139 102 25MB
English Pages [939]
Lecture Notes on Data Engineering and Communications Technologies 126
Subarna Shakya Klimis Ntalianis Khaled A. Kamel Editors
Mobile Computing and Sustainable Informatics Proceedings of ICMCSI 2022
Lecture Notes on Data Engineering and Communications Technologies Volume 126
Series Editor Fatos Xhafa, Technical University of Catalonia, Barcelona, Spain
The aim of the book series is to present cutting edge engineering approaches to data technologies and communications. It will publish latest advances on the engineering task of building and deploying distributed, scalable and reliable data infrastructures and communication systems. The series will have a prominent applied focus on data technologies and communications with aim to promote the bridging from fundamental research on data science and networking to data engineering and communications that lead to industry products, business knowledge and standardisation. Indexed by SCOPUS, INSPEC, EI Compendex. All books published in the series are submitted for consideration in Web of Science.
More information about this series at https://link.springer.com/bookseries/15362
Subarna Shakya · Klimis Ntalianis · Khaled A. Kamel Editors
Mobile Computing and Sustainable Informatics Proceedings of ICMCSI 2022
Editors Subarna Shakya Institute of Engineering Tribhuvan University, Pulchowk Campus Lalitpur, Nepal
Klimis Ntalianis University of Applied Sciences Athens, Greece
Khaled A. Kamel Department of Computer Science Texas Southern University Houston, TX, USA
ISSN 2367-4512 ISSN 2367-4520 (electronic) Lecture Notes on Data Engineering and Communications Technologies ISBN 978-981-19-2068-4 ISBN 978-981-19-2069-1 (eBook) https://doi.org/10.1007/978-981-19-2069-1 © 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 privileged to dedicate the proceedings of ICMCSI 2022 to all the participants and editors of ICMCSI 2022.
Preface
This conference proceedings volume contains the written versions of most of the contributions presented during the conference of International Conference on Mobile Computing and Sustainable Informatics (ICMCSI) 2022, held at Tribhuvan University, Nepal, during January 27–28, 2022. The conference provided a setting for discussing recent developments in a wide variety of topics including mobile computing, cloud computing, and sustainable expert systems. This conference has been a good opportunity for participants coming from various destinations to present and discuss topics in their respective research areas. ICMCSI 2022 conference tends to collect the latest research results and applications on mobile computing, cloud computing, and sustainable expert systems. It includes a selection of 65 papers from 322 papers submitted to the conference from universities and industries all over the world. All accepted papers were subjected to a strict peer-review system by 2–4 expert referees. The papers have been selected for this volume because of their quality and their relevance to the conference. ICMCSI 2022 would like to express our sincere appreciation to all authors for their contributions to this book. We would like to extend our thanks to all the referees for their constructive comments on all papers, and especially, we would like to thank the organizing committee for their hard work. Finally, we would like to thank the Springer publications for producing this volume. Guest Editors Lalitpur, Nepal Athens, Greece Houston, USA
Prof. Dr. Subarna Shakya Dr. Klimis Ntalianis Dr. Khaled A. Kamel
vii
Acknowledgments
The International Conference on Mobile Computing and Sustainable Informatics (ICMCSI) 2022 would like to thank the conference organizing committee and keynote speakers for their excellent work on January 27–28, 2022. The conference organizers also wish to acknowledge publicly the valuable services provided by the reviewers. On behalf of the organizers, authors, and readers of this conference, we would like to thank the keynote speakers and the reviewers for their time, hard work, and dedication to this conference. The conference organizers would like to acknowledge all the technical program committee members for the discussion, suggestion, and cooperation to organize the keynote speakers of this conference. The conference organizers also wish to acknowledge speakers and participants who attended this conference. Many thanks to everyone who has helped and supported this conference. ICMCSI 2022 wishes to acknowledge the contribution made to the organization by its many volunteers. Members contribute their time, energy, and knowledge at a local, regional, and international level. We also thank all the chairpersons and conference committee members for their support.
ix
Contents
XGBoost Design by Multi-verse Optimiser: An Application for Network Intrusion Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Milan Tair, Nebojsa Bacanin, Miodrag Zivkovic, K. Venkatachalam, and Ivana Strumberger Gateway-Based Congestion Avoidance Using Two-Hop Node in Wireless Sensor Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A. Revathi and S. G. Santhi
1
17
Stroke Prediction System Using Machine Learning Algorithm . . . . . . . . . Siddharth Purohit, Ankit Chahar, Anish Reddy Banda, Anson Antony, Abhishek Subhash Suryawanshi, and Chandan Vanwari
33
A Comprehensive Survey on Multilingual Opinion Mining . . . . . . . . . . . . Aniket K. Shahade, K. H. Walse, and V. M. Thakare
43
Development Features and Principles of Blockchain Technologies and Real Options as the Main Components of the Digital Economy . . . . . Radostin Vazov, Gennady Shvachych, Boris Moroz, Leonid Kabak, Vladyslava Kozenkova, Tetiana Karpova, and Volodymyr Busygin Technical Efficiency Analysis of China’s Telecommunication Infrastructure: A Copula-Based Meta-Stochastic Frontier Model . . . . . . Anuphak Saosaovaphak, Chukiat Chaiboonsri, and Fushuili Liu ATM Security Using Iris Recognition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Judy Simon, Ushus S. Kumar, M. Aarthi Elaveini, Reshma P. Vengaloor, and P. Phani Kumar
57
75 97
Reliable Data Acquisition by Master–Slave Approach in Marine-IoT Environment for Logistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109 Geetha Venkatesan and Avadhesh Kumar
xi
xii
Contents
Mobile Technology Acceptance of University Students: A Consolidated Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121 Zahra Hosseini, Jani Kinnunen, Mohammad Mehdizadeh, and Irina Georgescu Multistage Intrusion Detection System using Machine Learning Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139 N. Maheswaran, S. Bose, G. Logeswari, and T. Anitha Modeling Global Solar Radiation Using Machine Learning with Model Selection Approach: A Case Study in Tanzania . . . . . . . . . . . . 155 Samuel C. A. Basílio, Rodrigo O. Silva, Camila M. Saporetti, and Leonardo Goliatt Plant Disease Detection Using Transfer Learning with DL Model . . . . . . 169 Prakash Sahu and Vivek Kumar Sinha Tagging Fake Profiles in Twitter Using Machine Learning Approach . . . 181 Monika Singh A Machine Learning System to Classify Cybercrime . . . . . . . . . . . . . . . . . . 199 Shridevi Soma and Fareena Mehvin Analyses of Non-linear Effects with DCS and HOA Performance for 4 X 4 WDM/DWDM System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209 Tsegaye Menber Belay and Pushparaghavan Annamalai User Profiling and Influence Maximization . . . . . . . . . . . . . . . . . . . . . . . . . . . 221 Bahaa Eddine Elbaghazaoui, Mohamed Amnai, and Youssef Fakhri An Efficient IoT-Based Novel Design for Home Automation Using Node MCU Controller . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233 Naveen Rathee, Varnika Rathee, Sandeep Kumar, Archana Das, Yuliia Ivchuk, and Chornenka Liudmyla A Supervised Machine Learning Approach for Analysis and Prediction of Water Quality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247 Abhinav Mittra, Devanshu Singh, and Anish Banda Artificial Neural Network Established Hexagonal Ring- MPA Intended for Wideband Application . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265 Rohini Saxena, J. A. Ansari, and Mukesh Kumar EEGs Signals Artifact Rejection Algorithm by Signal Statistics and Independent Components Modification . . . . . . . . . . . . . . . . . . . . . . . . . . 275 Hussein M. Hussein, Kasim K. Abdalla, and Abdullah S. Mahmood IoT-Based Air Quality Monitoring System Using SIM900 . . . . . . . . . . . . . 291 P. Lavanya and I. V. Subbareddy
Contents
xiii
Predictive Modeling for Risk Identification in Share Market Trading—A Multiphase Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 301 R. V. Raghavendrarao, Ch. Ram Mohan Reddy, T. Sharmila, and H. M. Ankitha An Efficient Cross-Layered Approach Quality-Aware Energy-Efficient Routing Protocol for QoS in MANET . . . . . . . . . . . . . . . . 319 S. Jayaprada, B. Srikanth, Chokka Anuradha, K. Kranthi Kumar, Syed Khasim, and Padmaja Grandhe HealthCare Data Analytics: A Machine Learning-Based Perspective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333 Mangesh Bharate and Suja Sreejith Panicker A Novel Dual-Watermarking Approach for Authentication and Tamper Recovery of Colored Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . 355 Nahush Kulkarni, Reshma Koli, Nayana Vaity, and Dakshta Argade Blockchain-Based E-Pharmacy to Combat Counterfeit Drug Transactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 377 S. Kalarani, Keerthisree Raghu, and S. K. Aakash A Survey on Designing a Secure Smart Healthcare System with Blockchain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 391 Neelam Chauhan and Rajendra Kumar Dwivedi A Novel Ensemble of Classification Techniques for Intrusion Detection System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 405 Shaik Jakeer Hussain, N. Raghavendra Sai, B. Sai Chandana, J. Harikiran, and G. Sai Chaitanya Kumar An Resourceful System for Lossless Address Data Compression Using Novel Adaptive Algorithm in WSN . . . . . . . . . . . . . . . . . . . . . . . . . . . . 419 Sanjay Mainalli and Kalpana Sharma A New Paradigm in Cultivation Observing System Using NodeMCU and Blynk Application . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 431 G. Sasikala, Sai Venkata Raghavan Nookala, Nasir Shaik, and Veerendranath Reddy RTL Verification and FPGA Implementation of Generalized Neural Networks: A High-Level Synthesis Approach . . . . . . . . . . . . . . . . . . 447 Satyashil D. Nagarale and B. P. Patil Application of a Fuzzy Logic Model for Optimal Assessment of the Maintenance Factor Affecting Lighting in Interior Design . . . . . . . 463 Rahib Imamguluyev, Rena Mikayilova, and Vuqar Salahli
xiv
Contents
Comprehensive Analysis to Predict Hepatic Disease by Using Machine Learning Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 475 Reddy Shiva Shankar, P. Neelima, V. Priyadarshini, and K. V. S. S. R. Murthy Classification of Breast Tumor Using Ensemble Learning . . . . . . . . . . . . . 491 Aditya Singh and Vitthal Gutte Different Categories of Forwarding Routing Protocols in WSN (Wireless Sensor Network): A Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 509 Deepak Kumar, Vishal Kumar Arora, and Richa Sawhney Comprehending Object Detection by Deep Learning Methods and Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 523 Mallineni Priyanka, Kotapati Lavanya, K. Charan Sai, Kavuri Rohit, and Shahana Bano A Comprehensive Survey on Federated Learning: Concept and Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 539 Dhurgham Hassan Mahlool and Mohammed Hamzah Abed A Comprehensive Review of Optimisation Techniques in Machine Learning for Edge Devices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 555 P. Alwin Infant, P. N. Renjith, G. R. Jainish, and K. Ramesh AIC for Clients’ Perceived Risk on Online Buying Intention . . . . . . . . . . . 573 Dam Tri Cuong The DICE Framework: Efficient Computation Offloading through CASCADE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 585 P. Irene Komal, Anirudh Bathija, and K. Sindhu Depression Detection from Social Site using Machine Learning and Deep Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 599 Tushtee Varshney, Sonam Gupta, and Charu Agarwal An Improvised Machine Learning Approach for Wireless Sensor-Based Healthcare Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 613 V. Bharathi and C. N. S. Vinoth Kumar A Novel Deep Learning-Based Object Detector Using SPOTNET-SNIPER Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 627 Museboyina Sirisha and S. V. Sudha A Survey on Various Architectural Models Using Software-Defined Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 641 Alfred Raju M. and Narendran Rajagopalan A Survey on Alzheimer’s Disease Detection Using Machine Learning Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 659 N. L. Hemavathi, C. R. Aditya, and N. Shashank
Contents
xv
Lightweight Cryptography in IoHT: An Analytical Approach . . . . . . . . . . 665 Arnab Chakraborty and Payel Guria Prediction Model of the Burkholderia glumae Pest in Rice Crops Using Machine Learning and Spatial Interpolation . . . . . . . . . . . . . . . . . . . 681 Joel Perez-Suarez, Nemias Saboya, and A. Angel Sullon Application of Transfer Learning with a Fine-tuned ResNet-152 for Evaluation of Disease Severity in Tomato Plants . . . . . . . . . . . . . . . . . . . 695 R. Rajasree, C. Beulah Christalin Latha, and Sujni Paul Smart Home Technologies Toward SMART (Specific, Measurable, Achievable, Realistic, and Timely) Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . 711 P. Vinoth Kumar, B. Gunapriya, S. Sivaranjani, P. S. Gomathi, T. Rajesh, S. Sujitha, and G. Deebanchakkarawarthi Health Monitoring of Critical Care Patients Using Internet of Things . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 729 Y. Sri Lalitha, Varagiri Shailaja, K. Swanthana, M. Trivedh, and B. Chaitanya Kumar Quantum-Based Resource Management Approaches in Fog Computing Environments: A Comprehensive Review . . . . . . . . . . . . . . . . . 743 T. Veni Blockchain-Based Framework for Secure and Reliable Student Information Management System Using Artificial Intelligence . . . . . . . . . 753 Noor M. Abdulhadi, Noor A. Ibraheem, and Mokhtar M. Hasan Siamese Q&A: Distinguishing Unanswerable Questions Using Deep Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 763 L. Anantha Padmanabhan and G. Vadivu The State of the Art in Deep Learning-based Recommender Systems . . . 783 C. K. Raghavendra and K. C. Srikantaiah Varıous Frameworks for IoT-Enabled Intellıgent Waste Management System Usıng ML for Smart Cıtıes . . . . . . . . . . . . . . . . . . . . . 797 Karan S. Belsare and Manwinder Singh Multi-level Thresholding Partitioning Algorithm for Graph Processing in Cloud Computing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 819 J. Chinna and K. Kavitha Determining Trajectories for Hair Wash and Head Massage Robot Based on Artificial Neural Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 833 Nguyen Minh Trieu and Nguyen Truong Thinh Pineapple Eyes Removal System in Peeling Processing Based on Image Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 843 Nguyen Minh Trieu and Nguyen Truong Thinh
xvi
Contents
Threshold Optimization in Maximum–Minimum Eigenvalue-Based Detection in Cognitive Radio Using Ant Colony Optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 855 Anilkumar Dulichand Vishwakarma and Girish Ashok Kulkarni Glaucoma Diagnosis: Handcrafted Features Versus Deep Learning . . . . . 869 Shantala Giraddi, Swathi Mugdha, Suvarna Kanakaraddi, and Satyadhyan Chickerur Synthesis of the Variants of Adaptive Hexagonal Antenna Array Configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 883 Sridevi Kadiyam and A. Jhansi Rani Indexing, Clustering, and Search Engine for Documents Utilizing Elasticsearch and Kibana . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 897 Franz Frederik Walter Viktor Walter-Tscharf Q-Learning-Based Spatial Reuse Enhancement of Wireless Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 911 Gajanan Uttam Patil and Girish Ashok Kulkarni A Survey on Healthcare EEG Classification-Based ML Methods . . . . . . . 923 Abdulkareem A. Al-hamzawi, Dhiah Al-Shammary, and Alaa Hussein Hammadi The COVID-19 Images Classification by MobileNetV3 and Enhanced Sine Cosine Metaheuristics . . . . . . . . . . . . . . . . . . . . . . . . . . . 937 Miodrag Zivkovic, Aleksandar Petrovic, Nebojsa Bacanin, Stefan Milosevic, Vasilije Veljic, and Ana Vesic Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 951
About the Editors
Prof. Dr. Subarna Shakya is currently Professor of Computer Engineering, Department of Electronics and Computer Engineering, Central Campus, Institute of Engineering, Pulchowk, Tribhuvan University, Coordinator (IOE), LEADER Project (Links in Europe and Asia for engineering, eDucation, Enterprise and Research exchanges), ERASMUS MUNDUS. He received M.Sc. and Ph.D. degrees in Computer Engineering from the Lviv Polytechnic National University, Ukraine, 1996 and 2000 respectively. His research area includes E-Government system, computer systems and simulation, distributed and cloud computing, software engineering and information system, computer architecture, information security for E-Government and multimedia system. Dr. Klimis Ntalianis is Full Professor at the University of West Attica, Athens, Greece. He has worked as Senior Researcher in multi-million research and development projects, funded by the General Secretariat of Research and Technology of Greece (GSRT), the Research Promotion Foundation of Cyprus (RPF), the Information Society S.A. of Greece and the European Union. He is also serving as Senior Project Proposal Evaluator for GSRT, RPF, the European Union, the Natural Sciences and Engineering Research Council of Canada and the National Science Center of Poland. In parallel, he is Member of several master theses and Ph.D. evaluation committees in Greece, Cyprus, Germany and India. He is also serving as Promotion Evaluator for Saudi Arabia’s academic staff. He has served as General Chair in several conferences (IEEE, etc.). Dr. Ntalianis has published more than 160 scientific papers in international journals, books and conferences. His main research interests include social computing, multimedia analysis and information security. Dr. Khaled A. Kamel is currently Chairman and Professor at Texas Southern University, College of Science and Technology, Department of Computer Science, Houston, TX. He has published many research articles in refereed journals and IEEE conferences. He has more than 30 years of teaching and research experience. He has been General chair, Session Chair, TPC Chair and Panelist in several conferences and
xvii
xviii
About the Editors
acted as Reviewer and Guest Editor in referred journals. His research interest includes networks, computing and communication systems.
