. In both cases, approaches for evaluating the quality and quantity of ontology are necessary. Ontology evaluation provides the approaches and criteria to examine the quality and quantity of the ontology, which also concludes the best fit ontology in the specified domain as per requirement. The authors Raad and Cruz [10] have proposed following criteria: • Accuracy: It is determined by the definitions, descriptions of entities like classes, properties, and individuals. This criteria state that ontology is correct. • Clarity: It shows how effectively the meaning or definition of the term in the ontology is defined. The definition of a term should be independent of the context. • Completeness: It states that ontology covers complete information about a specified domain. • Adaptability: It measures the adaptability of an ontology. • Consistency: It shows that ontology does not have any contradictions. • Computational efficiency: It shows the flexibility of an ontology with the tools, specifically focus on the speed of the reasoner that infers the information from the ontology. Poveda-Villalón et al. [11] have proposed a tool called Ontology Pitfall Scanner! (OOPS!). It is a Web-based tool that shows the pitfalls or anomalies of an ontology. OOPS! shows the 41 types of pitfalls ranging from P01 to P41. Basically, OOPS! groups the pitfalls under three categories, namely minor pitfalls (these pitfalls are not serious and no need to remove them), important pitfalls (not very serious pitfalls but need to remove), and critical pitfalls (these pitfalls hamper the quality of an ontology and need to remove them before using ontology). These pitfalls are also classified by dimension and evaluation criteria—dimension (OOPS! tool has three types of dimension pitfalls, namely structural dimension, functional dimension, and usability-profiling dimension) and evaluation criteria (OOPS! determines the pitfalls under the three ontology evaluation criteria, namely consistency, completeness, and conciseness). The authors Lozano-Tello and Gómez-Pérez [12] have developed OntoMetrics tool. It is a Web-based tool that calculates the statistical information about an ontology. The current version of the OntoMetric tool is available at https://ontome trics.informatik.uni-rostock.de/ontologymetrics/. It has five types of metrics, namely base metrics, schema metrics, knowledge base metrics, class metrics, and graph metrics. As of now, many articles based on Covid-19 ontologies are available. However, a review of the current landscape shows that Covid-19 ontologies are not yet evaluated in terms of richness, anomalies, and pitfall rate that create a problem for choosing the suitable ontologies as per the need of the user. To overcome this problem, the paper evaluates the existing ontologies designed specifically for the Covid-19 context.

Evaluation of Covid-19 Ontologies Through …

355

3 Ontology at a Glance This section focusses on the ontology development methodologies and available Covid-19 ontologies that are accessible and downloadable.

3.1 Ontology Development Methodologies In the literature, various authors have developed ontologies for the semantical analysis of Covid-19 data. However, the major problem for the ontology developers is to choose the right methodology that builds correct, complete, and concise ontology as per requirement. Ontology development methodology describes the step-by-step process for the ontology development. The most famous used methodologies are TOVE, Enterprise Model Approach, METHONTOLOGY, and KBSI IDEF5 [13]. These methodologies have various steps for the ontology development, and some steps are common among them. Figure 1 shows the relationship among these methodologies via arrows (double-headed arrow). The most important step of ontology development is to identify the purpose and fix the boundary/scope of the ontology. This can be achieved by writing competency questions and motivation scenarios. The next step is to collect and analyse the information based on the defined scope and boundary of an ontology. For this purpose, ontology developers use different sources or repositories and conduct the interview with domain experts to collect the desired information. This step is also known as knowledge acquisition. After having required information, now need to formalize it. For this purpose, first, recognize the classes and their properties. The formal competency questions can be utilized here to identify the entities as classes, properties, and instances. Now, start to encode the ontology and imposed constraints on the classes and their properties as required. Ontology evaluation is the vital steps of ontology development methodologies. It shows completeness (ontology must contain all the required information as per domain need) and accuracy (ontology must be free from

Fig. 1 Most popular and extensively used ontology development methodologies

356

N. C. Debnath et al.

anomalies and is able to infer the correct answer). The last step of ontology development methodology is to document the ontologies that can become the base of other activities. Apart from these methodologies, three ontology development methodologies, namely Neon [14], YAMO [15], and SAMOD [16], are also available in the literature. The existing Covid-19 ontologies have been developed based on these above methodologies.

3.2 Available Covid-19 Ontologies Many ontologies have been developed that contain Covid-19-related data. However, only a few ontologies are available that have specifically designed for this pandemic. These ontologies are listed below: 1.

2.

3.

4.

5.

6.

7.

8.

An Ontology for Collection and Analysis of COviD-19 Data (CODO) [17]: The CODO ontology is a data model that publishes Covid-19 data on the Web as a knowledge graph. The CODO aims to show the patient data and cases of Covid-19. The latest version of COVID-19 was released in Sept 2020. COKPME [18]: This ontology is used to analyse the precautionary measures that help in controlling the spread of Covid-19. COKPME ontology is able to handle the various competence questions. The latest version of COKPME was released in Sept 2021. COVID-19 Surveillance Ontology (COVID-19) [19]: This ontology supports surveillance activities and is designed as an application ontology for the Covid19 pandemic. The developed COVID-19 surveillance ontology ensures transparency and consistency. The latest version of COVID-19 was released in May 2020. Long Covid Phenotype Ontology (LONGCOVID) [20]: It is RCGP RSC Long Covid Phenotype ontology. The latest version of LONGCOVID was released in Oct 2021. The COVID-19 Infectious Disease Ontology (IDO-COVID-19) [21]: It is an extension of two ontologies, namely IDO and VIDO, and provides integration and analysis of Covid-19 data. The latest version of IDO-COVID-19 was released in June 2020. WHO COVID-19 Rapid Version CRF semantic data model (COVIDCRFRAPID) [22]: It is a semantic data model of the Covid-19 cases provided by the WHO. This ontology aims to provide semantic references to the questions and answers of the form. The latest version of COVIDCRFRAPID was released in June 2020. COVID-19 Ontology [23]: It covers the wide spectrum of medical and epidemiological concepts linked to COVID-19. The latest version of COVID-19 ontology was released in May 2020. COVID-19OntologyInPatternMedicine (COVID-19-ONT-PM) [24]: It provides scientific findings of the Covid-19 that help to control the outbreak

Evaluation of Covid-19 Ontologies Through …

9.

357

of the pandemic. This ontology has the ability to make medical decisions in a systematic way. The latest version of COVID-19-ONT-PM was released in August 2021. Coronavirus Infectious Disease Ontology (CIDO) [25]: It is a communitybased ontology that imports Covid-19 pandemic-related concepts from the IDO ontology. CIDO is a more specific standard ontology as compared to IDO and encodes knowledge about the coronavirus disease as well as provides integration, sharing, and analysis of the information. The latest version of CIDO was released in August 2021.

4 Evaluation of Covid-19 Ontologies Ontology evaluation provides approaches and criteria to examine the quality and quantity of the ontology, which also concludes the best fit ontology in the specified domain as per requirement. The ontology evaluation aims to determine and check the following points: • Which ontology is optimal for the user among other available ontologies? • Which ontology has richer attribute values compared to other available ontologies? • Which ontology has no anomalies or errors? This section evaluates the richness, anomalies, and pitfall rate of the available Covid-19 ontologies. We use the OntoMetrics tool to examine the richness of an ontology and OOPS! tool for anomalies detection and pitfall rate.

4.1 OntoMetrics Tool It is a tool that evaluates ontology quantitatively. It measures the quantity (number of ontological attributes) of the ontology based on five metrics, namely base metrics, schema metrics, knowledge metrics, class metrics, and graph or structure metrics. These all metrics evaluate the different aspects of an ontology. The base metrics show the quantity of the ontology element, and they consist of simple metrics like classes, axioms, data property, object property, instances, etc. The schema metrics are used to describe the design of the ontology by indicating attribute richness, relationship richness, inheritance richness, etc. The knowledge metrics measure the amount of data that is encoded inside the ontology. It explains the effectiveness of the ontology design by counting instances of an ontology. The class metrics measure the classes and relationships of the ontology. The graph or structure metrics explain the structure of the ontology by examining the cardinality, depth, breadth, total number of paths, etc. We evaluate all the Covid-19 ontologies with respect to all these metrics, which examine the richness of these ontologies. Table 1 depicts the value of these metrics

358

N. C. Debnath et al.

Table 1 (a) and (b) shows the richness of the Covid-19 ontologies (a) Ontologies →Metrics ↓ Base metrics Axioms

CODO

COKPME

COVID-19

Long COVID

IDO-COVID-19

2047

406

165

33

5018

Logical axioms count

917

161

32

21

1032

Class count

91

26

33

13

486

Object properties

73

19

0

0

43

Data properties

50

19

0

0

0

Instances

271

41

0

0

94

Attribute richness

0.549

0.730

0.0

0.0

0.0

Inheritance richness

1.010

1.115

0.969

1.307

1.242

Relationship richness

0.471

0.472

0.0

0.190

0.340

Attribute class 0.0 ratio

0.0

0.0

0.0

0.0

Equivalence ratio

0.098

0.269

0.0

0.0

0.234

Axiom/class ratio

22.49

15.615

5.0

2.538

10.325

Inverse relations ratio

0.211

0.142

0.0

0.0

0.395

Class/relation ratio

0.522

0.472

1.031

0.619

0.530

Knowledge base metrics

Average population

2.978

1.576

0.0

0.0

0.193

Class richness 0.307

0.153

0.0

0.0

0.004

Graph metrics

Absolute root cardinality

1

1

1

1

1

Absolute leaf cardinality

67

17

28

4

306

Absolute sibling cardinality

91

26

33

13

486

Absolute depth

327

84

93

210

3588

3.230

2.818

6.176

7.382

Schema metrics

Average depth 3.59

(continued)

Evaluation of Covid-19 Ontologies Through …

359

Table 1 (continued) (a) Ontologies →Metrics ↓

CODO

COKPME

COVID-19

Long COVID

IDO-COVID-19

Maximal depth

6

5

3

8

13

Absolute breadth

91

26

33

34

486

Average breadth

3.64

2.6

5.5

2.428

2.685

Maximal breadth

13

6

11

5

18

Ratio of leaf fan-outness

0.736

0.653

0.848

0.307

0.629

Ratio of sibling fan-outness

1.0

1.0

1.0

1.0

1.0

Tangledness

0.021

0.076

0.0

0.307

0.213

Total number of paths

91

26

33

34

486

Average number of paths

15.16

5.2

11.0

4.25

37.384

(b) Ontologies →Metrics ↓

COVIDCRF RAPID

COVID-19

COVID-19-ONT-PM

CIDO

Base metrics

Axioms

6684

41,121

1232

134,742

Logical axioms count

1699

2630

472

28,118

Class count

399

2271

365

8775

Object properties

6

9

14

363

Data properties

7

1

1

18

instances

495

6

6

3646

Attribute richness

0.017

4.4E-4

0.002



Inheritance richness

1.932

1.153

1.257



Relationship richness

0.007

0.007

0.029



Schema metrics

(continued)

360

N. C. Debnath et al.

Table 1 (continued) (b) Ontologies →Metrics ↓

Knowledgebase metrics Graph metrics

COVIDCRF RAPID

COVID-19

COVID-19-ONT-PM

CIDO

Attribute class 0.0 ratio

0.0

0.0



Equivalence ratio

0.0

0.003

0.0



Axiom/class ratio

16.751

18.107

3.375



Inverse 0.0 relations ratio

0.0

0.0



Class/relation ratio

0.513

0.860

0.771



Average population

1.240

0.002

0.016



Class richness 0.057

0.0

0.002



Absolute root cardinality

1

1

1



Absolute leaf cardinality

344

1556

285



Absolute sibling cardinality

399

2271

365



Absolute depth

2092

26,269

2184



Average depth 5.23

9.555

5.983



Maximal depth

8

17

9



Absolute breadth

400

2749

365



Average breadth

7.142

2.899

4.506



Maximal breadth

91

247

36



Ratio of leaf fan-outness

0.862

0.685

0.780



Ratio of sibling fan-outness

1.0

1.0

1.0



Tangledness

0.666

0.130

0.079



Total number of paths

400

2749

365

– (continued)

Evaluation of Covid-19 Ontologies Through …

361

Table 1 (continued) (b) Ontologies →Metrics ↓ Average number of paths

COVIDCRF RAPID

COVID-19

COVID-19-ONT-PM

CIDO

50.0

161.705

40.555



for Covid-19 ontologies. The highest number of classes (8775) and object properties (363) are contained by CIDO ontology; however, the highest data properties (50) lie with CODO ontology among the existing Covid-19 ontologies. Hence, it is stated that CIDO ontology is rich in terms of classes and object properties, whereas CODO ontology is rich in terms of data properties.

4.2 Ontology Pitfall Scanner! (OOPS!) OOPs is a criteria-based tool that evaluates the ontology qualitatively, and it is available on the Web. This tool shows the errors or anomalies of the ontology under three categories, namely minor pitfalls (these are not serious concerns and no need to remove them), important pitfalls (these are moderate pitfalls, and it is good practice to remove them before using that ontology), and critical pitfalls (these pitfalls must be removed from the ontologies otherwise will provide a wrong reasoning results). OOPS! tool shows 41 types of different pitfalls ranging from P01 to P41 and their descriptions. Table 2 depicts the pitfalls that are available in the Covid-19 ontologies. The description of the obtained pitfalls is mentioned below. • Minor pitfalls: P04: Ontology has unconnected elements, P07: Concepts are merged in the same class, P08: Annotation properties are missing, P13: Property inverse is not defined explicitly, P20: Properties annotations are not used properly, P21: Ontology uses a miscellaneous class. P22: Ontology has different naming conventions, P32: Same labels are assigned to several classes. • Important Pitfalls: P10: Ontology lacks disjoint axioms between classes or between property, P11: Domain and range properties are missing, P24: Ontology uses recursive definitions, P25: Ontology relationship is defined as the inverse of itself. P30: Equivalence classes are not defined explicitly, P34: Ontology has untyped class, P38: owl:ontology tag is not declared, P41: No license declares. • Critical Pitfalls: P19: Ontology has multiple domains and ranges in properties. The numbers (e.g. 1, 2, 4,..) that are contained in Table 2 denote the total number of cases in accordance with the given pitfalls, like CODO ontology contains 1 case of pitfall P04. The sign × indicates that no pitfall case is available in the respective ontology. The pitfall describes the number of features that could create problems during reasoning. We have calculated the pitfall rate by using the following formula

1

×

1

P21

P22

P19

Critical pitfalls

×

3

×

×

P41

×

1

×

1

×

×

7

1

P34

×

×

2

P30

P38

×

×

×

×

4

4

×

1

1

×

×

×

×

× ×

1

×

1

14

P24

P11

×

×

×

× ×

×

×

1

P25

1

58

P10

Important pitfalls

P32

×

×

3

P20

×

12 ×

×

×

14

15

58

38

P08

×

1

LONGCOVID

×

4

COVID-19

P13

2

×

1

×

COKPME

P04

CODO

P07

Minor pitfalls

Ontologies → Pitfalls ↓

Table 2 Obtained pitfalls of Covid-19 ontologies

×

1

1

×

1

×

×

24

×

×

1

×

4

7

×

×

1

IDO-COVID-19

×

×

1

×

4

×

1

11

1

40

1

10

3

5

77

5

5

COVIDCRF RAPID

×

×

×

×

2129

×

×

9

×

×

1986

×

×

173

187

×

1

COVID-19

×

1

1

×

298

×

×

14

×

×

×

×

×

29

238

×

1

COVID-19-ONT-PM

×

×

×

×

7146

×

×

236

×

×

4376

×

×

259

7859

×

1

CIDO

362 N. C. Debnath et al.

Pitfall Rate

Evaluation of Covid-19 Ontologies Through … 0.5 0.45 0.4 0.35 0.3 0.25 0.2 0.15 0.1 0.05 0

363 0.472

0.454

0.147

0.125

0.109

0.086 0.042 0.007

0.024

Covid-19 Ontologies

Fig. 2 Pitfalls rate of Covid-19 ontologies

n i=1

Pi

N Pi represents the total number of pitfall cases according to the pitfall type Pi , and N is the total number of tuples (ontology size). The high value of the pitfall rate implies a more significant number of anomalies and vice versa. The pitfalls range lies between 0 and 1. Figure 2 shows the pitfalls rate of the available Covid-19 ontologies. The COVID-19-ONT-PM ontology has 0.472 pitfall rate, which is the highest as compared to other Covid-19 ontologies. COKPME ontology contains three cases of pitfall P19 which is a critical pitfall. So, before using COKPME ontology, it is mandatory to remove this pitfall.

5 Conclusion and Future Work It is challenging to choose a suitable ontology as many ontologies available in the same domain. The ontology evaluation approaches assess the quality of ontology among the available ontologies to get the most feasible ontology as per the need of the user. In this paper, we have collected those ontologies which are specifically designated in the Covid-19 context and then evaluated them in terms of richness, anomalies, and pitfall rate. OntoMetrics tool has been used for the calculation of the richness, and OOPS! tool has been utilized for the detection of anomalies. The evaluation results show that available Covid-19 ontologies are rich in terms of attributes; however, they are not free from anomalies. Since anomalies alter the reasoning results, the user must remove them before using ontology in an application. In the future, an attempt should be made to evaluate ontologies using software engineering process.

364

N. C. Debnath et al.

Acknowledgements This research is financially supported by Eastern International University, Binh Duong Province, Vietnam.

References 1. A. Patel, N.C. Debnath, A.K. Mishra, S. Jain, Covid19-IBO: A Covid-19 impact on Indian banking ontology along with an efficient schema matching approach. N. Gener. Comput. 39(3), 647–676 (2021) 2. M. Horridge, H. Knublauch, A. Rector, R. Stevens, C. Wroe, A practical guide to building OWL ontologies using the Protégé-OWL plugin and CO-ODE tools edition 1.0. Univ. Manchester (2004) 3. A. Patel, S. Jain, Ontology versioning framework for representing ontological concept as knowledge unit, in International Semantic Intelligence Conference, vol. 2786 (2021) 4. A. González-Eras, R. Dos Santos, J. Aguilar, A. Lopez, Ontological engineering for the definition of a COVID-19 pandemic ontology. Inform. Med. Unlocked 100816 (2021) 5. J. Kachaoui, J. Larioui, A. Belangour, Towards an ontology proposal model in data lake for real-time COVID-19 cases prevention (2020) 6. J.V. Fonou-Dombeu, T. Achary, E. Genders, S. Mahabeer, S.M. Pillay, COVIDonto: An ontology model for acquisition and sharing of COVID-19 data, in International Conference on Model and Data Engineering (Springer, Cham, 2021), pp. 227–240 7. Y. Sayeb, M. Jebri, H.B. Ghezala, Managing COVID-19 crisis using C3HIS ontology. Procedia Comput. Sci. 181, 1114–1121 (2021) 8. K.M. Kouamé, H. Mcheick, An ontological approach for early detection of suspected COVID19 among COPD patients. Appl. Syst. Innovation 4(1), 21 (2021) 9. A. Ahmad, M. Bandara, M. Fahmideh, H.A. Proper, G. Guizzardi, J. Soar, An overview of ontologies and tool support for COVID-19 analytics, in 2021 IEEE 25th International Enterprise Distributed Object Computing Workshop (EDOCW) (IEEE, 2021), pp. 1–8 10. J. Raad, C. Cruz, A survey on ontology evaluation methods, in Proceedings of the International Conference on Knowledge Engineering and Ontology Development, Part of the 7th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management (2015). 11. M. Poveda-Villalón, A. Gómez-Pérez, M.C. Suárez-Figueroa, Oops!(ontology pitfall scanner!): An on-line tool for ontology evaluation. Int. J. Semant. Web Inf. Syst. (IJSWIS) 10(2), 7–34 (2014) 12. A. Lozano-Tello, A. Gómez-Pérez, Ontometric: A method to choose the appropriate ontology. J. Database Manag. (JDM) 15(2), 1–18 (2004) 13. R. Iqbal, M.A.A. Murad, A. Mustapha, N.M. Sharef, An analysis of ontology engineering methodologies: A literature review. Res. J. Appl. Sci. Eng. Technol. 6(16), 2993–3000 (2013) 14. M.C. Suárez-Figueroa, A. Gómez-Pérez, M. Fernandez-Lopez, The NeOn methodology framework: A scenario-based methodology for ontology development. Appl. Ontol. 10(2), 107–145 (2015) 15. B. Dutta, U. Chatterjee, D.P. Madalli, YAMO: yet another methodology for large-scale faceted ontology construction. J. Knowl. Manag. (2015) 16. S. Peroni, SAMOD: an agile methodology for the development of ontologies (2016) 17. CODO Ontology, https://bioportal.bioontology.org/ontologies/CODO 18. COKPME Ontology, https://bioportal.bioontology.org/ontologies/COKPME 19. COVID19 Ontology, https://bioportal.bioontology.org/ontologies/COVID19 20. LONGCOVID Ontology, https://bioportal.bioontology.org/ontologies/LONGCOVID 21. IDO-COVID-19 Ontology, https://bioportal.bioontology.org/ontologies/IDO-COVID-19

Evaluation of Covid-19 Ontologies Through …

365

22. COVIDCRFRAPID Ontology, https://bioportal.bioontology.org/ontologies/COVIDCRFR APID 23. COVID-19 Ontology, https://bioportal.bioontology.org/ontologies/COVID-19 24. COVID-19-ONT-PM Ontology, https://bioportal.bioontology.org/ontologies/COVID-19ONT-PM 25. CIDO Ontology, https://bioportal.bioontology.org/ontologies/CIDO

Recognition of the Multioriented Text Based on Deep Learning K. Priyadarsini, Senthil Kumar Janahan, S. Thirumal, P. Bindu, T. Ajith Bosco Raj, and Sankararao Majji

Abstract The development and use of systems for analyzing visuals, such as photos and videos, using benchmark datasets is a difficult but necessary undertaking. DNN and STN are employed in this study to solve the challenge at hand. The study’s network design consists of a localization and recognition network. The localization network generates a sampling grid and locates and localizes text sections. In contrast, text areas will be entered into the recognition network, and this network will then learn to recognize text, including low resolution, curved, and multi-oriented text. Street View house numbers and the 2015 International Conference on Document Analysis and Recognition were used to gauge the system’s performance for this study’s findings (ICDAR). Using the STN-OCR model, we are able to outperform the literature. Keywords Spatial transformer networks · Deep neural networks · Recognition · STN-OCR · Multi-oriented text etc

K. Priyadarsini Department of Data Science and Business Systems, School of Computing, College of Engineering and Technology, SRM Institute of Science and Technology, Kattankulathur, Chennai, India S. K. Janahan (B) Department of CSE, Lovely Professional University, Phagwara, Punjab, India e-mail: [email protected] S. Thirumal Department of Computer Science and Engineering, Vels Institute of Science Technology and Advanced Studies, Chennai, India P. Bindu Department of Mathematics, Koneru Lakshmaiah Education Foundation, Vaddeswaram, AP, India T. A. B. Raj Department of Electronics and Communication Engineering, PSN College of Engineering and Technology, Tirunelveli, Tamil Nadu, India S. Majji Department of Electronics and Communication Engineering, GRIET, Hyderabad, India © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 I. J. Jacob et al. (eds.), Expert Clouds and Applications, Lecture Notes in Networks and Systems 444, https://doi.org/10.1007/978-981-19-2500-9_26

367

368

K. Priyadarsin et al.

1 Introduction Increasing demand for numerous computer vision jobs has pushed this community to focus on reading text in the wild (from scene photos). Despite substantial research in the last few years, finding text in uncontrolled contexts remains a difficult task [1, 2]. Even more challenging is recognizing text lines with random orientation, which takes into account a substantially greater number of hypotheses, which significantly expands the search field. In most cases, existing methods are able to recognize text that is either horizontal or close to horizontal. But when applied to multi-oriented text, the results of the recent ICDAR2015 competition for text detection show that there is still a considerable disparity. Text has a fairly different appearance and shapes when compared to generic objects since it can be handled as a sequence-like object with unlimited lengths [3–5]. This has led to the widespread use of scene image identification systems based on sliding windows and related components. ICDAR2013 and ICDAR2015 contests saw state-of-the art performance from component-based approaches using Maximally Stable Extremal Regions (MSER) as the fundamental representations. An extremely resilient representation of character components was recently learned through the use of a convolution neural network (CNN). To localize a word or a line of text, clustering algorithms or some sort of heuristic approach is usually required for this. Directly hits text lines from crowded photos, taking advantage of their symmetry and self-similarity features [6]. Text detection appears to need the use of both character components and text regions.

2 Related Work Natural picture text identification has piqued the curiosity of computer vision and document analysis professionals. Horizontal or near-horizontal-based text detection is the primary focus of most techniques. In order to create an end-to-end text recognition system, the first step is to locate word boundaries [7]. Here, we’ll take a look at some of the best examples of multi-oriented text detection. Lu et al. [8]. were the first to look at real-world multi-oriented text detection. Conventional detection pipelines can be compared to those that use connected component extraction and text line orientation estimation. Kang, et al. [8] turned the text identification problem into a graph partitioning problem by treating each MSER component as a node in a network. It has been proposed by Wei et al. [9]. Yin and others use multi-stage clustering methods in order to recognize multi-oriented text [10]. For multi-oriented text, an SWT-based end-to-end system was proposed by Yao and his colleagues. The ICDAR2015 text detection competition just announced a hard benchmark for multi-oriented text identification, and numerous academics have presented their results on it [11].

Recognition of the Multioriented Text Based on Deep …

369

3 Methodology It reads line by line, character by character, just like a person would using the STNOCR model. This human-like method to text analysis is no longer employed by text detection and recognition algorithms. An image is processed in its entirety, allowing these systems to retrieve all relevant information at once. Textual sections are found and localized progressively in photos using a human-based technique, and then recognized [12, 13]. Text detection and recognition are part of a Deep Neural Network (DNN) model that was created in this regard. This section focuses on the text detection stage’s attention mechanism and the complete approach for STN-OCR (Fig. 1). A.

Text Detection with Spatial Transformers

Jaderberg et al. employed the Spatial Transformer, a Deep Neural Networks learnable module that receives input I, spatially modifies the input feature map, and then outputs an output feature map O. This shift in location is made up of three main components. The initial part of the localization network computes the function f_ loc, which predicts the spatial transformation parameters 8. Based on projected parameters, the second portion generates a sample grid [14]. This portion generates the sampling grid, which is then delivered as input to the learnable interpolation algorithm in the third section, which produces the altered feature map O as an output. Part by part, we’ll go through all you need to know in this area. • Localization network: For example, an input feature map with dimensions such as height and width are fed into the localization network, which creates output parameters such as spatial transformation. The network of localization will locate and localize N letters, words, or lines of text. It will be necessary to use an affine transformation matrix in order to apply rotations, translations, skew, and zoom to the input feature map in order to achieve oriented text detection. When it comes to text rotation, translation, and zoom, this system has a lot to learn.