XGBoost Design by Multi-verse Optimiser: An Application for Network Intrusion Detection Milan Tair , Nebojsa Bacanin , Miodrag Zivkovic , K. Venkatachalam , and Ivana Strumberger
Abstract This article presents the results of an experimental study, which aims to assess the efficiency of the performance of a novel multi-verse optimiser algorithm for the optimisation of parameters of a network intrusion detection system event classifier. The article gives an overview of computer network intrusion detection, outlines common issues faced by software solutions tackling this problem, and proposes using a machine learning algorithm to help solve some of these common issues. An XGBoost classification model with a multi-verse optimisation algorithm for adaptive search and optimisation is used to solve the network intrusion detection system event classifier hyper-parameter optimisation problem. Results of this experimental study are presented and discussed, the improvements compared to previous solutions is shown, and a possible direction of future work in this domain is given in the conclusion. Keywords Optimisation · Multi-verse optimiser · Intrusion detection · Computer networks · Parameter tuning · Machine learning
M. Tair · N. Bacanin (B) · M. Zivkovic · I. Strumberger Singidunum University, Danijelova 32, 11000 Belgrade, Serbia e-mail: [email protected] M. Tair e-mail: [email protected] M. Zivkovic e-mail: [email protected] I. Strumberger e-mail: [email protected] K. Venkatachalam Christ (Deemed to be University), Bangalore, India e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 S. Shakya et al. (eds.), Mobile Computing and Sustainable Informatics, Lecture Notes on Data Engineering and Communications Technologies 126, https://doi.org/10.1007/978-981-19-2069-1_1
1
2
M. Tair et al.
1 Introduction At its core, intrusion detection (ID) has a very simple purpose. Its purpose is to monitor and analyse network traffic and user behaviour and detect unauthorised and malicious activity within a network. ID is commonly implemented on two fronts: network and host. Network ID systems (NIDS) analyse all network traffic from all hosts. In contrast, host ID systems (HIDS) analyse all network traffic coming to and emanating from a single host on the computer network [7]. The former are commonly implemented as network nodes on strategic points in the network, while the latter is usually in the form of an advanced firewall installed on the host computer [27]. According to [27], these two implementations perform differently for different types of threats. HIDS are more reliable when dealing with insider threats and for determining the extent of damage and for detecting irregular behaviour. On the other hand, NIDS are reliable when dealing with outside threats and intrusions and for detecting irregular behaviour, but are inefficient for determining the extent of damage from an attack. Two of the main problems with ID systems are rates of false positives (FP) and false negatives (FN). FP is when an ID system labels normal activity as an intrusion or another malicious type of activity. FN is when an ID system fails to identify a real intrusion or another malicious type of activity and allows it to go on undetected. To solve the problem of statistical mis-identification of events as FN or FP, different approaches are used. One of the possible solutions lies in the utilisation of machine learning (ML) algorithms. ML algorithms learn from patterns identified in training sets used to optimise the solution. It is important to solve the problem of ML algorithm’s hyper-parameter optimisation efficiently. This is an important step because for every classifier and used dataset, finding an optimal set of hyper-parameters, which give good classification results on the given dataset, is required to solve the given classification problem efficiently. Many ML algorithms can be used to optimise the machine learning model’s hyperparameters. Swarm intelligence (SI) algorithms are often used because of their good performance [3, 8, 22, 26]. In this study, a well-known XGBoost classifier is used for classifying NIDS events from a well-known ID dataset. The classifier’s hyperparameter optimisation is performed with the help of a novel optimisation algorithm called the multi-verse optimiser. The multi-verse optimiser (MVO) is inspired by concepts of white holes, black holes, and wormholes. It is designed for exploration, exploitation, and local search of solutions for single-objective optimisation problems [31]. Originally, MVO was tested on a selection of CEC 2005 benchmark functions [44]. The results of these tests and real engineering problem investigations from the original research have shown that the MVO algorithm was efficient in finding optimal solutions to these problems. In this article, the original MVO algorithm is applied on a real-life software engineering problem of network intrusion detection. Recent studies have explored weaknesses of concurrent intrusion detection applications [7]. There have been recent
XGBoost Design by Multi-verse Optimiser: An Application …
3
studies which have explored the use of metaheuristics and machine learning (ML) implementations for intrusion detection [21, 40, 46]. The original research [31] compares the results of the MVO algorithm to results by common competitor algorithms, such as the particle swarm optimisation algorithm (PSO), the grey wolf optimiser (GWO), the original genetic algorithm (GA), gravitational search optimisation algorithm (GSA), extreme learning machine (ELM) [33]. Performance results of all five algorithms on the problem of network intrusion detection is presented in this paper. The rest of the paper is structured as follows. Section 2 gives an overview of ID and SI and recent works in these fields. The 3rd section introduces the XGBoost classifier, the original MVO algorithm and presents the proposed MVO-XGBOOST framework. Section 4 details the experimental setup, findings of the comparative analysis, and the discussion of the results. Finally, the last section brings the final remarks, proposes future work, and concludes the research.
2 Background and Related Work This section introduces the problem of intrusion detection, its challenges, and some recently proposed and presented solutions in the literature. Common issues in solving this problem are presented as well. The section introduces ML as a viable and efficient method of finding optimised solutions for problems that have traditionally used time-consuming problem-solving methods. ML with metaheuristics algorithms has helped solve many problems in the industry where traditional solutions were having challenges. Even so, ML systems are not without their challenges [29]. ML approaches are viewed as enhanced forms of solutions to the problem of ID[39].
2.1 On the Problem of Intrusion Detection Traditional firewall systems rely on different detection techniques on the transport and the application layer [36]. These systems require precise policy configuration, network setup, protocol violation detection sub-systems, etc. These techniques allow firewalls to detect threats. However, recent advances in web technologies have seen a shift from the most common form of network traffic, based on the HTTP protocol, to more modern web application data exchange methods, such as SPDY, QUIC, WebRTC, WebSocket, IPFS. Threats for new data exchange methods have been identified. However, there is concern that the lack of research in the field of security of emerging technologies may hinder the identification of new threats and finding counter-measures [24].
4
2.1.1
M. Tair et al.
Existing Solutions of Intrusion Detection Systems
ID systems have existed since the 1980s. These systems were usually based on statistical analysis, protocol verification, state monitoring, and, very recently, ML approaches, including artificial neural networks (ANN), evolutionary algorithms (EA), and other similar supervised and unsupervised learning procedures [47]. Functional differences between a firewall and an ID system are that firewalls are active protective systems, used to prevent intrusions, while ID systems are passive [6]. As such, ID systems do not help directly prevent intrusions. Instead, they rely on the network administrator to react to possible detected intrusions and secure the network. In some cases, a synthesis of ID systems with intrusion prevention (IP) systems such as a network firewall can be automated to provide fast response time. Intrusion detection and prevention systems (IDPS) have become necessary due to the increasing number of attacks and ways attackers devise to counter even the strongest security strategies. The need for intelligent IDPS is evident. Intelligent ID systems use ML and artificial intelligence (AI) to facilitate their adaptability by self-managing their properties, configuration, optimisation, and knowledge used to distinguish malicious from regular activity [37]. To be able to create ML-supported intelligent ID system, it is required to understand some of the common issues faced by ID systems. This can help identify areas where ML support is most needed, where such methods can increase the ID system performance.
2.1.2
Common Issues and Intrusion Detection System Evaluation
The main problem of many NIDS is the mis-classification of events as false positives (FP) or false negatives (FN). These rates of FP and FN can be used as a performance metric when evaluating an ID system, in combination with rates of true positives (TP) and true negatives (TN). When metrics such as resource consumption and software dependencies are disregarded, a metric for determining the probability of correct classification can be defined as [20]: ACC =
TP + TN TP + FP + TN + FN
(1)
Equation 1 takes into account TP, FP, TN, and FN to calculate the ID system’s accuracy. With TP, FP, TN, and FN, it is possible to calculate the true positive rate (sensitivity), true negative rate (specificity), false positive rate (fallout), false negative rate (miss rate), and precision, as shown in Eqs. 2-6: Sensitivity = TP/(TP + FN)
(2)
Specificity = TN/(TN + FP)
(3)
Fallout = FP/(TN + FP)
(4)
XGBoost Design by Multi-verse Optimiser: An Application …
5
Miss rate = FN/(TP + FN)
(5)
Precision = TP/(TP + FP)
(6)
Metrics shown in Eqs. 1–6 are used to evaluate an ID classifier optimised using a ML algorithm, presented in the Methodology section.
2.2 Machine Learning-Supported Intrusion Detection Systems ML algorithms are used in many different fields of industry for optimisation and for finding solutions to computationally difficult problems. In computer network security, as support systems in NIDS solutions, different ML algorithms have been used [1, 45]. Some of these ML population-based algorithms include particle swarm optimisation (PSO) [28], artificial neural networks (ANN) [35], naive Bayesian (NB) [34], support vector machines (SVM) [35], K-nearest neighbour (KNN) [25]. Issues caused by incomplete data can be overcome by using the XGBoost model, due to its better performance in processing large amount of data, even when certain data points are missing, because of the specific structure of the ID datasets. Experimental research performed on ID datasets by using XGBoost-based models for classification has shown better performance compared to simply using the aforementioned base algorithms alone [14, 18, 28]. Among the mentioned applications of ML algorithms for the problem of ID, SI algorithms have proven efficient. An already mentioned PSO algorithm was successfully applied to the problem of ID [2, 28].
2.3 Swarm Intelligence Nature-inspired metaheuristic methods and swarm intelligence methods have frequently been used for solving NP-hard problems, in the field of information technology, such as global numerical optimisation [10], wireless sensor networks [5, 48], cloud-based system task scheduling problems [13], artificial neural network optimisation [4, 11, 17, 23, 30], predicting COVID-19 cases [49], MRI, and histological frame classifier optimisation [9, 12].
6
M. Tair et al.
3 Methodology This study proposes a new XGBoost-based model, supported by the MVO algorithm, for ID, tested on the NSL-KDD dataset [19, 38], which are effective in comparing and evaluating ID systems [38]. The methodology can be described as a series of steps. The first two steps are manual and include the acquisition and normalisation of the NSL-KDD dataset. These steps include loading the initial dataset training and testing sets, executing the optimisation algorithm with preconfigured parameters, using the training sets, testing the resulting solution using the testing sets, and finally, evaluating the solution’s performance against the whole dataset. Finally, another manual step of comparing the results of this study to the results of previous studies is performed.