Fig. 1 Text detection and recognition using STN-OCR

370

K. Priyadarsin et al.

STN-OCR uses a feed-forward CNN and an RNN to generate N affine transformation matrices. This network of localization makes use of the CNN model ResNet50. The system’s performance is superior to that of other network architectures, such as VGGNet, while using this network structure. As a result, it overcomes the problem of vanishing gradient and maintains a higher level of accuracy than alternative network structures. For the experiments, Batch Normalization was employed, and subsequently, RNN was used for the rest of the study. In this case, the RNN is a Bi-directional LSTM RNN. Hidden states are used to predict affine transformation matrices. BLSTM is primarily responsible for generating concealed states. • Localization Network Configuration ResNet architecture, or residual neural network, is utilized in the localization network. Pictures from this study will be sent to the network, which will then use them to locate the corresponding texts. The first layer of the network will use 32 filters to do a 3 × 3 convolution, the second layer will use 48 filters to accomplish the same convolution, and the third layer will use 48 filters to perform the same convolution. This process is followed by Batch Normalization and averaging 2 × 2 and stride two for each convolution layer. In each layer, ReLU is employed as an activation function. Batch Normalization is applied after each layer, followed by the usage of two residual layers with 3 × 3 convolution. Finally, a BLSTM with 256 neurons was applied to the last residual layer. A sampling grid with bounding boxes (BBoxes) retrieved for textual portions is constructed after the aforesaid model. Only the textual portion of the document, as seen in Fig. 2, is used to generate BBoxes.

Fig. 2 Localization network

Recognition of the Multioriented Text Based on Deep …

371

• GRID Generation Using the feature map as input, the system creates n grids of the input feature map I using grid G0 and the coordinates xw0 and yh0. This stage generates a total of N output grids, including the BBoxes of the network-located textual parts. • Image sampling The values of feature map I were sampled at their corresponding coordinates on each of the N grids after N sampling grids were created with the grid generator in the second section. I can’t fit these spots into the feature map grid since they don’t make sense. As a result, bi-linear sampling was employed to select individuals from the regions that are closest to the centre of the population. The grid generator and picture sampler in action can be seen in Fig. 2. To choose picture pixels at a certain point in an image, the image sampler uses the grids generated by the grid generator. The vertices of the sampling grids are used to generate BBoxes automatically by this technique. Hence, the Spatial Transformer is formed by merging these three components: localization network, grid generation, and picture sampling, and can be employed in any region of a Deep Neural Network. This system begins with Spatial Transformer. B.

Text Recognition

It returns N textual areas retrieved from the input image as a result of the text detection stage This stage of text recognition treats each of the N regions separately. CNN handles the processing of N regions. Because ResNet has been shown to produce better outcomes in text recognition systems, a ResNet variant is also used in this CNN. Text recognition is required to produce strong gradients for text detection. Probability distributions over label space are estimated at this stage. Probability distributions can be predicted using Softmax classifiers. X n = On

(1)

ytn = soft max( f rec (x n ))

(2)

For example, we get the output f rec (x) after convolution feature extraction. Its configuration is identical to that of a localization network except for convolution filters. There are three convolutional layers totaling 32, 64, and 128 filters in this network. C.

Training Network

An image training set X and a text file for each individual image are used to train the network/model in ICDAR 2015. Coordinates for the top-left, top-right, bottom-right, and bottom-left coordinates of × 1 and y1, × 2 and y2, × 3, y3, × 4, and Y4 in each image are included in each file. After learning localization and detecting possible text possibilities in the first step, the model employs labeling to identify the specific piece of text.

372

K. Priyadarsin et al.

By calculating the loss of predicted text labels, error gradients are used to search and locate text regions. Some pre-training activities are required because we discovered that the model fails to merge multi-line texts into a picture. Optimizing the network during model training has a substantial influence. Adam optimizer is used after pre-training the network using Stochastic Gradient Descent (SGD) in order to improve the network’s performance on more difficult tasks. The learning rate in the first step of text detection is kept constant for a longer period of time. This is leading to an improved ability to locate and identify textual sections. As a result, SGD is employed, and it performs admirably in this context. The next stage of text recognition involves learning to recognize text sections that have already been predicted in a prior stage.

4 Results and Discussions What can be accomplished by utilizing this study structure is examined in this section. This investigation uses the ICDAR 2015 and SVHN benchmark datasets. A discussion of the aforementioned datasets follows. • ICDAR 2015 Dataset: Robust Reading Competition makes use of the ICDAR 2015 dataset, which includes 1500 images and over 10,000 annotations. A total of 1000 photographs are utilized in the training process, and a further 500 images are used in the testing process. In addition, photographs can be annotated with text. There are three primary functionalities of ICDAR 2015: text recognition and word localization. Each image has a set of text-bound boxes (BBoxes) for localization. Each image’s BBoxes is kept in its own file, separated just by a single line. To aid in automatic word recognition, BBoxes are provided in addition to the word itself. • SVHN Dataset: Low-resolution photos and little data processing and formatting make the Street View House Numbers (SVHN) dataset a good benchmark. It could be compared to the MNIST database. Because it was compiled from house numbers in Google Street View pictures, this collection comprises a wide range of photographs, including blurry, low-resolution images. It is available in two formats: a chopped digits image format similar to MNIST and a complete home door image file with digit bounding boxes. Too many photos in this dataset: 73,257 for training, and 26,032 for testing. It’s a mess. Experiments on Datasets: ICDAR 2015 was the first dataset to be experimented with. For this dataset, the most difficult part is the variety of photos, which include a variety of background noises and clutter, as well as fuzzy photographs and lowresolution images. Pictures of the outcomes can be seen in Fig. 3. With a Recall, Precision, and H-mean score of 64.2, 79.53, and 72.86%, the STNOCR approach exceeds all other methods. The comparison is shown side by side in Fig. 4.

Recognition of the Multioriented Text Based on Deep …

373

Fig. 3 Findings from the incidental scene text category of the 2015 international conference on documentary arts research Fig. 4 STN-OCR performance

Table 1 STN-OCR performance

Method

Recall (%)

Precision (%)

H-mean (%)

STN-OCR

64.2

79.53

72.86

Baidu VIS

62.12

70.28

65.89

HoText_v1

62.26

67.25

66.78

FOTS

52.19

74.59

64.42

374 Table 2 Results on SVHN dataset

K. Priyadarsin et al. Method

Accuracy (%)

MaxoutCNN

95

ST-CNN

95.3

STN-OCR

97.8

According to the data in Table 1, the proposed strategy has produced superior outcomes when compared to the alternatives. On the SVHN dataset, the network design is evaluated to demonstrate that this model can be used for real data. House numbers in SVHN also contain noise. Finding, locating, and recognizing SVHN house numbers on a sampling grid was found to be a successful method of testing this study’s network architecture. While initializing random weights were used to train the research model, weights from an already-trained network were used to initialize the localization network for optimum results. Better outcomes are generally obtained when using the localization network stage. A comparison of text recognition performance using real house numbers and the SVHN dataset is shown in Table 2. This system’s accuracy on SVHN improved after ICDAR 2015, when it reached 97.8%. Even if previous research has dealt with some of these conclusions, this study model works well with those photos. An image with a colour backdrop is processed in 2–3 s using Google’s K80 GPU and 12 GB of RAM for testing purposes.

5 Conclusion In this study, a single DNN was utilized to perform text detection and recognition (STN-OCR) utilizing recent benchmark datasets, such as ICDAR 2015. Two of the most important parts of this system are its text detection and recognition components. Text detection models are supplied into a text recognition network, which uses that network’s output to recognize text areas in images. As a result, we were able to better detect text from several perspectives. According to the findings, our model outperforms current best practices by a wide margin on SVHN and ICDAR 2015 tests. Only whole sentences and lines are possible with this model. In the future, this model will be applied to other regional or well-known languages (such as Urdu/Hindi) and the geometric design will be adjusted to detect directly curved texts.

References 1. A. Alshanqiti, A. Bajnaid, A. Rehman, S. Aljasir, A. Alsughayyir, S. Albouq, Intelligent parallel mixed method approach for characterising viral Youtube videos in Saudi Arabia. Int. J. Adv. Comput. Sci. Appl. (2020) 2. Y. Xu, Y. Wang, W. Zhou, Y. Wang, Z. Yang, X. Bai, Textfield: Learning a deep direction field for irregular scene text detection. IEEE Trans. Image Process. (2019)

Recognition of the Multioriented Text Based on Deep …

375

3. S. Khan, D.-H. Lee, M.A. Khan, A.R. Gilal, G. Mujtaba, Efficient edge-based image interpolation method using neighboring slope information. IEEE Access 7, 133539–133548 (2019) 4. S.L. Xue, F. Zhan, Accurate scene text detection through border semantics awareness and bootstrapping, in Proceedings of the European Conference on Computer Vision (ECCV) (2018), pp. 355–372 5. A.R. Gilal, J. Jaafar, L.F. Capretz, M. Omar, S. Basri, I.A. Aziz, Finding an effective classification technique to develop a software team composition model. J. Softw. Evol. Process 30(1), 1–12 (2018) 6. A. Sain, A.K. Bhunia, P.P. Roy, U. Pal, Multi-oriented text detection and verification in video frames and scene images. Neurocomputing 275, 1531–1549 (2018) 7. C. Bartz, H. Yang, C. Meinel, See: Towards semi-supervised endto-end scene text recognition, in 32nd AAAI Conference on Artificial Intelligence, AAAI 2018 (2018) 8. M. Liao, B. Shi, X. Bai, X. Wang, W. Liu, Textboxes: A fast text detector with a single deep neural network, in Thirty-First AAAI Conference on Artificial Intelligence (2017) 9. Y. Wei, Z. Zhang, W. Shen, D. Zeng, M. Fang, S. Zhou, Text detection in scene images based on exhaustive segmentation. Sig. Process. Image Commun. 50, 1–8 (2017) 10. Y. Zhu, C. Yao, X. Bai, Scene text detection and recognition: Recent advances and future trends. Front. Comp. Sci. 10(1), 19–36 (2016) 11. B. Xiong, K. Grauman. Text detection in stores using a repetition prior, in Proceeding of the WACV (2016) 12. S. Qin, R. Manduchi, A fast and robust text spotter, in Proceeding of the WACV (2016) 13. Z. Zhang, W. Shen, C. Yao, X. Bai, Symmetry-based text line detection in natural scenes, in Proceeding of CVPR (2015) 14. I. Posner, P. Corke, P. Newman, Using text-spotting to query the world, in 2010 IEEE/RSJ International Conference on Intelligent Robots and Systems (2010), pp. 3181–3186

Fruit and Leaf Disease Detection Based on Image Processing and Machine Learning Algorithms S. Naresh Kumar, Sankararao Majji, Tulasi Radhika Patnala, C. B. Jagadeesh, K. Ezhilarasan, and S. John Pimo

Abstract Climate change and population growth have recently brought huge environmental and agricultural challenges. Many technologies have been used to monitor and improve agricultural productivity. Modern agriculture has greatly reduced the use of technology due to its low resolution, destructiveness, cost, sensitivity, and reactiveness. In this work, we have concentrated on the fruit and leaf disease detection based on image processing. FCM, K-mean, and ABC algorithms are used for the segmentation of images. The adapted ML algorithms are used healthy and unhealthy dataset to calculate the accuracy and error rate. The ABC algorithms give the 83% in fruit dataset and 95% in leaf dataset. Likewise very less error rate in 16% in fruit and 4% in leaf dataset has been identified. Keywords Machine learning algorithms · Image processing · Accuracy · Error rate · Fruit and leaf disease detection

S. Naresh Kumar School of Computer Science and Artificial Intelligence, SR University, Waragal, Telangana, India S. Majji (B) Department of Electronics and Communication Engineering, GRIET, Hyderabad, India e-mail: [email protected] T. R. Patnala Department of Electronics and Communication Engineering, GITAM University, Hyderabad, India C. B. Jagadeesh New Horizon College of Engineering, Bangalore, India K. Ezhilarasan Department of ECE, CMR University, Bangalore, Karnataka, India S. J. Pimo St. Xavier’s Catholic College of Engineering, Chunkankadai, Nagercoil, India © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 I. J. Jacob et al. (eds.), Expert Clouds and Applications, Lecture Notes in Networks and Systems 444, https://doi.org/10.1007/978-981-19-2500-9_27

377

378

S. Naresh Kumar et al.

1 Introduction Image processing is the process of enhancing or extracting valuable information from an image by performing various operations on it. When an image is sent into a signal processing algorithm, it can return a picture or other features related with that image as the output. The field of image processing is one that is exploding right now [1, 2]. In order to increase the quality of an image and to adapt it for usage in various applications, image processing is used to enhance, segment, feature extract, classify, etc. In order to improve an image, you can alter the brightness, change the colour tone and remove noise. If you want to break up an image into smaller pieces, you can use image segmentation [3]. Objects in digital photographs are often identified using this technique. Image segmentation can be done in a variety of methods, such as using threshold, colour, transform, or texture as a basis [4]. A form of dimensionality reduction known as “feature extraction” reduces the number of pixels in an image by extracting just the most important and visually appealing elements. Image matching and retrieval can be expedited by using a reduced feature representation and a high image size with this strategy [5]. When photos are classified, they are assigned to one of a number of predefined groups. Supervised and unsupervised are separated in the classification. Improving pictorial information for human interpretation, processing image data for storage, transmission, and representation in autonomous machine perception are the main objectives in digital image processing [6]. Leaf photos are grouped under four categories: fruit crops, vegetables, cereal crops, and commercial crops. Crop diseases are used to better categorise these photos. In this dataset, only illnesses that are prevalent in most crops are included. Fungus, viral, and bacterial diseased pictures are also present [7]. A total of 2912 photos are included in the dataset, which was gathered from the Web and universities. There are 728 photos in each category. A total of 364 photos were used in the training process, and a total of 364 images were utilised in the testing phase for each category. Figure 1 depicts a few representative images from the collection. To examine the efficiency of the presented technique, it is tested against fruit image gallery data set. We have collected data set, which is 609 fruit images [7] (Fig. 2).

2 Background Techniques for extracting features provide a useful set of reduced feature vectors, which can summarise virtually all of the information included in the original set of features. These reduced feature vectors are helpful for a variety of applications [8]. A machine learning (ML) classification model is used to assign class labels to the feature vectors in order to categorise them. The primary focus of this research is on image-based machine learning and pattern identification.

Fruit and Leaf Disease Detection Based on Image Processing …

379

Fig. 1 Data set of different types of leaf disease

The advancement of pattern recognition systems has resulted in the creation of a large number of classification models [9]. The features depicted in Fig. 3 were obtained through the pre-processing of the training dataset. During the pre-processing stage, noise reduction and picture verification are carried out. To construct a model that can represent multiple and full feature representations, features are extracted from their original forms. The unfortunate reality is that optimising this procedure is quite tough, and it is necessary in order to build functionality for all apps [10].

3 Methodology This technique is used to first preprocess the leaf image. Affected and unaffected parts of the image are first separated from the rest of the picture using thresholding. Once the non-disease zones have been cleared, only illness zones remain [11]. In the feature extraction stage, each disease region is represented by a different set of features (Fig. 4). A classifier model that has already been built is used to categorise the different areas of disease in the body [12]. Finally, the results of the detection tests are evaluated and analysed using the performance data collected during the process. In the following part, we will go over some of our suggestions for putting them into action. There are sections on pre-processing, segmentation, feature extraction, and classification included in the proposed study.

380

Fig. 2 Data set of different types of fruit disease

Fig. 3 Block diagram for the test phase of the detection

S. Naresh Kumar et al.

Fruit and Leaf Disease Detection Based on Image Processing …

381

Fig. 4 Flow chart of the classification process

For their general growth, transpiration, and nutritional function, plants’ leaves are completely reliant on water due to their highly sensitive capacities [13]. As a result, the timely and exact measurement of resource inputs such as water may be extremely beneficial to a sophisticated agricultural system in many ways. During the course of four days in the laboratory, the antioxidant properties of coffee, pea shoots, and spinach leaves were assessed. Following pre-processing for feature extraction, the data is fed into our suggested machine learning algorithms, which then perform automatic classification [14]. The K-mean, FCM, and ABC algorithms were used to evaluate a variety of machine learning methods.

382

S. Naresh Kumar et al.

4 Results and Discussion The classification system can be analysed with the help of K-mean, FCM, ABC algorithms. In this work we calculated on Accuracy and Error Rate Analysis. For the classifier, this experiment uses the ELM approach as the classifier because it gives the best result. Among five performance metrics, this experiment takes only detection accuracy and error rate as the performance metric. Table 1 shows the detection accuracy analysis of the K-means, FCM, and ABC methods. Table 1 gives the explanation about the accuracy analysis of the fruit and leaf disease detection with the help of the different segmentation methods. Graph can be plotted based on Table 1. ABC method gives more accuracy than the K-mean and FCM methods as shown in Fig. 5. Table 2 represents the error rate analysis with mentioned segmentation methods. ABC method gives the low error rate than K-mean and FCM methods (Fig. 6). Table 1 Segmentation method accuracy analysis

Type of test data set

K-Mean

FCM

ABC

Fruit data set

0.753

0. 762

0.832

Leaf data set

0.902

0.924

0.956

Accuracy Analysis

1.2 1 0.8 0.6 0.4 0.2 0

Fruit Dataset K-Mean

Leaf Dataset FCM

ABC

Fig. 5 Graph for accuracy analysis of the different segmentation methods

Table 2 Error rate analysis with mentioned segmentation methods

Type of test data set

K-Mean

FCM

ABC

Fruit data set

0.247

0. 238

0.168

Leaf data set

0.098

0.076

0.044

Fruit and Leaf Disease Detection Based on Image Processing …

383

Error Analysis

0.3 0.25 0.2 0.15 0.1 0.05 0 Fruit Dataset

K-Mean

Leaf Dataset FCM

ABC

Fig. 6 Graph for error analysis of the different segmentation methods

5 Conclusion Present days, most of the farmers are facing the issue in the form of leaf and fruit disease. ˙In this paper work on the detection of the leaf and fruit disease detection based on the image processing with the help of mechine learning classification algorithms. Here, K-mean, FCM, and ABC algorithms are used for the classification process. And calculate the accuracy and error rate. Finally, compare all the methods which are used in the classification. By observing the results, ABC method gives less error rate and more accuracy in terms of segmentation process.

References 1. R. Dhaya, Flawless identification of fusarium oxysporum in tomato plant leaves by machine learning algorithm. J. Innovative Image Proc. (JIIP) 2(04), 194–201 (2020) 2. A. Sungheetha, R. Sharma, Design an early detection and classification for diabetic retinopathy by deep feature extraction based convolution neural network. J. Trends Comput. Sci. Smart Technol. (TCSST) 3(02), 81–94 (2021) 3. A. Bashar, Survey on evolving deep learning neural network architectures. J. Artif. Intell. 1(02), 73–82 (2019) 4. Y. Yuan, Z. Xu, G. Lu, SPEDCCNN: Spatial pyramid-oriented encoder-decoder cascade convolution neural network for crop disease leaf segmentation. IEEE Access 9, 14849–14866 (2021). https://doi.org/10.1109/ACCESS.2021.3052769 5. J.S. Manoharan, Study of variants of extreme learning machine (ELM) brands and its performance measure on classification algorithm. J. Soft Comput. Paradigm (JSCP) 3(02), 83–95 (2021) 6. T. Vijayakumar, Comparative study of capsule neural network in various applications. J. Artif. Intell. 1(01), 19–27 (2019) 7. Y. Zhang, C. Song, D. Zhang, Deep learning-based object detection improvement for tomato disease. IEEE Access 8, 56607–56614 (2020). https://doi.org/10.1109/ACCESS.2020.2982456 8. K.S. Patle, R. Saini, A. Kumar, S.G. Surya, V.S. Palaparthy, K.N. Salama, IoT enabled, leaf wetness sensor on the flexible substrates for in-situ plant disease management. IEEE Sens. J. 21(17), 19481–19491, 1 Sept, 2021, https://doi.org/10.1109/JSEN.2021.3089722

384

S. Naresh Kumar et al.

9. J. Sun, Y. Yang, X. He, X. Wu, Northern maize leaf blight detection under complex field environment based on deep learning. IEEE Access 8, 33679–33688 (2020). https://doi.org/10. 1109/ACCESS.2020.2973658 10. Q. Zeng, X. Ma, B. Cheng, E. Zhou, W. Pang, GANs-based data augmentation for citrus disease severity detection using deep learning. IEEE Access 8, 172882–172891 (2020). https://doi.org/ 10.1109/ACCESS.2020.3025196 11. A. Khattak et al., Automatic detection of citrus fruit and leaves diseases using deep neural network model. IEEE Access 9, 112942–112954 (2021). https://doi.org/10.1109/ACCESS. 2021.3096895 12. M. Kumar, A. Kumar, V.S. Palaparthy, Soil sensors-based prediction system for plant diseases using exploratory data analysis and machine learning. IEEE Sens. J. 21(16), 17455–17468, 15 Aug 2021. https://doi.org/10.1109/JSEN.2020.3046295 13. S. Janarthan, S. Thuseethan, S. Rajasegarar, Q. Lyu, Y. Zheng, J. Yearwood, Deep metric learning based citrus disease classification with sparse data. IEEE Access 8, 162588–162600 (2020). https://doi.org/10.1109/ACCESS.2020.3021487 14. G. Yang, G. Chen, Y. He, Z. Yan, Y. Guo, J. Ding, Self-supervised collaborative multi-network for fine-grained visual categorization of tomato diseases. IEEE Access 8, 211912–211923 (2020). https://doi.org/10.1109/ACCESS.2020.3039345

A Survey for Determining Patterns in the Severity of COVID Patients Using Machine Learning Algorithm Prachi Raol, Brijesh Vala, and Nitin Kumar Pandya

Abstract In recent era of medical domain research, a decision support system is need of the time. Various technological analyses based on historical data and clinical results are used for prediction and classification in medical domain. Support vector machine, Naive Bayes and random forest are examples of machine learning algorithms that have been demonstrated to be effective in medical studies such as heart disease, Alzheimer’s disease and brain tumor categorization. As COVID infection is world wide issue, and symptoms and other diagnosis for COVID affected patient is vary by age group, region and variant of the virus. There is a large scope for the system that can help medical persons and government organizations to arrange the resources and better treatment to the patient. Early prediction for the severity of COVID patient helps for better treatment. Like other dieses, machine learning can also play an important role in COVID pattern prediction also. There are also numerous problems in this study project due to a shortage of raw data and the need to uncover patterns. In this paper, we have examined some of the research that is been done by different researchers. A prediction or forecasting model that can well characterize and specify the severity of current COVID-19 disease infection rate in clinical diagnosis and provide support for clinicians to help scientific and rational medical and treatmentrelated decision making may be built. We have tried out to find their pros and cons and tried to find out better solution that can improve the overall result. Keywords Machine learning · Medical · Prediction · COVID · SVM · Naïve Bayes · Pattern mining · Patient care · Deep learning · Classification

P. Raol (B) · B. Vala · N. K. Pandya Parul Institute of Engineering and Technology, Vadodara, India e-mail: [email protected] B. Vala e-mail: [email protected] N. K. Pandya e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 I. J. Jacob et al. (eds.), Expert Clouds and Applications, Lecture Notes in Networks and Systems 444, https://doi.org/10.1007/978-981-19-2500-9_28

385

386

P. Raol et al.

1 Introduction Health care and patient care are important sections which offer value-based care to millions of people. Resource management and in time care are main keys for the health sector [1]. In most cases apart from COVID, there have been enough patient care resources arranged in most of countries. Treatment strategy is also smooth due to various technical support in most of dieses. But COVID has tested the limitation of healthcare services in most of region due to its variant and different type of effect and its spreading methods. After evolution of machine learning, there is not any domain where ML has not left its impact. Health care is not also in exception. Machine learn ing technology and their algorithms have proven their efficiency in health care also. Many machine learning-based models are useful for diagnosis systems. Supportive systems are used in decision making in various health conditions like brain tumor, heart diseases, etc. For COVID, some research has been performed by various researchers. In this survey paper, we have reviewed some of the research work that has been carried out for use of machine learning algorithms for COVID. “Machine learning in health informatics can streamline recordkeeping, including electronic health records (EHRs). Using AI to improve EHR management can improve patient care, reduce healthcare and administrative costs, and optimize operations” [2]. Some of the algorithms used for this type of work are discussed below: Classification using Machine Learning The classification algorithm is a technique that is used to identify the category of new observations or samples based on provided training data. In classification, a model learns from the input dataset samples (training data) or observations and then classifies new sample into a number of classes or categories. The classes may be Yes or No, 0 or 1, spam or not spam, fake or not fake, etc. Classes can be referred as targets/labels or categories in technical terms (Fig. 1). Prediction using Machine Learning Prediction in machine learning refers to the outcome of any algorithm after training on a historical dataset and tested to new testing data with aim of forecasting. Example of prediction in machine learning includes weather prediction, customer purchase Fig. 1 Classification example [3]

A Survey for Determining Patterns in the Severity of COVID …

387

prediction, query prediction, etc. Mostly it works over probability criteria. Prediction has no general world meaning in machine learning. It is purely based on mathematical model. “Linear regression is perhaps one of the most well-known and well-understood algorithms in statistics and machine learning” [4]. Linear regression is represented by an equation that defines a line that best matches the relationship between the input variables (x) and the output variables (y) by determining precise weightings for the input variables, known as coefficients (B). Other algorithms like SVM, decision tree and XG Boost are also being used. Average results for these algorithms are shown at last section in this paper. The main challenges for the COVID Predictions are the lack of a dataset and insufficient information regarding COVID treatment and different symptoms. Due to lack of data feature selection methods, applying may not provide proper information. We have tried to find out study of current methodologies.

2 Related Works In a study [1] researchers used a machine learning algorithm to predict if COVID-19 patients require mechanical ventilation within 24 h of their initial hospital admission. They have tested their model over 197 patients [1]. The following graph shows the early warning score of their system (Figs. 2 and 3). In their study, the researchers used three classification algorithms: logistic regression (LR), random forest (RF) and extreme gradient boosting (XGB). In this study, 287 COVID-19 data of patients at Saudi Arabia’s King Fahad University Hospital was used. They tried to predict deceased from the dataset. Tables 1 and 2 displays the result founded by their algorithms. As per their result, they have applied SMOTE analysis. They have applied accuracy, sensitivity and specificity and F-score measure for their research parameter. Table 1 shows the detailed result for linear regression, random forest and XG Boost algorithm. In their research, lake of dataset is seen. And anomaly is the main issue for this research as death ratio for the COVID cases

Fig. 2 Early warning score used in model [1]

388

P. Raol et al.

Start

COVID 19 Patients data Data Cleaning Cleaned Data Data Transformation Transformed Data

Applying Machine learning Algorithms

Output Classifying COVID-19 cases into positive or negative

End Fig. 3 Flow for COVID data classification [5]

is nearly 2%. So, it is difficult to find the pattern for death cases. They have tried to get it by various machine learning algorithms and got significant result. As per their opinion, anomaly detection can be an important key point for death prediction. Various anomaly detection algorithms and methods can improve the result. In another research work “Predicting the COVID-19 infection with fourteen clinical features using machine learning classification algorithms” [5], researcher of this paper has also tried to classify COVID dataset. Based on 14 clinical variables, their study produced six prediction models by using six distinct classifiers for COVID19 diagnosis (i.e., Bayes Net (e.g., NB), logistic like regression, IBk algorithm,

A Survey for Determining Patterns in the Severity of COVID …

389

Table 1 Literature review table Title

Method/Parameters

Limitations

“Prediction of respiratory de-compensation in COVID-19 patients using machine learning: the READY trial” [1]

Linear regression Positive/Neg. score in percentage

More data is required for improving the result

“Machine learning-based model to predict the disease severity and outcome in COVID-19 patients” [13]

Researchers have used logistic regression (LR), random forest (RF) and extreme gradient boosting (XGB) evaluation parameters like accuracy and F-score were used

Imbalance data to be managed

“Predicting the COVID-19 infection with fourteen clinical features using machine learning classification algorithms” [5]

They have used BayesNet, Better feature selection method logistic, IBk, CR, PART and J48 can lead for a better result Evaluation parameters like accuracy and F-score were used