3.1 The Multi-verse Optimisation Algorithm This section introduces and describes the original MVO algorithm, its features, and gives an outline of the way it functions, as well as where it has found uses in recently presented research. The MVO algorithm’s inspiration is rooted in cosmological concepts of white holes, black holes, and wormholes [31]. Adding onto the theory of the Big Bang, which suggests that the universe was created as a result of an explosion of energy and matter, the multi-verse theory suggests that there is more than one universe created as a result of multiple Big Bangs [15]. Some theories suggest that white holes might be reoccurring spontaneous events, consider the Big Bang itself a white hole, and a step in an endless cycle of creation and destruction of multi-verses [43]. These theories suggest that, as white holes are causes of the births of universes, so are black holes causes of their death. Additionally, wormholes are considered tunnels that connect different parts of a universe, enabling the transfer of information. Multiverse visions of wormholes are space-time tunnels that may connect parts of different universes [41]. The MVO algorithm is an extension of population-based algorithms for exploration and exploitation of search spaces. White holes and black holes explore, while wormholes exploit. Each universe is a potential solution, and bodies are features of those solutions. Universes have inflation rates proportional to their fitness functions. The optimisation process follows several rules, as specified in the original study [31]: • The probability of a universe having white holes is directly proportional to the inflation rate. • The probability of a universe having black holes is inversely proportional to the inflation rate. • The probability of bodies being sent to other universes through wormholes is directly proportional to the inflation rate.
XGBoost Design by Multi-verse Optimiser: An Application …
7
• The probability of bodies being received from other universes through wormholes is inversely proportional to the inflation rate. • Bodies can randomly move to the best universe through wormholes, regardless of the inflation rate. The original study [31] presents the conceptual model of the MVO algorithm. According to these descriptions, bodies move from one universe to another through white and black holes, which form wormhole tunnels. The tunnel direction depends on the inflation rates of the connected universes. The universe with the higher inflation rate gains a white hole and the other universe a black hole. Bodies move from the higher inflation universe to the lower inflation rate universe. The roulette wheel mechanism is used to model the exchange of bodies between universes by sorting the universes by their inflation rates at each iteration and choosing one which will gain a white hole [31]. Aside from the roulette wheel mechanism, which helps determine which universes can send bodies, the diversity of universes is achieved by allowing universes to exchange bodies randomly, regardless of the inflation level. It is assumed that there are always wormholes present between each of the universes, and the probability of them exchanging bodies is dependant on a random body exchange parameter of the algorithm, which is a subject of the optimisation process [31]. In the MVO algorithm, each universe contains several bodies, whose values represent values of the parameters of the problem being solved by the algorithm, here indicated as p. The total number of universes is configurable and is indicated as N . According to this, the set of all universes can be expressed as: ⎞ ⎛ b1,1 b1,2 · · · b1, p ⎜ b2,1 b2,2 · · · b2, p ⎟ ⎟ ⎜ (7) U =⎜ . .. . . . ⎟ ⎝ .. . .. ⎠ . b N ,1 b N ,2 · · · b N , p
In Eq. 7, each indexed b represents an individual body existing in all N universes, a total of p bodies (solution parameters) per universe. The MVO model can be further explained by the following two expressions: bk, j , r1 < N I (Ui ) bi, j = (8) r1 ≥ N I (Ui ) bi, j , In Eq. 8, bi, j is the jth body of the ith universe, from U , r1 is a random number in [0,1], N I (Ui ) is the normalised inflation rate of the ith universe, and bk , j is the jth body of the kth universe.
8
M. Tair et al.
bi, j
⎧ ∗ ⎪ ⎨b j + TDR × (ubi − lbi ) × r4 + lbi , = b∗j − TDR × (ubi − lbi ) × r4 + lbi , ⎪ ⎩ bi, j
r2 < WEP ∧ r3 < 0.5 r2 < WEP ∧ r3 ≥ 0.5 r2 ≥ WEP
(9)
In Eq. 9, b∗j is the jth body of the current best universe, ubi and lbi are the lower and upper bound of the ith universe, parameters r2 , r3 , and r4 represent random values in range [0,1], and TDR and WEP are coefficients called travelling distance rate and wormhole existence probability, used to enhance exploration and exploitation, respectively. They are calculated as follows: t × (M − m) + m T
k TDR = 1 − t/T
WEP =
(10) (11)
In Eqs. 10 and 11, T is the total number of iterations, t is the current iteration, k is the exploration factor, m and M are the minimum and maximum values for the W E P coefficient, respectively. The MVO algorithm spawns an initially random set of universes, which evolve in each iteration by moving bodies between them. In each iteration, the fitness of each universe is determined based on feature values. Finally, the fittest solution is selected. The MVO algorithm pseudocode is shown in Algorithm 1 listing. Algorithm 1 Pseudo-code of the MVO algorithm i, r1 , r2 , r3 , r4 ← Set to random values in [0,1] Used in Equations 8 and 9 T ← Set the maximum number of iterations N ← Set the population size U ← Create a set of initial random universes, as per Equation 7 U ∗ ← Identify the current best universe define N I as The Normalised Inflation Rate for each t in [1, ..., T ] do t is the current iteration Calculate the fitness of all universes SU ← Sort all universes by their fitness U ∗ ← Identify the current best universe for each u i ∈ U do for each b j ∈ Ui do for each body of universe Ui if r1 < N I (Ui ) then Calculate bi, j according to Equation 8 end if Calculate bi, j according to Equation 9 end for end for end for
XGBoost Design by Multi-verse Optimiser: An Application …
9
3.2 The XGBoost Optimiser The XGBoost optimiser is an improved gradient boosting decision tree optimisation tool, based on the gradient boosting decision tree, with the possibility of constructing boosted regression and classification trees, which can operate efficiently in parallel to optimise the value of the objective function. XGBoost is extensible and configurable. It works by scoring the frequency of use of a selected feature and by measuring the count and coverage of its impact on the function’s output [16]. XGBoost allows for fast computation and good performance. Since the XGBoost takes advantage of additive training optimisation, each iteration depends on the previous result. This property of the algorithm can be seen in the way the ith iteration’s objective function is calculated: Fo i =
n l yk , yˆki−1 + f i (xk ) + R( f i ) + C
(12)
k=1
In Eq. 12, Foi is the ith iteration’s object function, l is the loss term in that iteration, C is a constant term, and R is the model’s regularisation term: λ 2 w 2 j=1 j T
R( f i ) = γ Ti +
(13)
In Eq. 13, w represents the weights, defined later, γ and λ are parameters for configuring the tree structure, where larger values produce simpler trees. Optimisation of these parameters solves the over-fitting problem. The score of the loss function used to evaluate the tree structure, whose smaller values indicate a better structure of the tree, is calculated as: Fo
∗
T 2 g 1 +γT =− 2 j=1 h+λ
(14)
In Eqs. 12–14, g and h are the 1st and 2nd derivatives, which can also be used to calculate the weights w. The following three equations show how these derivatives and w are calculated. (15) g j = ∂ yˆki−1 l y j , yˆki−1 h j = ∂ y2ˆ i−1 l y j , yˆki−1
(16)
k
w ∗j = −
gt
(h t + λ)
(17)
10
M. Tair et al.
Table 1 XGBoost parameters optimised by MVO Parameter Default Range eta max_depth
0.3 6
[0,1] [0,+∞]
min_child_weight gamma
1 0
[0,+∞] [0,+∞]
subsample
1
(0,1]
colsample_bytree
1
(0,1]
Details Learning rate Maximum depth of the tree Minimum leaf weight Related to loss function Controls sampling to prevent over-fitting Controls feature sampling proportions
3.3 Proposed MVO-XGBoost Framework The MVO-XGBoost framework aims to optimise the parameters of the XGBoost classifier efficiently. These parameters include general model parameters, booster parameters, and dataset’s training set parameters. This model uses the MVO to optimise six parameters, shown in Table 1, that greatly influence the performance. The proposed MVO method was tested using 100 individuals in the swarm and 800 iterations, a total of 8000 fitness function evaluations (FFE). The same approach was utilised in the referenced paper [28].
4 Experimental Setup, Analysis, and Discussion This section gives an overview of the setup for conducting an experimental evaluation of the proposed MVO-XGBoost framework. Also, the section describes the dataset, its standardisation, normalisation, and structure. Statistical and comparative analyses are shown later in this section, followed by a discussion of the results. Because the proposed model is tested on an enhanced NSL-KDD dataset, inherent problems of the original DD99 dataset are overcome.
4.1 Experimental Setup This study uses the NSL-KDD dataset for the evaluation of ID systems to test the proposed model. The main characteristics of the dataset are described in detail in [38]. Some general information about the dataset is shown in Table 2.
XGBoost Design by Multi-verse Optimiser: An Application …
11
The dataset has five classes of events: normal use, denial of service (DoS) attack (the first attack type by the total number of attacks [32]), probe attack, user to root attack (U2R), and remote to local user (R2L). The dataset is split into training and testing sets. The detailed structure is shown in Table 3 [38]. Earlier research has shown that XGBoost-based solutions are good optimisation when combined with population-based metaheuristic algorithms for many real-life problems. The proposed framework was tested on the NSL-KDD dataset, following all setup instructions from [28] and substituting the algorithm proposed and tested in that study with the MVO algorithm. Of all 41 features, 9 are discrete values and 32 are continuous due to varying methods of measuring. Because of this, the data point in the dataset is standardised into a continuous range, using the following method: di j
=
M × di j −
M
di j
i=1
M M 1 di j di j − M i=1 i=1
(18)
In Eq. 18, M is the total number of dataset records, d is an individual data point, corresponding to the ith feature of the jth dataset record, and d is the standardised value of the corresponding data point. Upon standardising the dataset, it is normalised, using the following method: di j = di j − dmin (dmax − dmin )
(19)
Table 2 General information about the NSL-KDD dataset Property Description Number of records Number of features Number of classes Groups of attacks Types of attacks Number of sets
126,620 41 2 (normal uses and attacks) 4 (Probe, DoS, U2R, and R2L) 38 in total (21 in training set) 2 (a training and a testing set)
Table 3 Structure of the NSL-KDD dataset Event type Training set Normal use DoS Probe U2R R2L Total
67,343 45,927 11,656 52 995 125,973
53.46 % 36.46 % 9.25 % 0.04 % 0.79 %
Testing set 9711 7456 2421 200 2756 22,544
43.08 % 33.07 % 10.74 % 0.89 % 12.22 %
12
M. Tair et al.
In Eq. 19, d is the normalised data point value of the ith feature of the jth dataset record, while dmin and dmax represent the minimum and maximum values of the jth feature. The evaluation of the proposed model is done with common metrics, such as precision, recall, and F-score, with the addition of the P-R curve in place of the standard ROC curve, because of its better ability to capture the impact in measuring rare binary event situations, as explained in [42]. Such situations occur in the used dataset because of a small number of cases of events of the U2R attack type, compared to the other three types of attacks and the most numerous normal use-case event type. Processed values, based on the P-R curve, such as the average precision (AP), mean average precision (mAP), and performing the macro-averaging calculations can further be used better to evaluate the overall performance of the proposed model.
4.2 Comparative Analysis and Discussion This section presents the experiment results, with metrics such as detection precision, recall, F-score measurements. Results for the PSO supported version of the XGBoost framework, shown in Table 4, are acquired from [28]. The best results are marked in bold. As the presented results show, the proposed MVO-XGBoost approach slightly outperforms the referred PSO-based approach for all observed metrics and all observed classes, except the U2R class, where two methods were tied. Table 5 shows AP values from the P-R curves of both the PSO-XGBoost and the MVO-XGBoost models for each dataset event type or class. Values for the PSO version are acquired from the original article [28]. Again, it can be seen that the proposed MVO-XGBoost approach obtains better values for all classes except U2R, where the results were tied between the two methods. The reason why the approaches are tied for the U2R is an extremely small number of samples for this class (less than 1% of the total data). Therefore, the models are not able to learn how to classify this class well. The evaluation of the proposed MVO-XGBoost, compared to PSO-
Table 4 Dataset testing set optimal parameters confusion matrix Class Precision Recall F-score PSO MVO PSO MVO PSO MVO Normal use DoS Probe U2R R2L Average/total
0.66 0.80 0.95 1.00 0.96 0.81
Best results are marked bold
0.68 0.81 0.96 1.00 0.97 0.82
0.97 0.52 0.83 0.01 0.05 0.75
0.98 0.55 0.84 0.01 0.06 0.77
0.78 0.63 0.88 0.01 0.09 0.71
0.80 0.65 0.90 0.01 0.11 0.72
Support 9711 2421 7456 200 2756 22,544
XGBoost Design by Multi-verse Optimiser: An Application … Table 5 Comparison of AP values for each class Classifier Normal Probe DoS PSOXGBoost MVOXGBoost
13
U2R
R2L
0.90
0.79
0.94
0.15
0.49
0.92
0.81
0.95
0.15
0.51
Best results are marked bold
Fig. 1 Comparison of AP values between AdaBoost, bagging, random forest, PSO-XGBoost, and the introduced MVO-XGBoost
XGBoost and other common machine learning approaches, is shown in Fig. 1. The statistics for the PSO-XGBoost, random forest, bagging, and AdaBoost approaches were extracted from the published article [28]. The obtained results indicate that all tested methods obtain higher detection rates for the classes containing larger entries (such as normal, probe, and DoS). The reported results for PSO-XGBoost, random forest, bagging, and suggested MVO-XGBoost approach proposed in this article are similar, however, slightly in favour of the proposed MVO-XGBoost (1–3%). PSO-XGBoost finished in second place, while traditional random forest ended up in the third position. Even for the minor classes, namely U2R and R2L, the suggested MVO-XGBoost performed better, marginally outperforming the PSO-XGBoost that finished in second place (by 1–2%). At the same time, the difference to the other approaches was much more significant: approximately 5–7% for U2R class and 12– 15% for R2L class.
14
M. Tair et al.
5 Conclusion This paper has presented a proposed MVO-XGBoost framework, based on the XGBoost classifier and improved by using a new MVO metaheuristic algorithm for feature optimisation. The proposed classifier modification is trained and tested for network NID. Training and testing are done using a well-known NSL-KDD dataset. The performance of the proposed framework has shown that it has a better performance compared to other implementations. More precisely, the experimental results show that the MVO-XGBoost model obtained the best accuracy compared to other cutting-edge approaches. These findings also suggest a great deal of potential in utilising swarm intelligence metaheuristics in NIDS, opening the possibility of future research in this area. In future, we plan to implement other swarm intelligence metaheuristics to enhance the XGBoost model and also to utilise the developed model to address other classification challenges.
References 1. Ahmed, L.A.H., Hamad, Y.A.M.: Machine learning techniques for network-based intrusion detection system: a survey paper. In: 2021 National Computing Colleges Conference (NCCC). pp. 1–7 (2021) 2. Ajdani, M., Ghaffary, H.: Introduced a new method for enhancement of intrusion detection with random forest and PSO algorithm. Security and Privacy 4(2), e147 (2021) 3. Bacanin, N., Bezdan, T., Tuba, E., Strumberger, I., Tuba, M.: Optimizing convolutional neural network hyperparameters by enhanced swarm intelligence metaheuristics. Algorithms 13(3), 67 (2020) 4. Bacanin, N., Bezdan, T., Zivkovic, M., Chhabra, A.: Weight optimization in artificial neural network training by improved monarch butterfly algorithm. In: Mobile Computing and Sustainable Informatics, pp. 397–409. Springer, Berlin (2022) 5. Bacanin, N., Tuba, E., Zivkovic, M., Strumberger, I., Tuba, M.: Whale optimization algorithm with exploratory move for wireless sensor networks localization. In: International Conference on Hybrid Intelligent Systems, pp. 328–338. Springer, Berlin (2019) 6. Beal, V.: Intrusion Detection (IDS) And Prevention (IPS) Systems (2005, July). https:// webopedia.com/insights/intrusion-detection-prevention 7. Benkhelifa, E., Welsh, T., Hamouda, W.: A critical review of practices and challenges in intrusion detection systems for IoT: toward universal and resilient systems. IEEE Commun. Surv. Tutorials 20(4), 3496–3509 (2018) 8. Bergstra, J., Bardenet, R., Bengio, Y., Kégl, B.: Algorithms for hyper-parameter optimization. In: Shawe-Taylor, J., Zemel, R., Bartlett, P., Pereira, F., Weinberger, K.Q. (eds.) Advances in Neural Information Processing Systems, vol. 24. Curran Associates, Inc. (2011) 9. Bezdan, T., Milosevic, S., Venkatachalam, K., Zivkovic, M., Bacanin, N., Strumberger, I.: Optimizing convolutional neural network by hybridized elephant herding optimization algorithm for magnetic resonance image classification of glioma brain tumor grade. In: 2021 Zooming Innovation in Consumer Technologies Conference (ZINC), pp. 171–176. IEEE (2021) 10. Bezdan, T., Petrovic, A., Zivkovic, M., Strumberger, I., Devi, V.K., Bacanin, N.: Current best opposition-based learning salp swarm algorithm for global numerical optimization. In: 2021 Zooming Innovation in Consumer Technologies Conference (ZINC), pp. 5–10. IEEE (2021)
XGBoost Design by Multi-verse Optimiser: An Application …
15
11. Bezdan, T., Stoean, C., Naamany, A.A., Bacanin, N., Rashid, T.A., Zivkovic, M., Venkatachalam, K.: Hybrid fruit-fly optimization algorithm with k-means for text document clustering. Mathematics 9(16), 1929 (2021) 12. Bezdan, T., Zivkovic, M., Tuba, E., Strumberger, I., Bacanin, N., Tuba, M.: Glioma brain tumor grade classification from mri using convolutional neural networks designed by modified fa. In: International Conference on Intelligent and Fuzzy Systems, pp. 955–963. Springer, Berlin (2020) 13. Bezdan, T., Zivkovic, M., Tuba, E., Strumberger, I., Bacanin, N., Tuba, M.: Multi-objective task scheduling in cloud computing environment by hybridized bat algorithm. In: International Conference on Intelligent and Fuzzy Systems, pp. 718–725. Springer, Berlin (2020) 14. Bhati, B.S., Chugh, G., Al-Turjman, F., Bhati, N.S.: An improved ensemble based intrusion detection technique using XGBoost. Trans. Emergi. Telecommun. Technol. 32(6), e4076 (2021) 15. Carr, B., Ellis, G.: Universe or multiverse? Astron. Geophys. 49(2), 2–29 (2008) 16. Chen, T., Guestrin, C.: XGBoost: A scalable tree boosting system. In: Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pp. 785–794 (2016) 17. Cuk, A., Bezdan, T., Bacanin, N., Zivkovic, M., Venkatachalam, K., Rashid, T.A., Devi, V.K.: Feedforward multi-layer perceptron training by hybridized method between genetic algorithm and artificial bee colony. Data Sci. Data Anal. Opportunities Challenges 279 (2021) 18. Dhaliwal, S.S., Nahid, A.A., Abbas, R.: Effective intrusion detection system using XGBoost. Information 9(7), 149 (2018) 19. Dhanabal, L., Shantharajah, S.: A study on NSL-KDD dataset for intrusion detection system based on classification algorithms. Int. J. Adv. Res. Comput. Commun. Eng. 4(6), 446–452 (2015) 20. Diwan, T.D., Choubey, S., Hota, H.: A detailed analysis on NSL-KDD dataset using various machine learning techniques for intrusion detection. Turk. J. Comput. Math. Educ. (TURCOMAT) 12(11), 2954–2968 (2021) 21. Elmasry, W., Akbulut, A., Zaim, A.H.: Evolving deep learning architectures for network intrusion detection using a double PSO metaheuristic. Comput. Netw. 168, 107042 (2020) 22. Feurer, M., Hutter, F.: Hyperparameter Optimization, pp. 3–33. Springer International Publishing, Cham (2019) 23. Gajic, L., Cvetnic, D., Zivkovic, M., Bezdan, T., Bacanin, N., Milosevic, S.: Multi-layer perceptron training using hybridized bat algorithm. In: Computational Vision and Bio-Inspired Computing, pp. 689–705. Springer, Berlin (2021) 24. Ghasemshirazi, S., Heydarabadi, P.: Exploring the attack surface of websocket. arXiv preprint arXiv:2104.05324 (2021) 25. Govindarajan, M., Chandrasekaran, R.: Intrusion detection using k-nearest neighbor. In: 2009 First International Conference on Advanced Computing, pp. 13–20. IEEE (2009) 26. Guo, X., Yang, J., Wu, C., Wang, C., Liang, Y.: A novel LS-SVMs hyper-parameter selection based on particle swarm optimization. Neurocomputing 71(16–18), 3211–3215 (2008) 27. Hodo, E., Bellekens, X., Hamilton, A., Dubouilh, P.L., Iorkyase, E., Tachtatzis, C., Atkinson, R.: Threat analysis of IoT networks using artificial neural network intrusion detection system. In: 2016 International Symposium on Networks, Computers and Communications (ISNCC), pp. 1–6. IEEE (2016) 28. Jiang, H., He, Z., Ye, G., Zhang, H.: Network intrusion detection based on PSO-XGBoost model. IEEE Access 8, 58392–58401 (2020) 29. Jordan, M.I., Mitchell, T.M.: Machine learning: trends, perspectives, and prospects. Science 349(6245), 255–260 (2015) 30. Milosevic, S., Bezdan, T., Zivkovic, M., Bacanin, N., Strumberger, I., Tuba, M.: Feed-forward neural network training by hybrid bat algorithm. In: Modelling and Development of Intelligent Systems: 7th International Conference, MDIS 2020, Sibiu, Romania, October 22–24, 2020, Revised Selected Papers 7, pp. 52–66. Springer International Publishing (2021) 31. Mirjalili, S., Mirjalili, S.M., Hatamlou, A.: Multi-verse optimizer: a nature-inspired algorithm for global optimization. Neural Comput. Appl. 27(2), 495–513 (2016)
16
M. Tair et al.
32. Mugunthan, S.: Soft computing based autonomous low rate DDOS attack detection and security for cloud computing. J. Soft Comput. Paradigm (JSCP) 1(2), 80–90 (2019) 33. Mugunthan, S., Vijayakumar, T.: 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) 34. Mukherjee, S., Sharma, N.: Intrusion detection using Naive Bayes classifier with feature reduction. Procedia Technol. 4, 119–128 (2012) 35. Mukkamala, S., Janoski, G., Sung, A.: Intrusion detection using neural networks and support vector machines. In: Proceedings of the 2002 International Joint Conference on Neural Networks. IJCNN’02 (Cat. No. 02CH37290), vol. 2, pp. 1702–1707. IEEE (2002) 36. Neupane, K., Haddad, R., Chen, L.: Next generation firewall for network security: a survey. In: SoutheastCon 2018, pp. 1–6. IEEE (2018) 37. Patel, A., Qassim, Q., Wills, C.: A survey of intrusion detection and prevention systems. Inf. Manag. Comput. Secur. 18(4), 277–290 (2010). https://doi.org/10.1108/09685221011079199. https://www.emerald.com/insight/content/doi/10.1108/09685221011079199/full/html 38. Proti´c, D.D.: Review of kdd cup’99, nsl-kdd and kyoto 2006+ datasets. Vojnotehniˇcki glasnik 66(3), 580–596 (2018) 39. Sathesh, A.: Enhanced soft computing approaches for intrusion detection schemes in social media networks. J. Soft Comput. Paradigm (JSCP) 1(2), 69–79 (2019) 40. Seth, J.K., Chandra, S.: MIDS: Metaheuristic based intrusion detection system for cloud using k-NN and MGWO. In: Advances in Computing and Data Sciences, pp. 411–420. Springer Singapore, Singapore (2018) 41. Shatskii, A.A., Novikov, I.D., Kardashev, N.S.: A dynamic model of the wormhole and the multiverse model. Phys. Usp. 51(5), 457 (2008) 42. Sofaer, H.R., Hoeting, J.A., Jarnevich, C.S.: The area under the precision-recall curve as a performance metric for rare binary events. Methods Ecol. Evol. 10(4), 565–577 (2019) 43. Steinhardt, P.J., Turok, N.: A cyclic model of the universe. Science 296(5572), 1436–1439 (2002) 44. Suganthan, P.N., Hansen, N., Liang, J.J., Deb, K., Chen, Y.P., Auger, A., Tiwari, S.: Problem definitions and evaluation criteria for the CEC 2005 special session on real-parameter optimization. KanGAL Rep. 2005005(2005), 2005 (2005) 45. Tama, B.A., Lim, S.: Ensemble learning for intrusion detection systems: a systematic mapping study and cross-benchmark evaluation. Comput. Sci. Rev. 39 (2021) 46. Thapa, N., Liu, Z., KC, D.B., Gokaraju, B., Roy, K.: Comparison of machine learning and deep learning models for network intrusion detection systems. Future Internet 12(10) (2020) 47. Verwoerd, T., Hunt, R.: Intrusion detection techniques and approaches. Comput. Commun. 25(15), 1356–1365 (2002) 48. Zivkovic, M., Bacanin, N., Tuba, E., Strumberger, I., Bezdan, T., Tuba, M.: Wireless sensor networks life time optimization based on the improved firefly algorithm. In: 2020 International Wireless Communications and Mobile Computing (IWCMC), pp. 1176–1181. IEEE (2020) 49. Zivkovic, M., Venkatachalam, K., Bacanin, N., Djordjevic, A., Antonijevic, M., Strumberger, I., Rashid, T.A.: Hybrid genetic algorithm and machine learning method for covid-19 cases prediction. In: Proceedings of International Conference on Sustainable Expert Systems: ICSES 2020, vol. 176, p. 169. Springer Nature (2021)
Gateway-Based Congestion Avoidance Using Two-Hop Node in Wireless Sensor Networks A. Revathi and S. G. Santhi
Abstract Wireless sensor network (WSN) plays an essential role in our lifestyle. WSN may be a collection of random devices that are connected to networks. WSN technologies are attracted by worldwide researchers. Every node is connected to the sensors. WSN has been developed to perform high-level transmission. The two-hop cluster is used to decrease network congestion, energy consumption and increase performance. The data are transmitted through the gateway node when it transmits through the gateway. The gateway knows the path and data transmission. It collects the information from all node’s delivery times and transmitting energy states. It collects the congestion level of the clustering paths and shows the traffic intimation of the path, after changing the transmitting path. Keywords WSN · Gateway · Congestion · Clustering · Data transmission
1 Introduction The sensor network works with both static and dynamic. WSN monitors the physical and environmental conditions like surveillance, personal digital assistant (PDA) and security applications. All the sensor devices are connected to the network through nodes [1]. WSN comprises various wireless devices to send the data from one device to another. The sensor node’s battery power memory and coverage area are limited [2]. WSN is a network of small sensors that are randomly distributed around the network region. Every sensor node detects the surrounding location and transmits the data to the sink node across a remote intermediate node using single-hop or multi-hop transmission. When compared to other sensor nodes, this one handles more traffic to sink because it processes their energy faster, even if previous sensor nodes normally keep up with underlying energy. The present circumstance is known as energy-opening or area of interest issue. This issue prompts network segment, A. Revathi (B) · S. G. Santhi Annamalai University, Tamil Nadu, Annamalai Nagar, India e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 S. Shakya et al. (eds.), Mobile Computing and Sustainable Informatics, Lecture Notes on Data Engineering and Communications Technologies 126, https://doi.org/10.1007/978-981-19-2069-1_2
17
18
A. Revathi and S. G. Santhi
Fig. 1 WSN architecture
and thus, the whole network should not point all the time [3, 4]. WSN architecture is shown in Fig. 1. The transport layer is an end-to-end and peer-to-peer process on the remote. The transport layer gets data from the application layer and converts it into segments and transmits it into the next layer. This layer is the earliest one that breaks the information, supplied by the application layer into smaller units called segments. This layer ensures the data must be transmitted and received within the same sequences. This layer provides end-to-end delivery between hosts which can or cannot belong to an equivalent subnet. This layer has the main challenges congestion control and loss recovery. Congestion is a crucial challenge in WSN. Congestion control is a hot topic among researchers, as network traffic is growing rapidly and buffer mechanisms are changing regularly [5]. Congestion from the source node to the neighboring node is depicted in Fig. 2. Because congestion control aids in avoiding areas of traffic congestion, it is a highpriority concern [6]. Congestion control is divided into several stages. After the congestion deduction, a challenging sensor node has been informed for congestion during the congestion detection phase, during which congestion is identified to the sensor nodes in the notification phase. Recent traffic control technologies have some limitations in terms of traffic management, as they do not incorporate traffic appraisal Fig. 2 Congestion in WSNs
Gateway-Based Congestion Avoidance Using Two-Hop Node …
19
on alternate routes. The fundamental concern of the path from source to sink node is packet delay [7]. WSNs often employ buffer queue length without link-layer acknowledgments for congestion detection, which cannot accurately depict the presence of congestion [8]. The amount of time it takes for a channel to load detects network congestion rapidly. To compute the channel loading time to watch the channel and which consumes more energy, the sampling method [9] is used to sense the channel loading time. A certain wireless sensor application necessitates a particular reporting level. In such an application, the reporting level is frequently used to determine the level of congestion. Due to congestion, we consistently receive a lower reporting rate than the planned reporting rate. Once the packet arrival time exceeds the packet service time, a large number of packets are delayed, and the service time is also extended. As a result, congestion can be detected using the ratio of packet service time to packet inter-arrival time. The congestion control is to alter the incoming traffic rate [10] to control congestion. The state of the congestion influences the growth or reduction in speed. The temporary type of congestion can be efficiently mitigated with control. The control mechanism is less expensive and easier to implement than the resource management algorithm. In control, there is frequently a high packet loss at the time of the observed event; therefore, resource control algorithms can perform better in such instances. End-to-end control and hop-by-hop traffic control are two types of managed congestion [11]. Reducing traffic rates during a crisis is undesirable since it will result in significant data loss because the trustworthiness of knowledge in a critical situation is exceptionally high. The resource control approach is used in these situations [12]. To alleviate congestion, additional resources are deployed around the congestion hotspot. Within the congestion hotspot, extra bandwidth or nodes is frequently deployed [13]. The congestion problem must be solved in a wireless environment with a random and dense topology for the systems to function properly. Wireless networks produce more unstable connections between nodes than cable networks. Packets are sent many times due to the unstable connection, which should induce network congestion. Furthermore, the typical environment for sensor networks is a random and dense topology, which collects network congestion. Multiple-object tracking applications, in particular, create a large number of data transmissions and so may be affected by this problem [14]. One of the normal issues in WSN is congestion which causes due to data being moved through the organization. Subsequently, it builds the deferral and causes information misfortune which is corrupting the norm of administration by diminishing the lifetime of the organization and the crumbling of geography in different parts. The clog occurring over the organization has the restricted energy for more retransmissions and parcel drops and hinders the occasion location dependability [15]. A fast congestion control (FCC) procedure is dependent on routing with hybrid improvement calculation. The proposed conspire comprises of two handling steps. The multiple input time on task improvement calculation for choosing legitimate
20
A. Revathi and S. G. Santhi
next hop with negligible undesirable delay. Then, at that point, we proposed adjusted gravitational analysis calculating energy efficiency between source to destination [16]. The remainder of the paper is organized as follows: In Sect. 2, related work is examined. Section 3 expands on the problem statement. In Sect. 4, the proposed GCA is outlined. The simulation result is used to assess Sect. 5 performance. Finally Sect. 6 ends with the conclusion.