“Clinical and inflammatory features-based machine learning model for fatal risk prediction of hospitalized COVID-19 patients: results from a retrospective cohort study” [6]

They have applied XG Boost Mostly data is from Wuhan and algorithm is for single variant of COVID Evaluation parameters like AUC More data can improve result and ROC were used

“Severity detection for the They used SVM algorithm Coronavirus disease 2019 Evaluation parameters like (COVID-19) patients using a accuracy and F-score were used machine learning model based on the blood and urine tests” [7]

The findings may not be applicable to people of other ethnicities because their study focused only Chinese patients with COVID-19

“Supervised machine learning models for prediction of COVID-19 infection using epidemiology dataset” [8]

Decision tree was used to develop the model Evaluation parameters like accuracy and F-score were used

According to their studies, the decision tree model has the maximum accuracy of 94.99%

“Prediction of COVID-19 cases using CNN with X-rays” [9]

They have used algorithms like CNN and GoogLeNet They have used accuracy as a evaluation parameter

Their outcomes have 99% training accuracy and 98.5% testing accuracy

“Coronavirus disease ANN-GWO MAPE CR (COVID-19) global prediction using hybrid artificial intelligence method of ANN trained with grey wolf optimizer” [10]

In their research ANN-GWO gave MAPE of 6.23, 13.15 and 11.4%

“Forecasting COVID-19 via RCNN algorithm was used registration slips of patients using Accuracy has been used as ResNet-101 and performance evaluation parameter analysis and comparison of prediction for COVID-19 using faster R-CNN, mask R-CNN, and Res-Net-50” [11]

The dataset was X-ray images. The best accuracy of 87% was achieved by a faster R-CNN

(continued)

390

P. Raol et al.

Table 1 (continued) Title

Method/Parameters

Limitations

“COVID-19 prediction and detection using deep learning” [12]

They have used LSTM in their model Accuracy was the evaluation parameter in their model

They achieved average accuracy of 94.80% and 88.43%

Table 2 Algorithm accuracy

Source

Algorithm

Accuracy (%)

[5]

Multivariate random forest

80

[5]

Decision tree

82–86

[13]

Logistic regression without SMOTE

87

[13]

Logistic regression with SMOTE

86

[5]

CR

84.21

[7]

SVM

81

[8]

ANN

89

CR, PART and J48 algorithm). Their research looked back at 114 instances from the Taizhou hospital in Zhejiang Province, China. This system’s detailed steps are shown in Fig. 2. We have reviewed research “Clinical and inflammatory features-based machine learning model for fatal risk prediction of hospitalized COVID-19 patients: results from a retrospective cohort study” [6]. Researchers used 48 features in their study, including clinical and laboratory information. The LASSO approach was used to screen all features. The importance of each feature selected using LASSO was then ranked using a machine learning model based on multi-tree extreme gradient boosting. The simple-tree XGBoost model was then used to develop death risk prediction model. Model’s performance was evaluated using AUC, prediction accuracy, precision and F1-scores. In another research [7], they have applied support vector machine (SVM) for their research work. In research paper, “Supervised Machine Learning Models for Prediction of COVID-19 Infection Using Epidemiology Dataset” [8], they employed supervised machine learning algorithm for COVID-19 [8]. They employed a variety of techniques, including logistic regression, decision trees, support vector machines, Naive Bayes and artificial neutral networks. The dataset was tagged for positive and negative COVID-19 instances in Mexico by the researchers. They also used correlation coefficient analysis on a variety of features to determine the strength of each dependent and independent feature’s link. Testing and training data ratio is 20:80. Their findings show that, with a 94.99%, its decision tree model is the most accurate. The highest sensitivity is 93.34% for the support vector machine (SVM) model, and the highest specificity is 94.30% for the Naive Bayes model.

A Survey for Determining Patterns in the Severity of COVID …

391

In another research, _A transfer learning model was used in the study “Prediction of COVID-19 Cases Using CNN with X-rays” [9]. For the COVID-19 prediction, they used GoogLeNet. They used chest X-ray photos for Dataset. GoogLeNet is used to classify images. GoogLeNet is basically a CNN architecture. This model is also called InceptionV1. They classified X-ray images. The positively classified images indicate that the COVID-19 is present in X-ray. Their outcomes have 99% training accuracy and 98.5% testing accuracy. In research work “Coronavirus Disease (COVID-19) Global Prediction Using Hybrid Artificial Intelligence Method of ANN Trained with Grey Wolf Optimizer” [10], their research seeks to use a grey wolf optimizer integrated by artificial neural network for COVID-19 predictions. They have used global dataset. Time-series data was used for training and testing purpose. This data is from January 22 to September 15, 2020. To assess the results, they employed mean absolute percentage error (MAPE) and correlation coefficient values. ANN-GWO gave MAPE of 6.23%, 13.15% and 11.4% in their study. In research work “Forecasting COVID-19 via Registration Slips of Patients using ResNet-101 and Performance Analysis and Comparison of Prediction for COVID-19 using Faster R-CNN, Mask R-CNN, and ResNet-50” [11], the study has main two dimensions. They used patients’ registration slips in the first dimension. They used ResNet-101 out of an indigenous data set of COVID-19 patients’ registration slips. They have used dataset of 5003 patient’s records with exact timing. According to their model, the prediction accuracy in terms of time was 82%. X-Ray data was used in their second dimension. 8009 X-rays of the chest were used. Three neural networks were used. Over the X-ray dataset, faster R-CNN, ResNet-50 and Mask-CNN are used. The best accuracy of 87% was achieved by using a faster R-CNN. In the study “COVID-19 Prediction and Detection Using Deep Learning”, they worked over a deep convolutional neural network-based artificial intelligence technique [12]. The main goal was to use real-world datasets to detect COVID-19 patients. For identifying COVID-19 patients, their system examines chest X-ray images. As per their research and findings, they proved that X-rays can be available faster and at lower costs. For testing purpose, they have tested A total of 1000 X-ray scans of real-life patients. By using that dataset, they confirmed that the ML system may be beneficial in recognizing COVID-19. They got a 95–99% F-measure range. They used three forecasting approaches in the next phase. The three methods used were the prophet algorithm, the autoregressive integrated moving average model and the long short-term memory neural network. In Jordan and Australia, they had average accuracy of 88.43% and 94.80%, respectively. Tables 1 and 2 explains pro and cons founded by our survey. As COVID is still in initial stage of research work, there is too much scope for improving the various result. Event current tested model may not be applicable for next variants as symptoms and features in different variant differ in various region.

392

P. Raol et al.

3 Conclusion We have studied some of the research work done for COVID data classification and prediction from various researchers. We have founded that data collection is main issue for the same problem and imbalanced data leads to inaccurate result. We are going to develop a system that manages the data in proper way and classify and predict the severity for the patient. In future we are going to find the various sources of COVID data from various regions and apply classification algorithm that best suits for this imbalanced data.

References 1. H. Burdick et al., Prediction of respiratory decompensation in COVID-19 patients using machine learning: the READY trial. Comput. Biol. Med. 124, 103949 (2020) 2. Harvard Business Review, “Using AI to Improve Electronic Health Records”. https://hbr.org/ 2018/12/using-ai-to-improve-electronic-health-records 3. University of Illinois Chicago, “Machine Learning in Healthcare: Examples, Tips & amp; Resources for Implementing into YourCarePractice”. https://healthinformatics.uic.edu/blog/ machine-learning-in-healthcare 4. Built in BETA, The top 10 machine learning algorithms every beginner should know. https:// builtin.com/data-science/tour-top-10-algorithms-machine-learning-newbies 5. I. Arpaci et al., Predicting the COVID-19 infection with fourteen clinical features using machine learning classification algorithms. Multimedia Tools Appl. 80(8), 11943–11957 (2021) 6. X. Guan et al., Clinical and inflammatory features based machine learning model for fatal risk prediction of hospitalized COVID-19 patients: results from a retrospective cohort study. Ann. Med. 53(1), 257–266 (2021) 7. H. Yao et al., Severity detection for the coronavirus disease 2019 (COVID-19) patients using a machine learning model based on the blood and urine tests. Front. Cell Dev. Biol. 8, 683 (2020) 8. L.J. Muhammad et al., Supervised machine learning models for prediction of COVID-19 infection using epidemiology dataset. SN Comput. Sci. 2(1), 1–13 (2021) 9. D. Haritha, N. Swaroop, M. Mounika, Prediction of COVID-19 cases using CNN with X-rays, in 2020 5th International Conference on Computing, Communication and Security (ICCCS) (IEEE, 2020) 10. S. Ardabili et al., Coronavirus disease (COVID-19) global prediction using hybrid artificial intelligence method of ANN trained with Grey Wolf optimizer, in 2020 IEEE 3rd International Conference and Workshop in Óbuda on Electrical and Power Engineering (CANDO-EPE) (IEEE, 2020) 11. H. Tahir, A. Iftikhar, M. Mumraiz, Forecasting COVID-19 via registration slips of patients using ResNet-101 and performance analysis and comparison of prediction for COVID-19 using Faster R-CNN, Mask R-CNN, and ResNet-50, in 2021 International Conference on Advances in Electrical, Computing, Communication and Sustainable Technologies (ICAECT) (IEEE, 2021) 12. M. Alazab et al., COVID-19 prediction and detection using deep learning. Int. J. Comput. Inf. Syst. Ind. Manage. Appl. 12, 168–181 (2020) 13. S.S. Aljameel et al., Machine learning-based model to predict the disease severity and outcome in COVID-19 Patients. Sci. Program. 2021 (2021)

Relation Extraction Between Entities on Textual News Data Saarthak Mehta, C. Sindhu , and C. Ajay

Abstract Relation extraction is one of the key components in information extraction. The primary step in relation extraction is entity recognition. Identifying the entities in a sentence is a highly complex task and requires multilayered models which use computationally heavy algorithms. The performance of these models was highly dependent on the natural language processing tools. Neural networks are used which use different algorithms and activation functions which are not dependent on these tools. Begin, interior, outside tags are also used as the tagging scheme for the dataset. Each of the word in the line or paragraph is tokenized, and then, each word goes through a named entity recognition process. The tokenized words are tagged in the begin, interior and outside tag format. An attempt to tackle the issue of relation extraction is done using conditional random fields to identify our entities and gated recurrent unit for the purpose of contextual data recognition and finally extract the relations between these entities. Begin, interior, and outside tagging scheme are used to figure out the type of entity, which will help us in classifying the entities better. Keywords Gated recurrent units · Conditional random fields · Begin input and outside tagging scheme · Named entity recognition · Relation extraction

1 Introduction Natural language processing came up as a means for humans to automate the process of our day-to-day work with the help of a computer. It was specifically for the ability of computers to understand human language semantically and lexically. Thus continued, the quest for automating more Natural Language Processing problems. It is in demand for the purpose of bringing the level of understanding between computers and humans to a higher ground. Natural language processing has the ability to solve many issues and come up with a new innovation. These tasks are speech recognition, part of speech tagging, word sense disambiguation [1], sentiment S. Mehta · C. Sindhu (B) · C. Ajay Department of Computing Technologies, SRM Institute of Science and Technology Kattankulathur, Kattankulathur 603203, India e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 I. J. Jacob et al. (eds.), Expert Clouds and Applications, Lecture Notes in Networks and Systems 444, https://doi.org/10.1007/978-981-19-2500-9_29

393

394

S. Mehta et al.

analysis [2, 17, 20] and natural language generation. To solve these issues, natural language processing needed some tools to be able to solve specific tasks. For this purpose, the natural language ToolKit (NLTK) library was introduced. The NLTK library has the ability to tokenize the sentence, stemming and lemmatization of each and every word and also provide more functionality. Today, natural language processing is going beyond the scope of rule-based systems. It is turning toward machine learning and deep learning to solve the issues posed by the researchers. There are a number of use cases for natural language processing which include spam detection, machine translation, virtual agents, chatbots, text summarization, etc. An interesting task, relation extraction [16] is the most important fields in natural language processing. This was one such problem in natural language processing which was not worked on by researchers until recently. In this problem, there exist entities which have relations between each other, and the task is to extract these relations between these entities. The entities may be an organization, person, location or any other designation possible. The relation between two entities could be to signify whether one person lives in, was born in, etc. The goal is to correctly identify the entities as well as correctly give the relation which exists between them using deep learning and get a high accuracy. Before this task the process of named entity recognition needs to take place. In named entity recognition, the tokenized words are individual fed into the algorithm after which a tag is provided to the word which might indicate what a word signifies if the word is from the premade groups of in the library. The use cases of relation extraction are in the fields of medical, chemical research as well as general research. This could work to identify symptoms and causes of diseases, or to even something completely different, such as, finding out which individual was born in which city. The scope of relation extraction reaches to not only science but also to the fields of art and journalism. Previous works on the relation extraction used NLP tools such as named entity recognition, dependency parser and POS tagging. These tools would complete the task but proved to be inefficient because of the dependency on these tools and external features made specifically for the task. Liu et al. [3] used a dependency-based neural network which would extract relations and improve the working with other external features and thus requires these features. Kanjirangat and Rinaldi [4] also used the shortest dependency parser which is an external feature on which the models are based. Another issue was that multiple entities could have multiple relations, but generally a relation extraction system could only get back two entities and one relation in one sentence. To tackle this problem, the process of extracting multiple relations between entities was realized by [5, 6]. However, we noticed that neither of the researchers tried to apply GRU which is a computationally effective RNN. GRU has the update and reset gate which has similar working as a LSTM which has three gates which provide the same functionality. GRU thus proves to be a better option. In this paper, the task was to make a model which works on the basis of gated recurrent units and conditional random fields to handle the task of relation extraction and entity classification. The issue of multi-head instance where multiple relations can exist between multiple entities is also handled. Our model will help in being more computationally efficient than [5]. The model is applied on the CONLL04 dataset,

Relation Extraction Between Entities on Textual News Data

395

and the goal is to extract relations and entities which will show us the results without using external features of parser, part of speech tagging, etc.