2 Related Work The identification and avoidance of congestion are a critical feature of sensor networks. Floyd and Jacobson’s random early detection (RED) scheme proposes that when a transitional node encounters congestion, it drops the packets. As a result, the source is sufficiently informed by a break or a negative acknowledgment (NACK). We investigate various solutions [17] because the dropping of packets is determined by sensor networks. Chieh Yih Wan et al. Congestion detection and avoidance (CODA) was proposed by the authors, and numerous strategies were approached with traffic congestion. It includes congestion detection via receivers, open-loop hop-by-hop backpressure, and closed-loop multi-source regulation [18]. Yogesh Sankara Subramaniam et al. The event-to-sink reliable transport protocol was proposed by the authors (ESRT). ESRT could be a revolutionary transport resolution designed to achieve the consistency of event detection in WSN while consuming minimal power. It has a congestion detection module that is processed for two reasons: dependability and energy conservation. ESRT’s planning is mostly based on the sink, with a negligible value desired at the expense of sensor nodes [19]. Xiaoping Yang et al. Particle swarm–neural PID congestion control is a standard particle swarm–neural PID congestion control, according to experts (PNPID). Initially, the PID mechanism was implemented to the remote sensor network’s executive line. The purpose of achieving an online change of loads to guide the extent, essential, and differential limitations of the PID regulator was to achieve selflearning and self-coordinating capacity of neurons. Finally, for online optimization, the conventional particle swarm–neural PID congestion control (PNPID) calculation of quantity initial values, integral and differential calculation, and neuron learning rates were used [20]. Saneh LataYadav et al. offered an efficient congestion avoidance strategy based on the Huffman coding algorithm (ECA-HA) and ant colony optimization to increase network performance. This technique combines source-oriented optimization with accessible traffic. This is the first time ant colony optimization has been used to locate many traffic-free alternative paths. Moving forward, the ant ensures that paths from the sink to the source node, as well as multiple-congestion paths from the source to the sink node, are created [21].
Gateway-Based Congestion Avoidance Using Two-Hop Node …
21
Jeongyeup Paek et al. 2010, The rate-controlled reliable transport (RCRT) protocol is designed by sensor nodes with limited resources. End-to-end express packet loss retrieval is used in RCRT, but all congestion detection and rate transformation effectiveness are done at the sinks. Sinks decide on rate allocation options, and because they have a broader perspective on network methods, they can achieve greater efficiency. It is possible to change the rate assignment selections for a similar reason [22]. Faisal Karim Shaikh et al. TRCCIT stands for tunable reliability with congestion control for information transport in WSNs. TRCCIT uses limited techniques including probabilistic flexible retransmission, crossover verification, and retransmission time management to contribute probabilistically assured adjustable reliability. TRCCIT reduces network congestion by opportunistically transferring information via numerous paths [23]. Chonggang Wang et al. The node requirement list in the priority-based congestion control protocol (PCCP) is well-known for expressing the importance of each node. PCCP measures a parameter called congestion unit based on packet arrival time and transmission time and provides hop-by-hop traffic control based on the measurement of congestion unit as the significance node [24]. Haoxiang et al. The routing protocol is so important in improving the performance of the WSN; the proposed method aims to build an efficient routing protocol that takes into account all of the metrics that are important for maintaining link stability and extending the network’s life. Because the sensors are powered by a battery and have limited battery power, one of the important factors determining the network’s longevity is energy usage [25]. Jeena Jacob et al. The artificial bee colony optimization technique and its evaluative features have a favorable impact on wireless network communication. In this method, the basic behavior of bee agents assists in making synchronous and decentralized routing decisions [26].
3 Problem Statement For the network traffic depicted in Fig. 3 and the reliability of specific packets, the congestion transport layer is established. In many WSN applications, similar to Fig. 3 Traffic around the network area in WSNs
22
A. Revathi and S. G. Santhi
the networks, the congestion management method requires the level of traffic that is delivered into the WSN in order to reduce packet loss and provide end-to-end depending packet delivery. WSN experts are questioning the need for a transport layer because WSN is essentially the same as the internet. In WSNs, the queue length is not fixed. As a result, data overflow causes congestion. In WSN, a queue is required. Researchers, in addition to admitting the network-scaffolding limitation of sensor devices, also acknowledge the network-scaffolding constraint of sensor devices. The multi-hop nodes are works with large distance. Every node is send the acknowledgement to the source. The transmission energy will be affected for all nodes. So, the energy loss will be happened in WSN. In Fig. 3, the congestion is occurred in the upper nodes.
4 Proposed Work In this section, the proposed technique is used to avoid congestion in WSN. The two-hop intermediate node gateway-based congestion avoidance is shown in Fig. 4.
4.1 Network Model • The nodes are randomly initialized and placed in the network region. • The cluster head (CH) chosen depends on the node’s energy and distance. • The base station (BS) energy is based on the energy limits of the data packets received and is aware of the nodes’ environmental locations. • Every sensor node has identical capabilities (sensing, processing, and communication), as well as the ability to dynamically modify its transmission power level based on signal strength.
Fig. 4 Two-hop gateway-based clustering
Gateway-Based Congestion Avoidance Using Two-Hop Node …
23
• The communication range of the sensor node is two-hop intermediate node communication. • The gateway collects the information from all the nodes and selects the CH using energy level, distance, and communication bandwidth. The main role of the CH is to collect the data from sensor nodes and deliver it to the BS. The CH must maintain the cluster data, which consume more energy and pass the data to the BS. G(x, y) =
R(x).E d(y, r )2 2 2 2 + d(x, y) + d(x, z) + d(y, z) + R(i).Max d(x, r )2
(1)
where Eq. 1 represents the gateway of x and y value calculation. • • • • • • • •
R(x) CH’s remaining energy is represented by E. R(i) The original energy is represented by Max. The distance between CH x and CH y is represented by d(x,y) The distance between CH x and member node z is represented by d(x,z). The distance between CH y and member node z is represented by d(y,z). The distance between CH y and BS is represented by d(y,r). The distance between CH x and BS is represented by d(x,r). d(x,r) represents distance between CH x to BS
4.2 Gateway-Based Congestion Avoidance (GCA) Mechanism The source neighboring nodes and low-energy consumption occurred nodes are selected as the gateway node. The two-hop cluster is used to reduce the network traffic and increase performance. Here, we have to add a gateway node to identify the congestion and avoid the congestion in WSN. The gateway node is made to regulate the traffic and transmit the data through another path. The gateway node is predicated on signal strength. The maximum weightage of the node and nearest node of the source node are considered as the gateway node. Gateway = N x(D,E) ≤ Nn(D,E) where Eq. 2, N x(D,E) Nn(D,E) Nn
represents node x distance and energy. represents node n distance and energy. represents the number of nodes.
(2)
24
A. Revathi and S. G. Santhi
4.3 Congestion Notification The gateway node identifies the load and gets notifications from intermediate nodes. The notification information must be sent to the congested node’s neighbors. Load Balancing = (1 − TG ) × intervalprev ) − (TG × intervalnew )
(3)
Equation (3) indicates the exponential gateway task’s weighted traffic load, where T G stands for traffic gateway function and the before variables intervalprev and intervalnew stand for the interval between the previous and new data packets. Congestion avoidance procedures with gateway support are the proposed mechanism. Each component has its own algorithm, which is detailed in the algorithm. After examining the buffer, the interval between two sequential packets is calculated when the node is congested. This is due to the fact that if the interval is large, the chances of filling the queue or buffer are also large, resulting in packet loss. Source knows the data transmission and transmission path. Here, we have to feature the gateway node in clustering. The data are transmitted through the gateway
Gateway-Based Congestion Avoidance Using Two-Hop Node …
25
node when it transmits through the gateway. The gateway knows the path, and data are transmitted. It collects the knowledge from all node’s delivery times and transmitting energy states.
4.4 Congestion Avoidance Gateway nodes are connected around the source to the BS. It collects the traffic level of the clustering paths and shows the traffic intimation of the path after changing the transmitting path. It works with two-hop intermediate node transmission. The twohop intermediate node is easy to identify the shortest path and traffic level. Therefore, we avoid the congestion using a two-hop intermediate node.
Algorithm 2: Conges on No fica on and Avoidance Start IF CE≤ Th//CE: The amount of energy u lised by the node. //ETh: The maximum amount of energy required to be classified as a weak node. b< 0.7 × max( max) // b: The buffer's consumed and residual capacity, respec vely. //c min: The me between the two data packets must be kept to a minimum. // max: The buffer's maximum capacity // : The sensor node to be congested If is near to then Discover the less congested route Else If nc is close to D or in the middle of the route//D: Data packets Call: GCA Else If (1 – TG) × intervalprev) − (TG × intervalnew)> 0.7 Call: GCA End If Send a message to predecessor nodes no fying them of the change End If To send to preceding nodes, create a message (frame) If a path is free then Send the data/control frame signals that are required Else Bk Off = Bk Off + 1 // Bkoff the counter for the rest of the perıod If Bk off ≥ 6 Call: GCA Else Interval between back-offs End If End If End
Figure 5 explains the workflow of the proposed work. The cluster is formed randomly and selects the CH using the GCA mechanism.
26
A. Revathi and S. G. Santhi
Fig. 5 Gateway-based congestion avoidance (GCA)
5 Results and Discussion The proposed GCA mechanism is implemented in the network simulator tool (NS2). The simulation settings listed in Table 1 are used to implement the algorithm. The BS is located in the sensing field’s center, a 100 * 100 m area with hundred nodes. Hundred nodes are the medium size of the network area and also able to add more than hundred nodes in NS-2. Table 1 Simulation parameters
Parameters
Values
Number of nodes
100
Number of sink nodes
1
Number of mobile nodes
5
Number of clusters
5
Network area
100 m ∗ 100 m
Gateway-Based Congestion Avoidance Using Two-Hop Node …
27
5.1 Performance Metrics The performance of the proposed mechanism GCA is examined using NS2 in this section. The network life is defined as the time it takes for the primary node’s remaining energy to reach zero. GCA is compared with TRCCIT and ESRT, and both existing mechanism are used to control the congestion previously. In Equation 4, the delay is based on the packet transmission time in the network. The parameter was calculated using milliseconds (ms). Delay =
Packet Received Time − Packet Sent Time n
(4)
Figure 6 shows the proposed method produces more clusters than the conventional method. Packet loss and end-to-end delay are reduced as the number of nodes. In network, each process is based on energy using Eq. 5; the packet transmission takes the energy level is called energy consumption. Energy consumption = Threshold energy − current energy
(5)
From Fig. 7, the graphical representation of the energy consumption is 10% reduced to the proposed mechanism. Fig. 6 Node versus delay
Fig. 7 Node versus energy
28
A. Revathi and S. G. Santhi
Fig. 8 Node versus packet delivery ratio
In Fig. 8 the packet delivery ratio is compared with the existing mechanism. In the above, ratio is 10% better than the existing system. Packet Delivery Ratio =
Packets Received Time
(6)
Packet delivery ratio is calculated using an Eq. 6, packet transmission successfully. Total packets and received packets are calculated and compared with the existing system. Figure 9 shows the number of data that can be sent from the source to destination throughput of around packets per second increased the speed of the network using the GCA mechanism calculation based on Eq. 7. Throughput =
Packets Received + Packet Size No. of Nodes
(7)
From Fig. 10, the proposed method gives an increased speed than the existing method. The proposed GCA reduces the packet loss and delay. Fig. 9 Node versus throughput
Gateway-Based Congestion Avoidance Using Two-Hop Node …
29
Fig. 10 Speed versus Delay
Fig. 11 Speed vs energy
Speed =
Distance Time
(8)
Equation 8 used to calculate delay in proposed GCA. Figure 11 shows the energy consumption is reduced using the GCA mechanism and the speed level is increased, when compared with the previous mechanism. Energy Consumption =
Total Energy No. of Packet Transferred
(9)
In GCA mechanism, an energy consumption is calculated using Eq. 9. From Fig. 12, the proposed method gives an increased speed of the packet delivery ratio than the existing method. Packet Delivery Ratio =
(Total Packet Received time) Total Packet Sent time
(10)
In Eq. 10, the packet delivery ratio is calculated using packet sent and packet received time. Figure 13 exhibits a high throughput of 1 Mb/s with a long network lifetime using the GCA technique.
30
A. Revathi and S. G. Santhi
Fig. 12 Speed versus packet delivery ratio
Fig. 13 Speed versus throughput
Throughput =
sum(No. of Packets tranmitted ∗ average packet size Transmission Time
(11)
In Eq. 11, the throughput is calculated using speed with transmission time.
6 Conclusion The two-hop intermediate node is used to reduce the o minimize the network’s energy consumption. The gateway could analyze the congestion in the network area and send the data through another path. The gateway knows the path of the data transmission. It collects the information from all the node’s delivery times and transmitted energy levels. It collects the congestion level of the transmitting paths, shows the traffic level, and decides to transmit the data another way. This mechanism improves the network lifetime, traffic reduction, and packet delivery at a particular time.
References 1. Li, X., Li, D., Wan, J., Vasilakos, A.V., Lai, C.F., Wang, S.: A review of industrial wireless networks in the context of industry 4.0. Wirel. Netw. 23(1), 23–41 (2017)
Gateway-Based Congestion Avoidance Using Two-Hop Node …
31
2. Abdur Rahman, Md., Saddik, A.E., Gueaieb, W.: Wireless sensor network transport layer: state of the art. Sensors. Springer, Berlin (2008) 3. Mehto A., Tapaswi S., Pattanaik, K.K.: Rendezvous point-based delay-efficient trajectory formation for mobile sink in wireless sensor networks. In: Proceedings of IEEE International Conference on Computing, Communication and Networking Technologies (ICCCNT), pp. 1–6. Kanpur, India (2019) 4. Mehto, A., Tapaswi, S., Pattanaik, K.K.: PSO-based rendezvous point selection for delay efficient trajectory formation for mobile sink in wireless sensor networks. In: Proceedings of IEEE International Conference on Communication Systems and NetworkS (COMSNETS), pp. 252– 258. Springer, Bengaluru, India, (2020). https://doi.org/10.1109/COMSNETS48256.2020.902 7330 5. Zhang, J.: Congestion avoidance and control mechanism for multi-paths routing in WSN, In: 2008 International Conference on Computer Science and Software Engineering, pp. 1318–1322 (2008) 6. Misra, S., Tiwari, V., Obaidat, M.S.: Adaptive learning solution for congestion avoidance in wireless sensor networks, In: IEEE/ACS International Conference on Computer Systems and Applications, pp. 478–484 (2009) 7. Kafi, M.A., Othman, J.B., Badache, N.: A survey on reliability protocols in wireless sensor networks. ACM Comput. Surv. 50(2) (2017) 8. Kandris, D., Vergados, D.J., Vergados, D.D., Tzes, A.: A routing scheme for congestion avoidance in wireless sensor networks. In: 6th Annual IEEE Conference on Automation Science andEngineering, Marriott Eaton Centre Hotel, Toronto, Ontario, Canada, August, pp. 21–24 (2010) 9. Babar Nawaz, K.M., Khan, J., ul Hassan, M., Shah, A.M., Saeed, M.K.: Congestion control techniques in WSNs: a review. (IJACSA) Int. J. Adv. Comput. Sci. Appl. 10(4), 194–199 (2019) 10. Kharat, S.: Congestion control techniques in wireless sensor networks: a survey. Int. J. Eng. Res. Technol. (IJERT) 4(2), 617–620 (2015) 11. Narawade, N.E., Kolekar, U.D.: congestion avoidance and control in wirelesssensor networks: a survey. In: International Conference on ICT in Business Industry and Government (ICTBIG), Electronic, (2016). ISBN 978-1-5090-5515-9 12. Moon, S., Lee, S., Cha, H.: A congestion control technique for the near-sink nodes in wireless sensor networks. Springer, Berlin (2006) 13. Chen, L.J., Low, S.H., Chiang, M., Doyle, J.C., Cross-layer congestion control, routing and scheduling design in ad hoc wireless networks. In: Proceedings 25th International Conference on Computer Communications, pp. 1–13 (2006) 14. Palattella, M.R., Accettura, N., Dohler, M., Grieco, L.A., Boggia, G.: Traffic-aware scheduling algorithm for reliablelow-power multi-hop IEEE 802.15.4e networks. In: 2012 IEEE 23rd International Symposium on Personal, Indoor and Mobile Radio Communications-(PIMRC), 327–332 (2012) 15. Heikalabad, S.R., Ghaffari, A., Hadian, M.A., Rasouli, H.: DPCC: dynamic predictive congestion control in wireless sensor networks. IJCSI Int. J. Comput. Sci. Issues 8(1) (2011) 16. Raman, C.J., James, V.: FCC: fast congestion control scheme for wireless sensor networks using hybrid optimal routing algorithm. Springer, Berlin (2018) 17. Floyd, S., Jacobson, V.: Random early detection gateways for congestion avoidance. IEEE/ACM Trans. Netw. (1993) 18. Wan, C., Eisenman, S., Campbell, A.: CODA: congestion detection and avoidance in sensor networks. In: The 1st International Conference on Embedded Networked Sensor Systems. ACM Press, Los Angeles, pp. 266–279 (2003) 19. Sankara Subramaniam, Y., Akan, O.B., Akyildiz, I.: ESRT: event-to-sink reliable transport in wireless sensor networks. IEEE Mob Hoc 2004. ACM Press, Annapolis, pp. 177–188 (2003) 20. Yang, X., Chen, X., Xia, R., Qian, Z.: Wireless sensor network congestion control based on standard particle swarm optimization and singleneuron PID. Sensors 18(4) (2018) 21. Yadav, S.L., Ujjwal, R.L., Kumar, S., Kaiwartya, O., Kumar, M., Kashyap, P.K.: Traffic and energy aware optimization for congestion control in next generation wireless sensor networks. Hindawi (2021)
32
A. Revathi and S. G. Santhi
22. Paek, J., Govindan, R.: RCRT: rate-controlled reliable transportprotocol for wireless sensor networks. ACM Trans. Sens. Netw. 7(3) (2010) 23. Shaikh, F.K., Khelil, A., Ali, A., Suri, N.: TRCCIT: tunable reliability with congestion control forınformation transport in wireless sensor networks. IEEE Xplore (2010) 24. Wang, C., Sohraby, K., Lawrence, V., Li, B., Hu, Y.: Priority-based congestion control in wireless sensor networks. In: Proceedings of the IEEE International Conference on Sensor Networks. Ubiquitous, and Trustworthy Computing (SUTC’06) (2006) 25. Haoxiang, W., Smys, S.: Soft computing strategies for optimized route selection in wireless sensor network. J. Soft Comput. Paradigm (JSCP) 2(01), 1–12 (2020) 26. Jacob, I.J., Ebby Darney P.: Artificial bee colony optimization algorithm forenhancing routing in wireless networks. J. Artif. Intell. 3(1), 62–71 (2021)
Stroke Prediction System Using Machine Learning Algorithm Siddharth Purohit , Ankit Chahar , Anish Reddy Banda , Anson Antony , Abhishek Subhash Suryawanshi , and Chandan Vanwari
Abstract Stroke is one of the heart diseases and is very dangerous to very much people in the world as it is the world’s third-largest disease which is causing death to the people in this world. This is a very big health issue in the world and is addressed by the World Health Organization (WHO) and can be seen from the statistics that have been provided by the organization and is very much approved by the other organizations too. The symptoms that are shown by the patients who are suffering from the stroke disease or which are very much prone to the stroke disease and may suffer from a very chronic stroke disease show heart disease, declination of metabolism and problems in the artilleries. The main issue due to which this problem is occurring is due to the flow of the blood, which is not able to reach the part of the heart through which the cleaning of the blood takes place or in other words the carbon dioxide is taken out of the blood and the oxygen is given which purifies the blood and is sent to the other body parts. In this medical industry, there are many machine learning and deep learning methods that are incorporated by the research
Supported by organization x. S. Purohit University Institute of Technology, RGPV, Bhopal, India A. Chahar RV College of Engineering, Bengaluru, India A. R. Banda Mahatma Gandhi Institute of Technology, Gandipet, India A. Antony (B) JSPM’s Rajarshi Shahu College of Engineering, Pune, India e-mail: [email protected] A. S. Suryawanshi SCTR’S Pune Institute of Computer Technology College in Pune, Pune, India C. Vanwari Thapar University, Patiala, India © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 S. Shakya et al. (eds.), Mobile Computing and Sustainable Informatics, Lecture Notes on Data Engineering and Communications Technologies 126, https://doi.org/10.1007/978-981-19-2069-1_3
33
34
S. Purohit et al.
community and different novelties have been researched by the community. By the method proposed, we could mitigate the strokes occurring by approximately 96% of the items from the data received from the patient. Keywords Stroke prediction · Random forest classifier · Machine learning · Predictive modelling
1 Introduction The central and peripheral nervous systems are both affected by neurological diseases. Some of these illnesses are treatable, while others are not. Neurological disorders such as Alzheimer’s disease and Parkinson’s disease affect people mostly beyond the age of 60, making ageing a key role in the development of these diseases [2, 3]. Genetic abnormalities, infections and lifestyle choices are among the reasons, as are other health issues that may impact the brain. There are over 600 nervous system disorders, including stroke, brain tumours, epilepsy and many others. Around 15 million people live in the area [4]. Stroke is the second-greatest cause of mortality and adult impairment in the globe, as we know from the statistics provided by the different agencies around 450–550 people get affected by this dangerous disease per 0.1 million people, and taking the yearly statistics, they show that around 15 million cases of very intense strokes occur and this is not only the case there are many other things after which needs to be taken care of because the person experiencing stroke may be disabled for the remaining life or the years which is left to live which is a very worst case that needs to be looked into. Hence, an early sign or signal that the condition might worsen or something is not right with the patient and that patient needs medical attention needs to be addressed and is solved with this solution that is provided. The model or the case that is produced by the authors would be revolutionary and would be beneficial to many people.
2 Related Work Alzami et al. [5] proposed a system in his paper which is focussed on developing the selection of features technique for classification of the episodes that occur in epilepsy. The contribution that has been presented to the Bonn university present in Germany which is used to study the EEG dataset. Rank aggregation was used to merge the subgroups obtained. In addition, the aggregation subgroups were fed into a basic classifier to generate the learning model and prediction. Finally, forecasting the classification and detection tasks was done by voting, which can be used by us as an example for the detection task. Another research [6] of Alzheimer’s disease includes the study by Durongbhan and Zhao [6] which aimed to obtain biomarkers using quantitative analysis of electroencephalography through a framework consisting of data
Stroke Prediction System Using Machine Learning Algorithm
35
augmentation, feature extraction, K-nearest neighbour (KNN) classification, quantitative evaluation and topographic visualization. Twenty HC and 20 AD subjects participated in this research and data was collected from them. MATLAB had been used for research purposes. The proposed framework was able to accurately classify the records and found important features as biomarkers for proper diagnosis of disease progression. “Development of an Algorithm Fortrose Prediction: A National Health Insurance Database Study”—Min S. N., Park S. J., Kim D. J., Subramaniam M., Lee K.S.—In this research, this paper aimed to derive a model equation for developing a stroke pre-diagnosis algorithm with the potentially modifiable risk factors [7]. “Stroke prediction using artificial intelligence”—M. Sheetal Singh, Prakash Choudhary—In this paper, here, decision tree algorithm is used for feature selection process, principal component analysis algorithm is used for reducing the dimension and adopted backpropagation neural network classification algorithm, to construct a classification model [8]. Chen and Hengjinda [21] state that the survey of the neural networks which gives one idea about the difference in evolving neural networks.