2 Literature Survey The idea of adding semantic information to turn unstructured text into structured text is a popular one. In order to make an accurate annotation, the computer must be able to distinguish a piece of text with a semantic attribute of interest. As a result, extracting semantic relationships between elements in natural language text is a critical step in developing natural language understanding systems. The main focus is to target strategies for identifying relationships between elements in unstructured text in this paper. In this section, the different methods currently being used or which were in use in the past are listed and discussed about (Table 1). Recurrent neural networks, gated recurrent units, conditional random fields and named entity recognition were majorly used. Other methods used include: Sentence splitting for large textual document sentence splitting plays a vital role as processing a large paragraph would require high computational capabilities, lexical processing is the method in which each word of the sentence is broken down and the algorithm tries to recognize each of the words. Following are the list of related works: Table 1 Papers with unexplored fields Reference No. Task

Unexplored Fields

[7]

• Extract lexical and sentence-level information using deep learning neural networks

• Self-attention mechanism is something to add to this

[6]

• Independent of external features • Novel tagging scheme • Exploits self-attention to capture long-range dependencies

• Implementing GRUs is something which was not explored

[8]

• Implemented shortest dependency path from the dependency parse tree

• The word embedding of BioWord2Vec were not touched upon to check for performance

[5]

• A LSTM [18] and CRF method to extract entity and relation

• Applying other neural methods were unexplored

[9]

• CNN with character embedding • A Bi-LSTM [15] embedding also with tanh activation takes place

• Implementing GRUs is something which was not explored

396

S. Mehta et al.

2.1 Recurrent Neural Networks In the recurrent neural network, the input of the next step is based on the output from the last step. Usually, the inputs are independent from each other. Although in some cases, the previous words are needed, and the preceding words must be recalled/remembered. RNN has a hidden layer to overcome this issue and has a hidden layer to solve the problem. The most important component of RNN is the hidden state, which remembers specific information about a sequence. Although a step forward in many directions, the RNN has multiple shortcomings: (i) Problems exploding gradient and vanishing gradient, (ii) Difficult to train RNN, (iii) When utilizing tanh or RELU as an activation function, it will not be able to handle very long sequences.

2.2 Gated Recurrent Units Gated recurrent units (GRU) use an update gate and a reset gate, current memory content and final memory at the current step to address the vanishing gradient problem of a normal recurrent neural network (RNN). Essentially, they are two vectors that determine what information should be sent to the output. They are unique in that they can be trained to retain knowledge from a long time ago without being washed away by time or to discard information that is unnecessary to the prediction [18]. Update gate: Holds the information for the previous t−1 units and is multiplied by its own weight U(z). Both results are added together, and a sigmoid activation function is applied to squash the result between 0 and 1. It essentially eliminates the risk [19] of vanishing gradient problem. Reset gate: This gate is used from the model to decide how much of the past information to forget. The reset gate uses the same formula as the update gate. Current memory content: This step is required for collecting knowledge from the past that is relevant and is used with the input of the current cycle. Final memory at current step: The network then calculates the vector that holds information for the current unit and feeds it down to the network as the final stage of the cycle. In order to do that, the update gate is needed.

2.3 Conditional Random Fields A conditional random fields (CRF) is a sequence modeling algorithm which is used to identify entities or patterns in text, such as parts of speech (POS) tags. It is considered to be the best method for entity recognition. Conditional random fields are a type

Relation Extraction Between Entities on Textual News Data

397

of discriminative model that is well suited to prediction tasks in which the current prediction is influenced by contextual information or the status of the neighbors. CRF overcomes the label bias issue found in maximum entropy Markov mode (MEMM) by using global normalizer.

2.4 Named Entity Recognition There are special terms that indicate specific entities that are more informative and have a unique context in any written composition. These are referred to as named entities (Table 2), which are terms that describe real-world items that are frequently signified by proper names. A basic method may be to look for these in text documents’ noun phrases. To recognize and segment named items, as well as classify and sort them into several predetermined categories. SpaCy is one of the most widely used libraries.

3 Methodology In this paper, a model is provided which identifies the entities, that is, the type and all the relations which can be inferred from them and also tries to solve the problem related to multi-head issue. The multi-head issue is the presence of multiple relations between multiple entities in a single sentence. The model has different layers for the architecture. The layers are of word embedding, the Bi-GRU layer, the CRF layer and finally the sigmoid layer. For our model, the dataset which is being used is CONLL04 dataset with the BIO tags present. The sentence is tokenized, and the tokenized words are then going through a word embedding layer. The word embedding layers provide mathematical reference on which the statistical model is built and the machine can interpret the sentence. The Bi-GRU layer is utilized and Table 2 Types of named entities [10]

Type

Description

Human

People, including fictional

Nation

Country

Object

Vehicles, items, etc.

Event

Sports or business events

Law

Law-related document

Language

Particular language

Date

Format in date to identify dates

Time

Hours, minutes

Percent

Percentage

398

S. Mehta et al.

Fig. 1 Architecture diagram for relation extraction

Fig. 2 Layer-wise representation of the model

provides a complex contextual reference to the words. The CRF and sigmoid layers are eventually used to output the prediction made for the entities and the relation between them. If there is no relation between the entities, “N” tagging is used to signify as such (Figs. 1 and 2).

3.1 Dataset CONLL04 dataset: This dataset provides entity types, namely person, organization, location and other. The relation types are kill, located in, work for, etc. It contains 910 training examples,

Relation Extraction Between Entities on Textual News Data

399

243 for validation and 288 for testing. Zhao et al. [11] uses this for experimental purposes for the same. Implementation and hyperparameters: Python with the TensorFlow machine learning library is used. ADAM optimizer with a learning rate 10–2 is the main optimizer being applied in the network. The size of GRU was around 64. The dataset is regularized by applying dropout. It is applied to the hidden layers as well input embeddings. The character embedding size is set to 25.

3.2 Word Embedding A sentence may contain n number of words, which may have different meanings. These words need to have a mathematical reference for the machine to be able produce a statistical model. This word embedding layer helps in giving that mathematical reference to each and every word provided in the sentence. The pretrained embedding of GloVe is utilized, which is used for a contextual-level model, used universally. The 200-dimensional word embedding is applied. Character embedding is implemented. This takes each and every character of a word and provides it with a certain value. The whole purpose of this is to be able to obtain the power of prefixes and suffixes used. For example, a word like “famous” can have a prefix of “un” and can change the meaning only on a contextual level. The use of prefixes and suffixes thus becomes extremely useful. The embeddings are put through a BiGRU, and the two states from the forward and backward feeding are attached together. The character embedding vector is appended to the word embedding to obtain the final word embedding vector.

3.3 Bi-Directional Gated Neural Networks RNNs were introduced as a tool to tackle many NLP-related tasks which had lots of sequential data available. For our purpose, the same is required. BiGRU layer is applied. Deepa et al. [12] uses the same for relation extraction purpose. This layer has the ability to encode information from two directions and thus give more information at the time of output.

3.4 Conditional Random Fields Apart from using the BiGRU layer, the task at hand requires us to identify entities using the BIO tagging scheme. The BIO tag signifies B as the beginning of the entity,

400

S. Mehta et al.

I as the inside tag which suggests that it is a continuation of the entity tag from before and O as the outside tag which suggest that it is not an entity. Here, the CRF layer helps in calculating the most highly probable tag for the token. Referring from [5], the value for each token is calculated with this formula below.   S(l) (h i ) = V(l) f U(l) h i + b(l)

(1)

l is used in the named entity recognition task. f (.) is an activation function where V (l) ∈ Rp×g , U (l) ∈ Rg×2d , b(l) ∈ Rg . The CRF layer is utilized to identify the entity recognition work which was required to be done. Chen and Hu [13] applies for the same purpose. For this named entity recognition, the linear-chain CRF will find out the probability with this formula, where we assume the word w, has a sequence of score s, and prediction y and we refer [5] for the formula. n n−1     S y1(l) , . . . , yn(l) = S (l) i, y (l) i + Ty (l)i y(l)i+1 i=0

(2)

i=1

where S ∈ Rsi , yi (e)(e) is the score for the tag which is predicted for token wi , T is a transition matrix in which each value entered denotes the transition scores from one tag to another. Referring [5], the probability of the entity having one particular tag over all possible tag sequence for the input provided is          P y1(l) , . . . , yn(l) |W = e s y1(l) , . . . , yn(l) s y1(l) , . . . , yn(l) e s y1(l) , . . . , yn(l) (3) Cross-entropy loss is minimized. The entity tags for classification [21] purposes by which learns the label embedding are implemented.

3.5 Multi-Head Selection In this paper, the term multi-head is mentioned multiple times. Each token available in the sentence can have multiple relations between other entities. The vector, that is, the other entities and the relation between the entities (y, c) are predicted. Our aim is to find the most probable entities which have relations with the entity w and the most probable relation they have, with the following equation.     S(r ) z y , z x , rk = V(r ) f U(r ) zy + W(r ) zi + b(r )

(4)

The goal is to find the probability of getting W y and Rk when the word W x is observed. This is used in [5] for the same. Cross-entropy loss is minimized during training.

Relation Extraction Between Entities on Textual News Data Table 3 Accuracy score

S. No.

Reference No.

401 Model

Accuracy

1

[4]

Transformers

0.52

2

[14]

BiLSTM

0.84

3

[3]

DepNN

0.83

4

[7]

CNN

0.82

5

[8]

LSTM

0.92

6

[5]

LSTM with CRF

0.80

7

[6]

BiLSTM

0.82

4 Results and Discussion The proposed model uses CRF and sigmoid layer for the purpose of finding entities and relations. There are different evaluation types which are being applied here which are strict, boundaries and relaxed. Strict is used to identify an entity and relation between the entities exactly as it is labeled. Boundaries suggest that if it is within the scope of being classified as a certain type of entity, it will be accepted as a positive result. Whereas in relaxed, if one of the token types is correct for a multi-token entity, then we can give it a positive result. Evaluation of the data will help us in the process of finding out how accurately does the model work in reaching its target. The model is evaluated on the basis of precision and recall and the F1-score. These are the various performance metrics on which models are evaluated on. The formula for these metrics are as follows: Precision = True Positive/(True Positive + False Positive)

(5)

Recall = True Positive/(True Positive + False Negative)

(6)

F1 Score = 2 · (Precision × Recall)/(Precision + Recall)

(7)

Upon applying the BiGRU on the CONLL4 dataset, the accuracy score of 60% is achieved (Table 3).

5 Conclusion In this paper, a model is presented to jointly extract relations and entities using the textual news data. CRF is applied to identify the entities, and sigmoid layer is used to get relations between them. This model overcomes the previously used models which applied NLP tools to get the result for relation extraction.

402

S. Mehta et al.

References 1. E. Barba, L. Procopio, R. Navigli, CONSEC: word sense disambiguation as continuous sense comprehension, in Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing (2021). https://doi.org/10.18653/v1/2021.emnlpmain.112 2. S. Shakya, A self monitoring and analyzing system for solar power station using IoT and data mining algorithms. J. Soft Comput. Paradigm 3(2), 96–109 (2021) 3. Y. Liu, F. Wei, S. Li, H. Ji, M. Zhou, H. Wang, A dependency-based neural network for relation classification. ACL Anthology (2015). https://aclanthology.org/P152047.pdf 4. V. Kanjirangat, F. Rinaldi, Enhancing biomedical relation extraction with transformer models using shortest dependency path features and triplet information. J. Biomed. Inform. 122, 103893 (2021). https://doi.org/10.1016/j.jbi.2021.103893 5. G. Bekoulis, J. Deleu, T. Demeester, C. Develder, Joint entity recognition and relation extraction as a multi-head selection problem. Expert Syst. Appl. 114, 34–45 (2018). https://doi.org/10. 1016/j.eswa.2018.07.032 6. J. Chen, J. Gu, Jointly extract entities and their relations from biomedical text. IEEE Access 7, 162818–162827 (2019). https://doi.org/10.1109/access.2019.2952154 7. D. Zeng, K. Liu, S. Lai, G. Zhou, J. Zhao, Relation classification via convolutional deep neural network. ACL Anthology (2014). https://aclanthology.org/C141220.pdf 8. D. Ningthoujam, S. Yadav, P. Bhattacharyya, A. Ekbal, Relation extraction between the clinical entities based on ... EZDI. https://www.ezdi.com/wpcontent/uploads/2018/10/Relation_extrac tion_between_the_clinical_entities_based_on_the_shortest_dependency_path_based_LSTM. pdf. 9. F. Li, M. Zhang, G. Fu, D. Ji, A neural joint model for entity and relation extraction from biomedical text. BMC Bioinform. https://doi.org/10.1186/s12859-017-1609-9 10. D. Sarkar, Named entity recognition: a practitioner’s guide to NLP (2018). KDnuggets. Retrieved from https://www.kdnuggets.com/2018/08/named-entity-recognition-practitionersguide-nlp-4.html 11. S. Zhao, M. Hu, Z. Cai, F. Liu, Modeling dense cross-modal interactions for joint entity-relation extraction, in Proceedings of the Twenty-Ninth International Joint Conference on Artificial Intelligence (2020). https://doi.org/10.24963/ijcai.2020/558 12. C.A. Deepa, P.C. Raj, A. Ramanujan, Improving relation extraction beyond sentence boundaries using attention, in International Conference on Computational Sciences-Modelling, Computing and Soft Computing (CSMCS 2020) (2020). https://doi.org/10.1063/5.0046136 13. T. Chen, Y. Hu, Entity relation extraction from electronic medical records based on improved annotation rules and BILSTM-CRF. Ann. Transl. Med. 9(18), 1415–1415 (2021). https://doi. org/10.21037/atm-21-3828 14. S. Zhang, D. Zheng, X. Hu, M. Yang, Bidirectional long short-term memory networks for relation ... ACL anthology (2015). https://aclanthology.org/Y15-1009.pdf 15. Z. Huang, W. Xu, K. Yu, Bidirectional LSTM-CRF models for sequence tagging. arXiv. http:// export.arxiv.org/pdf/1508.01991 (2015) 16. S. Pawar, G.K. Palshikar, P. Bhattacharyya, Relation extraction: a survey. arXiv.org. https:// arxiv.org/abs/1712.05191 (2017) 17. C. Sindhu, D.V. Vyas, K. Pradyoth, Sentiment analysis based product rating using textual reviews, in 2017 International conference of Electronics, Communication and Aerospace Technology (ICECA), pp. 727–731 (2017). https://doi.org/10.1109/ICECA.2017.8212762 18. H.K. Andi, An accurate bitcoin price prediction using logistic regression with LSTM machine learning model. J. Soft Comput. Paradigm 3(3), 205–217 (2021) 19. S. Shakya, S. Smys, Big data analytics for improved risk management and customer segregation in banking applications. J. ISMAC 3(03), 235–249 (2021) 20. A. Sungheetha, R. Sharma, Transcapsule model for sentiment classification. J. Artif. Intell. 2(03), 163–169 (2020) 21. T. Vijayakumar, R. Vinothkanna, Capsule network on font style classification. J. Artif. Intell. 2(02), 64–76 (2020)