3 Methodology 3.1 Data Preparation Data has been taken from healthcare dataset of strokes, which has played an important part in many researches. As the data is pretty tough to collect, the only action possible to perform the model to do the prediction was to take a data from a good source. So the dataset has been taken from healthcare dataset of strokes. Based on input factors including gender, age and numerous illnesses and smoking status, this dataset is used to predict whether a patient is likely to have a stroke. For machine learning and data visualization purposes, a subset of the original train data is selected using the filtering approach. The glimpse of the dataset can be seen in Fig. 1 with 11 clinical features in it. The dataset has approximately 5000 entries in it.
Fig. 1 Dataset glimpse
36
S. Purohit et al.
3.2 Data Preprocessing The most important thing that the model or the entire solution needs is the data and the dataset without which it is not possible to do anything of the kind and is very tough to deal with the model. To satisfy it one needs very good data to prove something very novel and which is very useful to the community and other people can also work on it irrespective of the model or different problem statements provided as they can take help of the features that are present in the dataset. The next step would be to scale the variation in the features, since feature scaling is a means to normalize the independent characteristics included in a data in a set range, in this instance BMI, age and avg glucose level is scaled. And finally, the last step would be to drop the id feature because the number of entries isn’t required and finding the null value as null creates a problem as it hinders with the accuracy of the model. After that, the dataset is divided into training and testing data, and in this case, the authors are using the 70–30 ratio which means 70% data for training and 30% for testing (Fig. 2).
3.3 Feature Selection and Visualization This universe is made up of a lot of different things, and the features or variables contained in the dataset are critical to the dataset’s success. Without them, it is
Fig. 2 Scalability features overview
Stroke Prediction System Using Machine Learning Algorithm
37
difficult to go anywhere, and it should be done with caution. So the authors must handle the duplicates that exist in the dataset, whether they are features or variables, and this is done by the authors. The best solution to a problem is one that needs the fewest assumptions, according to “Occam’s Razor’s Law of Parsimony”. The results that have been achieved are very important in the most valuable thing which is developing the ml models. In this case, the author has 11 features that would help to get a proper accuracy and help the model perform better. In Fig. 3, it is observed that most people who have had a stroke do not have any heart disease, but that does not prevent it being an influential factor. Being a smoker has increased the chances of strokes, which can be observed in Fig. 4. The work type relatability to stroke is very unusual, as it is observed that people working in private companies have a great probability of getting stroke (Fig. 5). Age plays an important role, more the age more is the chance of getting stroke. And finally BMI is an important factor for the stroke as stroke is nothing but a blockage in our heart, and this could occur due to obesity so the box plot shown in Fig. 6 tells that most people receiving strokes are in overweight to obesity category.
Fig. 3 Heart disease correlation
Fig. 4 Smoking feature correlation
38
S. Purohit et al.
Fig. 5 Work-type feature correlation
Fig. 6 BMI Box plot
3.4 Model Architecture The model that is being used by the authors is based on the ensemble technique that is used by the different researchers in the main motive to increase the accuracy of the model that the authors are working on. In this problem statement, the technique that is used is the bagging technique and the main aim of it is to provide a better decision model than the other existing models and the decision tree is used as the estimators in the current research. Hence, when the case of classification comes into the scene, the decision trees come to the play and they play an important role in the voting. Since the random forest combines multiple trees to predict the class of the dataset, it is possible that some decision trees may predict the correct output, while others may not. But together, all the trees predict the correct output. We can avoid parameter tweaking and decrease overfitting this way. The main reason to use
Stroke Prediction System Using Machine Learning Algorithm
39
Fig. 7 Accuracy and performance matrix
random forest model with the algorithm is because it can handle a large dataset very well and provide a good accuracy.
4 Experimental Results The process that has been followed by the authors which is ranging from the data collection process and cleaning and making the dataset appropriate to use for the model has made a good impact and with the use of the random forest classifier which helped boost the accuracy to a great extent. The authors have achieved an accuracy of 96% which is better than the other models out there. The main novelty of the output is that the model used by us has performed better than the other models; hence, the accuracy has been increased the authors were getting an accuracy of 93% for the svm model, 91 for the logistic regression model and 94 for the naive Bayes. After that, a confusion matrix is important as it helps in summarizing the performance of the algorithm. Along with the performance matrix, the entire accuracy score of the algorithm is also required to get how much accurate it is. The precision and recall are 0.95 and 1.00, respectively (Fig. 7).
5 Conclusion The decision support system for predicting strokes aids and assists physicians in making the best, most accurate and quickest decisions possible, as well as lowering the cost that is occurring to the patient by a huge amount and the risk by a very big amount. The suggested method significantly lowers treatment costs and enhances the quality of life. The use of the ensemble techniques which is incorporated by the authors has received an accuracy of 96.1% which is a massive growth in this medical field and to the other healthcare sector too. And in the views of science and technology, this is a very great and novel innovation to the research community.
40
S. Purohit et al.
Future work Furthermore, we can use this dataset to do a comparative approach to show how our algorithm used with the help of the random forest classifier tops all the other methods out. After this, the authors or the other researchers can use the current algorithm or the current model on the dataset developed by the different people or collected from various/multiple sources and to see how is it performing after the testing stage. This research may aid in the development of more effective and reliable illness prediction and diagnostic systems, which will aid in the development of a better healthcare system by decreasing overall costs, time and death rates.
References 1. Akash, K.S., Srikanth, H.S., Thejas, A.M.: Prediction of stroke using machine learning (2020) 2. What is Alzheimer’s disease? Symptoms & Causes |alz.org. Alzheimer’s Association 3. Parkinson’s disease-symptoms and causes. Mayo Clinic, 2018. Available: https://www. mayoclinic.org/diseases-conditions/parkinsonsdisease/symptoms-causes/syc-20376055 4. Stroke statistics. The Internet Stroke Center. Available: http://www.strokecenter.org/patients/ about-stroke/stroke-statistics/ 5. Alzami, F., Tang, J., et al.: Adaptive hybrid feature selection-based classifier ensemble for epileptic seizure classification. IEEE Access 6, 29132–29145 (2018) 6. Durongbhan, P., Zhao, Y., et al.: A dementia classification framework using frequency and timefrequency features based on EEG signals. IEEE Trans. Neural Syst. Rehab. Eng., pp. 1–10 (2018) 7. Min, S.N., Park, S.J., Kim, D.J., Subramaniyam, M., Lee, K.S.: Development of an algorithm for stroke prediction: a national health insurance database study 8. Sheetal Singh, M., Choudhary, P.: Stroke prediction using artificial intelligence 9. Zang, Y., Lu, H., Zhang, Y., Alghannam, E., Guo, Z., Li, L.: A straightness control system for motor shaft straightening with the stroke prediction algorithm. In: 2019 6th International Conference on Systems and Informatics (ICSAI), 2019, pp. 57–62. https://doi.org/10.1109/ ICSAI48974.2019.9010553 10. Kobrinskii, B.A., Donitova, V.V.: Building a knowledge base of an expert system for personalized stroke risk prognosis. In: IEEE Conference of Russian Young Researchers in Electrical and Electronic Engineering (ElConRus) 2021, pp. 2815–2817. https://doi.org/10.1109/ ElConRus51938.2021.9396306 11. Shah, V., Modi, S.: Comparative analysis of psychometric prediction system: smart technologies. Commun. Robot. (STCR), pp. 1–5 (2021). https://doi.org/10.1109/STCR51658.2021. 9588950 12. Park, S.J., Hussain, I., Hong, S., Kim, D., Park, H., Benjamin, H.C.M.: Real-time gait monitoring system for consumer stroke prediction service. In: IEEE International Conference on Consumer Electronics (ICCE) 2020, pp. 1–4 (2020). https://doi.org/10.1109/ICCE46568.2020. 9043098 13. Kansadub, T., Thammaboosadee, S., Kiattisin, S., Jalayondeja, C.: Stroke risk prediction model based on demographic data. In: 2015 8th Biomedical Engineering International Conference (BMEiCON), 2015, pp. 1–3. https://doi.org/10.1109/BMEiCON.2015.7399556 14. Indarto, E.U., Raharjo, S.: Mortality prediction using data mining classification techniques in patients with hemorrhagic stroke. In: 2020 8th International Conference on Cyber and IT Service Management (CITSM), 2020, pp. 1–5. https://doi.org/10.1109/CITSM50537.2020. 9268802.
Stroke Prediction System Using Machine Learning Algorithm
41
15. Modi, S., Bohara, M.H.: Facial emotion recognition using convolution neural network. In: 2021 5th International Conference on Intelligent Computing and Control Systems (ICICCS), 2021, pp. 1339–1344. https://doi.org/10.1109/ICICCS51141.2021.9432156 16. Lien, C.-H., Wu, F.-H., Chan, P.-C., Tseng, P.-C., Lin, H.-H., Chen, Y.-F.: Readmission prediction for patients with ischemic stroke after discharge. In: 2020 International Symposium on Computer, Consumer and Control (IS3C), 2020, pp. 45–48. https://doi.org/10.1109/ IS3C50286.2020.00019 17. Chen, J., Chen, Y., Li, J., Wang, J., Lin, Z., Nandi, A.K.: Stroke risk prediction with hybrid deep transfer learning framework. IEEE J. Biomed. Health Inf. https://doi.org/10.1109/JBHI. 2021.3088750 18. Cho, J., Hu, Z., Sartipi, M.: Post-stroke discharge disposition prediction using deep learning. In: SoutheastCon 2017, pp. 1–2 (2017). https://doi.org/10.1109/SECON.2017.7925299 19. Zamsa, E.: Medical software user interfaces, stroke MD application design. In: E-Health and Bioengineering Conference (EHB) 2015, pp. 1–4 (2015). https://doi.org/10.1109/EHB.2015. 7391403 20. Zheng, L., et al.: Risk prediction of stroke: a prospective statewide study on patients in Maine. In: IEEE International Conference on Bioinformatics and Biomedicine (BIBM) 2015, pp. 853– 855 (2015). https://doi.org/10.1109/BIBM.2015.7359796 21. Chen, J.I.Z., Hengjinda, P.: Early prediction of coronary artery disease (CAD) by machine learning method—a comparative study. J. Artif. Intell. 3(01), 17–33 (2021) 22. Haoxiang, W., Smys, S.: Big data analysis and perturbation using data mining algorithm. J. Soft Comput. Paradigm (JSCP) 3(01), 19–28 (2021)
A Comprehensive Survey on Multilingual Opinion Mining Aniket K. Shahade, K. H. Walse, and V. M. Thakare
Abstract In a current scenario use of multimedia, gadgets have increased the usage of social websites and the Internet. Twitter, Facebook, Instagram, Telegram, and WhatsApp are the generally used platforms in the Internet community. Sharing reviews, feedbacks, and personal experiences are the most common task on social media. Such data is available in an unorganized and immensurable manner on the Internet. Opinion Mining can be carried out on such data available on the Internet. Most of the analyzers are working on the analysis of Chinese and English language sentiments, data available on the Internet is also in different languages which needs to be analyzed. The main purpose of this paper is to discuss the different frameworks, algorithms, Opinion Mining processes, classification techniques, evaluation methods, and limitations faced by the analyzers while bringing off the sentiment analysis on different languages. Keywords Sentiment analysis · Multimedia · Data · Machine learning technique · Deep learning technique · Opinion mining
1 Introduction Opinion Mining is very popular for processing social media data like reviews, feedbacks, personal experiences, and discussions on the product. Sentiment Analysis also called Opinion Mining is the computing research branch that is generally used for the classification of text but sometimes audio as well as the video is also classified either as neutral, positive, or negative. Opinion Mining also aims to extract people’s opinions and people’s reactions that will aslo be analyzed and understood, views expressed by people on the Internet regarding products or issues. Nowadays, as the A. K. Shahade (B) · V. M. Thakare Department of CSE, SGBAU, Amravati, India e-mail: [email protected] K. H. Walse Department of CSE, AEC, Chikhali, India © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 S. Shakya et al. (eds.), Mobile Computing and Sustainable Informatics, Lecture Notes on Data Engineering and Communications Technologies 126, https://doi.org/10.1007/978-981-19-2069-1_4
43
44
A. K. Shahade et al.
use of the Internet has increased all the information is available easily with a click on the Internet about different technologies, books, restaurants, services, movies, etc. People generally expressed their views, opinions, reactions about services on the Internet. For example suppose a person has purchased a new camera, as the person started using a camera, he/she first mark’s feedback whether the camera has met expectation, which features of the camera are most liked, or the features disliked. Such feedback taken from the buyer is very much important for the industry. With the help of feedback from the buyer the manufacturer tries to improve the product that is if the feedback is negative then that should be upgraded and if the feedback is positive, then the thing must be remembered while developing a new product. Medagoda et al. [1] work on a language other than English, as the work on other languages is very less as most of the researchers are continuously working on English datasets. The work which is performed on languages other than English also used the same approach as used for the English language but this one has a limitation as every approach has some properties relevant to the specific languages. Therefore, it is essential to recognize and analysis of data that is available in different languages as it may give meaningful perceptions to industries. Consider the countries like India and China are diverse in languages as the countries have the world’s largest population so it will be useful to all the industries as they have maximum share of customers in India and China.
2 Opinion Mining of Code-Mix Languages In code mixing, two or more languages are mixed while communicating or sharing views on the Internet. Multilingual people generally used code-mixing. Code-mixing is one of the challenges to the researchers working in sentiment analysis. In India, most people used the mixing of Hindi and English while communicating or sharing views on the Internet. For example, suppose the person has shared feedback about the newly bought smartphone “Best Smartphone.. camera bahut accha hai”. In this phrase, Best smartphones are words written in the English language, and the remaining words are written in Hindi language, so the identification of language is the initial step here followed by sentiment analysis which increases the overload of the researcher and becomes the process time-consuming.
3 Opinion Mining Process The sentiment analysis process consists of six different steps as data extraction, annotation, pre-processing, feature extraction, modeling, evaluation. The Figure 1 shows the steps of the sentiment analysis process.