A Comprehensive Survey on Compact MIMO Antenna Systems for Ultra Wideband Applications V. Baranidharan, S. Subash, R. Harshni, V. Akalya, T. Susmitha, S. Shubhashree, and V. Titiksha

Abstract 5G and beyond 5G technologies play a most crucial role in the current high speed communications systems. The Ultra Wideband (UWB) technologies are having the most important perspective in modern wireless systems. These communication systems will give desirable features over the support high data rate, reduce the cost, support more of the users, etc. In this paper, we give a comprehensive survey on different types of proposed UWB MIMO antennas and its designs proposed by the various researchers in recent literature. This paper elaborates the state of art of research in various 5G Ultra Wideband Multi Input Multi Output (MIMO) with good isolation antennas with their suitable performance enhancement techniques. This paper gives a detailed review about the 5G wideband antenna designs, structural differences, substrate chosen, comparison, and future breakthroughs in a detailed way. This will be more useful for the various wireless communication applications such as the UWB, WLAN, and Wi-Max, etc., with a very less isolation of 15 dB over the larger dimensions. Keywords Ultra Wideband · High speed communication · Data rate · Enhancement techniques · Substrate · Antenna design

1 Introduction Nowadays, there is a huge demand for low power and cost with high data rate in Ultra Wideband communication systems. The Federal Communications Commission (FCC) allocates the band for UWB communication between 3.1 and 10.6 GHz. This will induce the researchers towards the UWB antennas and its design. Wide impedance matching, low profile, radiation stability, and low cost are some of the UWB antenna designing difficulties of feasible UWB antenna design. Similar to other wireless systems, UWB systems are also affected by Multipath Fading. In order to V. Baranidharan · S. Subash (B) · R. Harshni · V. Akalya · T. Susmitha · S. Shubhashree · V. Titiksha Department of Electronics and Communication Engineering, Bannari Amman Institute of Technology, Sathy, India e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 I. J. Jacob et al. (eds.), Expert Clouds and Applications, Lecture Notes in Networks and Systems 444, https://doi.org/10.1007/978-981-19-2500-9_30

403

404

V. Baranidharan et al.

overcome this issue, in UWB systems, Multiple Input Multiple Output (MIMO) technology is introduced. This technology will provide diversity gain, multiplexing gain, link quality, and making further increase in capacity [1]. In the designing stage of MIMO antennas for UWB systems, two major challenges are faced. First is to minimize the antenna elements. Next is to enhance the isolation between the antenna elements. To overcome design issues, many methods are introduced. One is UWB diversity antennas, the second method involves decoupling structures for high isolation properties [1]. The third method is a hybrid combination of both structures. A compact UWB MIMO antenna is explained here. For achieving a wide bandwidth in UWB applications, on the radiating element, there is formation of a staircase structure. Design of MIMO antenna seeks to minimize mutual coupling between the elements of the antenna. To suppress the coupling, a comb-line structure works as an electromagnetic band-gap. The antenna proposed has a size of 2631 mm2 , which is compact in size than many of the UWB antennas with a single element. The observed bandwidths range from 2.8 to 11.9 GHz [1]. The paper is structured as follows: Sect. 2 explains the different UWB MIMO antenna types and various techniques to enhance its performance. Section 3 gives the detailed discussion about the recently designed antennas by the researchers and their findings and antennas future breakthroughs. This paper concludes with all the findings in Sect. 4.

2 UWB MIMO Antenna, Classifications, and Its Performance Enhancement Techniques In recent years, a large number of 5G antennas have been designed by different researchers worldwide. This section explains UWB MIMO, antenna types, and its performance enhancing techniques.

2.1 UWB MIMO Antennas Generally, the wireless communication systems are always affected by severe interference, radiation loss, and more multipath fading. In order to overcome these issues, we need Multi Input Multi Output (MIMO) antennas to achieve better transmission range without signal increasing power. This effective design will help to achieve better throughput and high efficiency [2]. Different wideband and multi-band antennas are designed to achieve the efficient different frequency bands. Even-though the antennas may be a compact device, this will help to achieve the higher transmission rate, we need a proper isolation between the antennas [3]. Different types of enhancement techniques are available to improve the various antenna designs and its structures to increase the gain, better and efficient isolation from other single antenna

A Comprehensive Survey on Compact MIMO Antenna Systems …

405

element terminals, improve bandwidth, envelope Correlation Coefficient (ECC), and radiation efficiency [3].

2.2 Suitable Antennas for 5G Applications Based on the available literature, some of the antennas are most suitable for most of the 5G applications. They are, Dipole antenna: Two straight microstrip lines with each of its length λ/4, where λ is defined as the wavelength of the antenna’s resonant frequency [4]. For this effective antenna structures, proper feeding may be given in between the two strip lines used in Microwave Integrated circuits. The entire length of this dipole antenna is λ/2. Monopole antenna: Generally, the monopole antennas are having its length as λ/4 [5]. These monopole antenna structures are modified into different shapes based on the applications. Magneto-electric Dipole antenna: This antenna structure consists of two dipole structures. The first one is planar electric dipole and the second one is vertically shorted planar magnetic dipole. The signal feeding is given at the bottom side of the substrate. Loop antenna: Different types of loop antennas are designed such as rectangular, circular, and square or any other shapes based on the application. Antipodal Vivaldi antenna: This antenna structure consists of two different conductors which are placed on the both sides of the substrates. One side is always a mirror to the opposite side [6]. It consists of two conductors, one will act as a radiator, and the second will act as a ground. Fractal antenna: This type of the antennas consist of similar structures repeated multiple more times. Mathematical rules are always used to design fractal antennas [7]. Different shapes are used while designing the fractal antennas based on the applications. Inverted F antenna: This type of antenna structures consist of microstrip lines with some bends. The signal feeding point is given at the bend part and straight portion of microstrip lines [8]. So, the overall structure will look like an inverted F antenna. Planar Inverted F antenna: This type of antenna structures consist of ground plane and patch antenna. This ground plane is always connected with a short pin and signal feeding point which is given at the bottom side of the substrate.

2.3 Antenna Performance Enhancement Techniques Some of the performance enhancement designs are used to enhance bandwidth, efficiency, mutual coupling reduction, and size reduction. Some of these techniques are discussed here. They are,

406

V. Baranidharan et al.

Choice of substrate: Different types of permittivity and loss tangent values are very important for choosing the substrates for an effective antenna design [9]. This will automatically increase the gain and reduce the power loss. Multi-element: By using the multi-antenna elements will increase the gain of an antenna [9]. Moreover, this will increase the antenna bandwidth and radiation efficiency. Single antenna element design will not fulfil the requirements. Corrugation: An effective radiator design and its shapes (sine, triangular, square, and rectangular) will helps to improve the bandwidth ratio. Dielectric lens: The dielectric lens are uses the electrostatic radiations which are used to transmit it in the one direction. This will helps to lead the increase of the gain and antenna directivity [10]. Mutual coupling based reduction: In this MIMO antenna-multi-element design, the different single elements performance will affect the performance of another element in the antenna [10]. In order to reduce this interference, different types of mutual coupling techniques are used in MIMO antennas. This will help with isolated or decoupling techniques proposed by the different researchers.

3 UWB MIMO Antenna In this section, we discussed the various recently proposed antenna structures by the different researchers.

3.1 Printed UWB MIMO Antenna In this work, the authors proposed a new UWB frequency based MIMO antenna with a compact size. It’s printed on a substrate FR4 with a relative permittivity value of 4.4.and substrate thickness value as 0.8 mm, the dimensions of the antenna are improved for a reduced size. The described MIMO antenna is made up of two L-shaped slot antenna elements which are marked as LS 1 and LS 2. To provide good isolation between antenna elements, the two LSs are arranged perpendicular structures. An L-shaped slot and a rectangular patch comprise the element antenna, which is fed by a 50-microstrip line [11]. At the rectangular patch, a T-shaped stub is attached, which has a vertical stub and a horizontal stub, to gain the bandwidth increase for UWB applications. A tiny rectangular gap is carved on the left bottom of the ground plane to increase the isolation between antenna elements at low bands. Simulation using the electromagnetic (EM) simulation tool CST is used to acquire the appropriate geometrical parameters and correct numerical analysis (Fig. 1). To improve simulation precision, the SMA connector was incorporated in the simulated model. The MIMO antenna’s final optimised parameters are verified. The plotted simulated ECC curves show that the measured ECCs are less than 0.04 in

A Comprehensive Survey on Compact MIMO Antenna Systems …

407

Fig. 1 Geometry of the proposed printed MIMO antenna. Source [11], p. 1

the frequency range 3.1–10.6 GHz, which is less needed to satisfy better diversity performance for the suggested MIMO antenna.

3.2 High-Isolation Compact MIMO Antenna In this work, the authors proposed two MIMO antenna elements made up of Rogers RO4003 26 × 31 mm2 with a relative permittivity of 3.55 and of 0.7874 mm thickness [11]. The loss tangent is 0.0027. The UWB MIMO antenna is made up of 2 monopole elements. The element radiating is U-shaped, and there is a staircase design that is identical to the radiating element’s two corner’s bottom. When multiple antenna elements are installed in a narrow space, the mutual coupling might be very strong between the UWB antenna elements. As a result, designing UWB MIMO antennas with low mutual coupling and confined size is a challenge. The elements of the antenna are positioned in the H-plane [11]. The edge-to-edge gaps between the elements is set to 8 mm, equals to 0.180, where the free space wavelength is 0. A 50-microstrip line connects each antenna element. These microstrip lines have the dimensions W mm2 [11]. The ground plane has dimensions of WLg mm2 , and a rectangular slot at high frequencies with parameters of Wfs Lfs mm2 is removed on the top corner of the ground plane. The last stage of design included the metal strip connected by an addition of ground stubs with dimensions of Ws Ls mm2 , which formed a structure of comb-line on the antenna’s ground plane Several techniques for combining both techniques to reduce mutual coupling in elements while maintaining a confined size have been investigated (Fig. 2). It is demonstrated that comb-line design can effectively expand impedance bandwidth across the entire UWB band and improve isolation [11]. The authors persuaded that the MIMO antenna is a good choice for portable UWB applications based on measured and simulated antenna performance.

408

V. Baranidharan et al.

Fig. 2 Geometry of proposed highly isolated MIMO antenna. Source [11], p. 2

3.3 Compact UWB Four Antenna Element MIMO Antenna Using CPW Fed The authors proposed the UWB MIMO antenna’s configuration for high frequency data communications. At first, the authors designed it for a single element. HFSS is used for simulating the antenna, and the antenna’s S-parameters are determined. It is indicated by the S-parameter that a single antenna element has the capability of working in UWB [12]. Whereas, after 17 GHz, the antenna’s performance degrades. The performance of the antenna is boosted by adding parasitic branches to the feeder. Matching of a single element antenna in the frequency region higher than 10.6 GHz has clearly improved after the stub was added. The 4-element based modified MIMO antenna can then be further designed [13]. The antenna construction and size were chosen to suit the required operating frequency. All measurements are in millimetres. The estimated and simulated FRS were found to be 4 and 3.7 GHz, respectively [13]. The estimated value is almost similar to the simulated value, as can be observed. The radiating patch and ground plane were printed on the substrate’s surface. A circular monopole evolved into the patch shape that is currently in use. The impedance matching of the antenna is heavily influenced by the size of the half-circle ground construction (Fig. 3). The R3-radius semi-circular protrusion structure at the top of the patch, as well as the stub attached to the feeder line, can both help with antenna matching and Ultra Wideband performance. The stub, in specific, plays a vital role in impedance matching. The little rectangular notches on the left and right sides of the semi-circular protrusion are also referred to as W2 and L2, and they aid in impedance matching. The simulation results reveal that the antenna ports have excellent impedance matching at 3 GHz to 20 GHz bandwidth. In terms of isolation, the four ports are separated by more than 17 dB [13]. The available bandwidth emphasised, however, that the

A Comprehensive Survey on Compact MIMO Antenna Systems …

409

Fig. 3 Geometry of the proposed multiple input multiple output (MIMO) antenna with CPW feed. Source [13], p. 1

suggested S11 has been measured. At high frequencies, the UWB MIMO antenna system’s results aren’t completely equal to the measured results.

3.4 8-Element Based UWB MIMO Antenna with Band-Notch and Reduced Mutual Coupling A small 4-element based Ultra Wideband frequency based antenna with a ground plane of slotted is described. The bandwidth is expanded by etching a modified slotted ground plane with a compact size of 0.33 × 0.33 × 0.01, further the isolation between its elements is 14 dB [14]. A wideband MIMO antenna has more number of the common parts that can be used as a LTE wireless access point and Wi-Fi. The latter is composed of 4 feedline printed microstrip antennas consisting of a simple element radiating and a ground plane which is round shape; the impedance bandwidth is raised from 1.8 GHz to 2.9 GHz [14] (Figs. 4 and 5). The suggested antenna structure includes a relative permittivity of 4.4 which is made of FR4 substrate and with a width of 1.6 mm that is a low-cost component with a 0.02 tangent loss. To estimate S-parameters, gain, efficiency, and radiation, the electromagnetic simulator HFSS is used. The surface waves of the patterns ECC travel over the ground. By CST MWS, the EBG structures are focused.

410

V. Baranidharan et al.

Fig. 4 Geometry of designed single antenna element (front view). Source [14], p. 1

Fig. 5 Geometry of an antenna single element b back view. Source [14], p. 2

3.5 WLAN Band Rejected 8-Element Based Compact UWB MIMO Antenna For Future Fifth Generation (5G) terminal equipment a compact UWB MIMO antenna with eight elements has the ability of supplying high data rates. An Inductor Capacitor is used to achieve band rejection from 4.85 GHz to 6.35 GHz [14]. A (LC) stub is shown at the ground plane. The included stub also allows for bandwidth control and the rejection of any desired band. The printed monopoles’ orthogonal positioning allows for polarisation diversity and excellent isolation. As suggested in this antenna structure, monopole pairs which are located on opposite corners of a planar substrate of 50 mm2 [14] (Fig. 6).

A Comprehensive Survey on Compact MIMO Antenna Systems …

411

Fig. 6 Geometry of 8-element compact UWB MIMO antenna. Source [15], p. 4

On the same board, 4 more perpendicularly placed monopole antennas are found, with a total volume of only 50 × 50 × 25 mm3 [15]. On proper comparisons, the results are concluded. The target criteria was met with a coefficient of reflection which is less than 10 dB over the whole bandwidth of 2 to 12 GHz [15] (excluding the rejected band), greater than 17 dB isolation, minimal strength variation, high signal rejection, minimal envelope correlation, and stable radiation pattern in the Wireless Local Area Network band. The ability to reject WLAN bands in a 3D with vertical and horizontal antennas. The suggested design’s planar configuration, i.e. 50 × 50 mm2 , which is small when compared to other structures.

3.6 Comparison of UWB MIMO Wideband Antennas Structures In this section, we have discussed the pros and cons of the recently proposed MIMO wideband antenna structures (Table 1).

4 Conclusion In this paper, different types of 5G UWB antennas are analysed critically, and comparisons are made based on their performance enhancement techniques. These UWB MIMO antennas are highly categorised by the multi-element wideband and single

412

V. Baranidharan et al.

Table 1 Comparison of UWB Wideband structures UWB MIMO wideband antennas structure

Pros

Cons

Printed UWB MIMO antenna • Provides a high efficiency [10] • This decoupled structure based multi-element is always comparatively compact, light in weight, smaller size, and easy to design

• Structure is highly complex and fabricate

High-isolation compact MIMO antenna [10]

• It always provides the optimised level of design, gives a good bandwidth, and high efficiency • Antenna size is always compact based on the low ECC value

• It requires external components for an effective design

Compact UWB MIMO antenna with CPW feed [11]

• Design is very easy to fabricate • This antenna is more compact and provides an enhanced gain, bandwidth and narrow bandwidth of operations

• Difficult design process and uses metamaterial type of unit cells

Band-notch and reduced • In this antenna, high • Provides the low bandwidth mutual coupling based UWB isolation is provided by using and analyse its challenging MIMO antenna [12] DRA and using the common issue ground structures • Capacity of the channel is improved by the good orthogonal polarisation and diversity WLAN band rejected 8 × 8 MIMO antenna for UWB applications [13]

• The feeding network is • Difficult to design and designed by complicated placing the slot or parasitic dual polarisation elements with optical Slotted ground structure gives elements is not easy a better efficiency to elevate the ECC, polarisation diversity techniques

element antenna structures. The recently proposed antenna types are explained in detail based on the structures with its performance enhancement techniques. The antennas electrical and physical properties are analysed to enhance the overall performances. This paper also emphases on the upcoming breakthroughs of the 5G smartphones or terminal devices, 5G IOT, mobile terminals, and base stations. This paper will shed some light on 5G UWB antenna design for selecting the antenna substrate, suitable performance enhancement techniques to meet out the 5G applications.

A Comprehensive Survey on Compact MIMO Antenna Systems …

413

References 1. V. Baranidharan, G. Sivaradje, K. Varadharajan, S. Vignesh, Clustered geographic–opportunistic routing protocol for underwater wireless sensor networks. J. Appl. Res. Technol. 18(2), 62–68 (2020) 2. B. Varadharajan, S. Gopalakrishnan, K. Varadharajan, K. Mani, S. Kutralingam, Energyefficient virtual infrastructure-based geo-nested routing protocol for wireless sensor networks. Turk. J. Electr. Eng. Comput. Sci. 29(2), 745–755 (2021) 3. C.R. Jetti, V.R. Nandanavanam, A very compact MIMO antenna with triple band-notch function for portable UWB systems. Progr. Electromagn. Res. C 82, 13–27 (2018) 4. Z.J. Tang, X.F. Wu, J. Zhan, Z.F. Xi, S.G. Hu, A novel miniaturized antenna with multiple band-notched characteristics for UWB communication applications. J. Electromagn. Waves Appl. 32(15), 1961–1972 (2018) 5. X. Zhao, S.P. Yeo, L.C. Ong, Planar UWB MIMO antenna with pattern diversity and isolation improvement for mobile platforms based on the theory of characteristic modes. IEEE Trans. Antennas Propag. 66(1), 420–425 (2018) 6. L. Wu, Y. Xia, X. Cao, Z. Xu, A miniaturized UWB MIMO antenna with quadruple bandnotched characteristics. Int. J. Microw. Wirel. Technol. 10, 948–955 (2018) 7. A. Chaabane, A. Babouri, Dual band notched UWB MIMO antenna for surfaces penetrating application. Adv. Electromagn. 8(3) (2019) 8. Z. Tang, X. Wu, J. Zhan, S. Hu, Z. Xi, Y. Liu, Compact UWB-MIMO antenna with high isolation and triple band-notched characteristics. IEEE Access (2019) 9. M. Irshad Khan, M. Irfan Khattak, S. Ur Rahman, A.B. Qazi, A.A. Telba, A. Sebak, Design and investigation of modern UWB-MIMO antenna with optimized isolation. MDPI (2019) 10. J. Ren, W. Hu, Y. Yin, R. Fan, Compact printed MIMO antenna for UWB applications. IEEE Antennas Wirel. Propag. Lett. 13, 1517–1520 (2014) 11. N. Malekpour, M.A. Honarvar, Design of high-isolation compact MIMO antenna for UWB application. Prog. Electromagn. Res. C 62, 119–129 (2016) 12. D.A. Sehrai, M. Abdullah, A. Altaf, S.H. Kiani, F. Muhammad, M. Tufail, M. Irfan, A. Glowacz, S. Rahman, A novel high gain wideband MIMO antenna for 5G millimeter wave applications. MDPI (2020) 13. S. Kumar, A.S. Dixit, R.R. Malekar, H.D. Raut, L.K. Shevada, Fifth generation antennas: a comprehensive review of design and performance enhancement techniques. IEEE Access (2020) 14. W. Yin, S. Chen, J. Chang, C. Li, S.K. Khamas, CPW fed compact UWB 4-element MIMO antenna with high isolation. Sensors 21(8), 2688 (2021) 15. M.I. Khan, M.I. Khattak, G. Witjaksono, Z.U. Barki, S. Ullah, I. Khan, B.M. Lee, Experimental investigation of a planar antenna with band rejection features for ultra-wide band (UWB) wireless networks. Int. J. Antennas Propag. (2019)

A Review of Blockchain Consensus Algorithm Manas Borse, Parth Shendkar, Yash Undre, Atharva Mahadik, and Rachana Yogesh Patil

Abstract The advent of Blockchain started when a mysterious person or organization with an alias Satoshi Nakamoto published a white paper named “Bitcoin: A Peer-to-Peer Electronic Cash System.” This paper introduced a digital currency with no middlemen and no central authority. This meant, no transaction taxes, secure transactions, and a uniform currency. Hence, started the expedition of Blockchain Technology. Blockchain Technology needs consensus algorithms to insert a valid block of data to the Blockchain and maintain its state. Due to the rapid developments in Blockchain Technology and it’s adaptation to a plethora of wide areas (games, digital art, medical records, etc.), a study of consensus algorithms is essential to help researchers and developers to adapt a consensus algorithm according to their needs (proof of resource or majority voting). Keywords Consensus algorithm · Blockchain · Proof of work · Proof of stake · Proof of elapsed time · Byzantine fault tolerance

1 Introduction 1.1 Blockchain Technology Blockchain Technology is a decentralized, immutable, consensus based, distributed ledger. It is a peer-to-peer network with its nodes spread all over the world, reaching an agreement about the form and validity of transactions. Once this consensus, as we call it is reached, the transactions are stored in a block of data and linked to the previous block. This forms an unending immutable series of data blocks, hence Blockchain. There are diverse consensus algorithms which are being used today. This paper discusses the consensus algorithms. All the parties involved in the Blockchain network has to agree upon a valid form of ledger so that the data block can be added to the block chain. The protocol that M. Borse (B) · P. Shendkar · Y. Undre · A. Mahadik · R. Y. Patil Pimpri Chinchwad College of Engineering, Pune, India e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 I. J. Jacob et al. (eds.), Expert Clouds and Applications, Lecture Notes in Networks and Systems 444, https://doi.org/10.1007/978-981-19-2500-9_31

415

416

M. Borse et al.

Fig. 1 Taxonomy of consensus algorithms

allows these parties to come to a “consensus” is called a consensus Algorithm. The classification of consensus algorithm is shown in Fig. 1.

2 Related Work In this section, the existing blockchain consensus algorithms are studied and their advantages and disadvantages are analyzed.

2.1 Proof of Work Consensus Algorithm Bitcoin is still the most influential cryptocurrency system. Its Proof of Work consensus is a system that rewards the first individual who can solve a hard math problem. This system limits the number of miners who can successfully solve the problem. The authors of [1] gives a modification to the Bitcoin by taking advantage of the missed computational effort of miners. It would allow miners to justify their work and the network’s difficulty would then decrease accordingly. Due to the unique features of Blockchain, it has inspired a wide range of new social applications. However, its decentralized nature and its use of Proof of Work (PoW) technology can be very challenging to manage. The algorithm can only keep trail of the network hash rate quick enough [2]. However, it cannot set the target BPT on time. We introduce a (linear predictor based difficulty control algorithm) that takes into account the relationship between the hash rate and the difficulty. It achieves better stability and flexibility in terms of BPT. The crypto (Bitcoin) has utilized Proof of Work consensus to prevent double spending attacks [3]. However, it is not yet clear how this will affect the decentralized network’s security. Gavrilovic and Ciric [4] proposes a new protocol that takes into account the computing control of each server and regulates the trouble of making a block on the base of the evaluation.

A Review of Blockchain Consensus Algorithm

417

Majority of the emails traffic is spam. It is an abuse that is used for the determination of mass dissemination of unwanted messages. In [5, 6], solution needs a specific quantity of work from the recipient before the email message is sent. The extended SMTP protocol method would be evaluated through the use of the Proof of Work. assessment of the servers work and the influence of distributed spam on it will be shown. The proposed solution will be presented to help minimize server load and reduce spam traffic. A Blockchain is an open ledger technology that enables people to verify transactions. There is a way to improve scale and transaction speed is to find an answer that provides quicker Proof of Work algorithm. In [6], parallel mining method is introduced. The proposed method involves a selection process for a manager, a reward system, and a distribution of work. It was tested using a variety of test cases. Bitcoin, the world’s first digital money, is founded on the PoW (Proof of Work) consensus procedure. Its widespread adoption has raised the bar for Blockchain Technology. Feng and Luo [7] has also covered the various concepts of Blockchain and Proof of Work consensus algorithm along with its performance analysis. In [3], they propose a new protocol that will allow miners to generate blocks with equal difficulty and reward them with equal opportunity. The new method will also evaluate the computing power of the nodes. Based on the evaluation, the reward for the user’s incentive is also changed (Table 1).

2.2 Proof of Stake Consensus Algorithm Over the years Proof of Stake (PoS) consensus algorithms is developed and is to be added in the systems so that a distributed ledger can be added in the system and then being processed over the whole working of the system [10, 11]. The validators do not receive any amount of reward on the blocks that have been validated by them, instead they receive networking fees as their type of reward. Hence, they can be awarded in a way that has been a true form of reward. However, the technology used or developed in this system is currently limited. The Proof of Stake consensus algorithm’s operation is based on transactions, and their validation by the validators for the transaction or it being called as node to be added in the system [12]. The nodes are the ones who make the transactions (or the amount of transactions) in the system. All nodes who wish to be validators for the next block must stake a certain amount of money [13]. The highest amount of stake which was being validated is added in the system or is being elected as the validator [14]. Then, the validator verifies all the transitions which have being taken place in the system then if the transaction which was being processed was an authentic and proved transaction, then the validator does hold the transaction as valid and it adds it or publishes it to the block [15]. Now, if the block being added by the validator is verified then the transaction is valid and then the validator gets a stake back with a reward from the transaction

418

M. Borse et al.

Table 1 Reference papers studied of proof of work Reference

Advantages

Disadvantages

[1]

PoE utilizes computational power and increases mining power. Miners who fail at computational tasks might get experience

High energy consumption is advantageous to nodes that accumulate computational power

[2]

The prediction-based difficulty management method provides significantly improved stability and flexibility on BPT, as it’s based on the link between hash rate, difficulty, and BPT

Author prove that the prediction-based difficulty control technique cannot be implemented using smoothed BPT as a calculating method

[3]

This ensures that all miners have an equal chance of success. In addition, based on the evaluation, the reward for the user’s incentive is modified

Miner can easily attack against blockchain

[4]

The suggested approach reduc