A Comprehensive Survey on Multilingual Opinion Mining
45
Fig. 1 Opinion mining
Fig. 2 Opinion mining process
3.1 Data Extraction Initially, the extraction of data is done in the opinion mining process. Data extraction can be done either manually or automatically. The data from the Internet can be extracted automatically with the help of different web scraping algorithms. Text pattern matching is the most used web scraping technique, which is used to extract targeted data that match the mentioned search criteria given in the algorithm. The data extraction process can be done by application programming interfaces in the process of extracting data from the Internet. Data extraction is a part of opinion mining process (Fig. 2).
3.2 Annotation Labeling of data is done after completing a data extraction process. In the Annotation process comments, issues, observations, notes, questions will be added to the file.
46
A. K. Shahade et al.
Labeling is an annotation process that is generally used in the classification of data as neutral, positive, or negative. The manual or automatic process will be followed while labeling the data.
3.3 Pre-processing Internet collected data is unprocessed and not well-structured data. In pre-processing stage raw and unstructured data is converted to knowable and structure form. Data cleaning as well as Data transformation and Data reductions are the most pre-processing important steps in pre-processing. Data Cleaning: The Data cleaning process is used to handle missed values in the dataset and noise available in the data. A manual process is used to fill the missed values or we may fill missed values by finding the mean of attributes by calculating the probability. Noisy data can be due to errors while entering data or errors during the collection of data. Clustering algorithms are used to handle noisy data. In clustering similar type of data is merged and forms a cluster and not required data generally goes outside the cluster. Data Transformation: Data transformation is the necessary step as in some cases mining process is not possible as the data is not available in a suitable form. Normalization of data, attribute derivation are some data transformation methods available. In normalization data values are scaled in the range 0 to 1 and −1 to 1. In attribute derivation data present in multiple attributes are extracted and newer attributes are created. For E.g. sex is a derived attribute from the salutation mentioned in front of the name. Data Reduction: A huge data is present on the Internet. To perform processing of such a huge data efforts are required and it is a very time-consuming process. There are some attributes in the data which can be removed as they are not much important. Attribute selection, numerosity reduction methods are used in the data reduction process. In the attribute selection task only main and appropriate attributes are selected from the set of data and the last remaining attributes were omitted. In numerosity reduction, sample data is stored despite complete data. The number of pre-processing methods used by the analyzer are like tokenization of data, removal of stop word, stemming, lemmatization, POS. The tokenization process will be more clear from Fig. 3.
Fig. 3 Tokenization
A Comprehensive Survey on Multilingual Opinion Mining
47
Frequently used words in the sentences are termed stop words, Removing stop words from the sentences will not affect the sentiment polarity. The most commonly used stop words in English sentences are “he”, “she”, “they”, “that”, “is”, “was” etc. In POS Tagging the words are going to be tagged supported by the parts of speech they are available in. Suppose, “He is clever” for the given term POS tagging method will split words in a sentence as ‘He’-pronoun, ‘is’-verb, ‘clever—adjective [2]. In the process of Stemming, suffixes and the prefix present in the word are removed and the word is reduced to the root form. Consider, in the word “tagging” after performing stemming it is “tag” as suffix “ing” is omitted from the word. The limitation of this process is that many times we lose the dictionary meaning of a word. The disadvantage of stemming is overcome by the process of Lemmatization. It first finds the root from the word and then suffix and prefix will be removed.
3.4 Data Vectorization Machine learning algorithms work on numeric data so the textual data must be converted to numeric data. Vectorization is the process to convert the data available in textual format into an equivalent numeric or vector format. There are various Vectorization methods like count vectorizer, TF-IDF, and a bag of words. Bag of Words (BOW) is a widely used method for vectorization. In this method list of predefined words that are BOW is available and a comparison between the word and sentence is done. If in the BOW list the word from the sentence is present, then 1 will be marked and if the word is not present then 0 will be marked. The following figure explains the Bag of Words in detail (Fig. 4). Term Frequency- Inverse Term Frequency (TF-IDF): In the document importance of a word can be calculated by TF-IDF. The most used feature extraction technique is TF-TDF. Term frequency is that the ratio of the number of times your word appears within the document to the entire words within the document. Inverse term frequency is used to find out the weightage of the word in whole documents [3]. Count Vectorization is the technique where a matrix of a document is maintained. In matrix documentation, every word is maintained with the frequency of occurrence of the word within the document. The figure shows the example of count vectorization (Fig. 5).
4 Classification Techniques Data classification can be done in different ways like a machine learning approach, deep learning approach, lexicon-based approach, or hybrid approach.
48
A. K. Shahade et al.
Fig. 4 Steps of bag of words
Fig. 5 Count vectorizer
4.1 Machine Learning Approach In machine learning approach different supervised, unsupervised, and semisupervised learning algorithms are applied on different datasets to perform the analysis and predictions. Related Work: Shi and Li [4] used an SVM algorithm for analysis and predictions. Analysis of English data of online hotel reviews by unigram feature was done. Features of the data like term frequency and TF-IDF are used to analyze the polarity of data as negative or positive. Data instances were used to separate data into a training dataset and testing dataset. Training datasets cover the target values. SVM is used as it gives
A Comprehensive Survey on Multilingual Opinion Mining
49
better results as compared with other classifiers [5], though Tong and Koller [6] used the Naïve Bayes algorithm and SVM algorithm as the most efficient classifier in machine learning approaches [7]. The reviews collected of hotels are 4000 reviews; pre-processing is performed on reviews and labeled as positive reviews or negative reviews. The developed classifier is used to classify reviews as positive or negative. The Term Frequency Inverse Term Frequency gives a better result as compared to simple [4]. Boiy and Moens [8] work on supervised machine learning techniques for identifying the sentiment in data. The model proposed by the authors applied to different datasets available on the Internet. The standard algorithms are used to extract the features. To work on a deeper level for analysis parsing of sentences is performed. Active learning is used to avoid annotation workload. After applying pre-processing different features extracted like negation, stems, unigrams, and discourse features. The machine learning algorithms like Naïve Bayes, SVM, Maximum entropy are applied. It is observed that SVM gives minimum error as compared with others for linearly separable data. The Naïve Bayes is the simple one as compared with others for classification. The Maximum Entropy classifier performs better in information extraction that gives a better result [9]. Corpora were collected from the Internet. It gives an accuracy of 83% which is high as compared with Naïve Bayes and SVM. Habernal et al. [10] worked on the Czech language. The supervised machine learning algorithm was proposed for the classification of sentiments of social media. In the study, three datasets were considered for analysis. The first dataset is of Czech Facebook Pages. Facebook data contain some bipolar, neutral, negative, and positive comments. Movie reviews were the second dataset considered in the study. The third dataset is reviews of products collected from online Czech shops. N-gram features have been extracted after data pre-processing. The binary features such as unigrams and bigrams were used. Evaluation of dataset is done by Maximum Entropy and SVM Classifier. The evaluation was done with F-measure. The measure of the proposed approach is 0.69. Prasad et al. [11] proposed an efficient algorithm for the Chinese dataset. POS tagger is used for tagging and parsing the text. After applying POS tagging selection of features was performed to find discriminative aspects. Lastly, the machine learning technique is used for classification. Information gain, chi-square feature, document frequency, and mutual information are included in feature selection. While training the corpus, the threshold document frequency of words and phrases is defined, the words having a frequency of document lower than the threshold or higher than the threshold were omitted. Mutual information was used for statistical language modeling. Information gain measures the quantity of data useful for the prediction of the category that’s contributed by the presence or absence of a given term within the document. The results of this approach show that information gain is that the best-suited feature which can be useful in future applications [11].
50
A. K. Shahade et al.
4.2 Deep Learning-Based Approach The Deep Learning approach based on deep neural networks is a sub-branch of machine learning techniques. Nowadays, deep learning algorithms are blooming in sentiment analysis. Related Work: Dang et al. [12] come up with the results of 32 different deep learning-based papers and analyzed Recurrent Neural Networks (RNN), Convolutional Neural Networks (CNN), and Deep Neural networks (DNN) of different eight datasets. In pre-processing phase word embedding and TF-IDF is used to prepare input for classification. It is observed that the Recurrent Neural Networks model gives good performance in terms of word embedding as compared with other algorithms. CNN algorithms have a low processing time as compared with other algorithms. Deep Learning Algorithms are the researcher’s choice in different 32 papers researchers used deep learning algorithms like Convolutional Neural Network, Simple Recurrent Networks, Recurrent Neural Network, Latent Rating Neural Network, Long-Short Term Memory, Gated Recurrent Units and Recurrent Neural Tensor Network. Yadav and Vishwakarma [13] presented a review of 130 papers that used deep learning algorithms. Authors observed that different deep learning techniques used for sentiment classification are: Capsule Network, Attention-based network, Deep Belief Networks, RNN (GRU and LSTM), Recursive Neural Networks, and Convolutional Neural Networks. It is observed that LSTM performs better as compared with other algorithms and deep learning techniques are much more efficient in sentiment classification. The only limitation is that it requires a large amount of data. Zhang et al. [14] presented applications of deep learning techniques used in the analysis of sentiments. The authors reviewed several papers which address different levels as aspect level, sentence-level, and document-level sentiment analysis. It has been observed that different algorithms were applied at different levels. In aspect level sentiment classification, algorithms used are Recurrent Attention Network, Attention-based LSTM, Adaptive RECURSIVE Neural Network, LSTM, and BiLSTM. In document-level sentiment classification, algorithms used are Memory Network, Artificial Neural Network, GRU-Based Encoder, Convolutional Neural Network, Stack Denoising Auto Encoder, LSTM, GRU. In sentence-level sentiment classification algorithms used are CNN, Recurrent Random Walk Network, LSTM, Bi-LSTM, Recursive Neural Tensor Network, Recursive Neural Networks.
4.3 Lexicon-Based Approach In the lexicon-based technique, the corpora or dictionary is used to perform an opinion mining task. In this lexicon-based approach, the words in the dictionary or corpora have polarity values given to each of them. The words present in the dataset are
A Comprehensive Survey on Multilingual Opinion Mining
51
looked at in the lexicon and if the word is matched in the lexicon then the polarity of the word is assigned. E.g. suppose the task is to find out vowels from the sentence. This can be done by keeping vowels as a dictionary and then searching the vowels from the sentence in it. Related Work: Singh et al. [15] works on the Cornell movie review dataset and apply an unsupervised technique with POS tagging. Firstly, the feature extraction was done on the whole dataset. Calculation of semantic orientation was done by the authors first then adjective extraction was performed and finally semantic orientation values were assigned. Aggregation was performed for semantic orientation; +1 is added for each positive term and −1 for each negative term. Therefore, the semantic orientation of each review is the total of positive and negative for all extracted terms. The threshold value is defined and used to classify the data as positive or negative based on obtained aggregation score. Cambria et al. [16] proposed approach is based on SentiWordNet. Firstly, the features are extracted and SentiWordNet was applied to calculate the score of the selected features. It provides scores between 0.0 and 0.1. Two datasets were used for the analysis purpose. The first dataset contains 1000 positive and 1000 negative reviews whereas another second dataset contains 700 positive and 700 negative reviews. The proposed approach can be easily employed in different languages. Multiword expressions and sarcasm can be easily detected by the approach. Some languages like Persian contains sarcasm and multiword expression largely. The only limitation of the approach is its time-consuming process. The performance also is not improved and it is less than machine learning techniques. Bhaskar et al. [17] proposed an unsupervised approach for classification. SentiWordNet is used for classification purposes. The method consists of many steps. Firstly, collect the online reviews from the Internet. Pre-processed collected reviews dataset. After pre-processing build a list of noun features and extract the noun phrases. Sentences are classified as subjective or objective sentences. After this opinion sentence detection was performed which calculates the semantic orientation of words. Finally, calculates the weight of the sentence and its polarity. This approach gives good accuracy. The dataset used for evaluation was the online reviews of Canon and Nikon cameras. The main limitation of this approach is the use of SentiWordNet. The result of the method shows that SentiWordNet is not efficient in classification and finding sentiment words.
4.4 Hybrid Approach Hybrid Approaches are trending nowadays. Many hybrid approaches are based on lexicon-based knowledge and machine learning techniques. The main hybrid approach is to combine both the techniques and find optimum results by extracting the effective features so that we can overcome the disadvantages of both approaches.
52
A. K. Shahade et al.
Currently, most of the researchers are focused on the combination of symbolic and sub-symbolic AI techniques for the analysis of sentiments [18]. Deep learning and Machine learning both are bottom-up approaches that used sub-symbolic Artificial techniques. These approaches are very much useful in handling a large amount of data and finding an area of interest from the data. It is observed that these approaches give a good result in image data but are not that effective in natural language processing applications. Therefore, the combination of symbolic and sub-symbolic approaches is used for effective communication as a pattern from a text can be recognized by using sub-symbolic AI and represent this in the form of knowledge by using symbolic AI. They concluded that the hybrid approach of symbolic and sub-symbolic AI gives a better performance in natural language processing tasks [18]. Related Work: Minaee et al. [19] developed a hybrid approach that is based on LSTM and CNN algorithms and demonstrated that this hybrid approach gives a better result as compared with individual output. Mizumoto et al. [20] developed an unsupervised machine learning approach to find the polarity and applied it to stock market data. Polarity dictionary was used to recognize the polarity of stock market news data. In the proposed approach for small words, polarity is found manually. The polarity of a new or long sentence is determined automatically. For unlabeled news, the dictionary method is used. This dictionary contains small words with polarities as positive or negative. If some sentence has positive as well as negative words then polarity is the co-occurrence of the frequency of positive and negative words. The co-occurrence bias is measured. Many words have a positive and negative polarity. The co-occurrence rate of positive and negative polarity is determined. Two different thresholds are used as thresholdP and thresholdN. The polarity of positive words is added in thresholdP and the polarity of the negative word is added in thresholdN. This threshold value may be between 0.5 and 1. Those words that have an occurrence frequency of less than 10 were omitted as those words are not reliable. A stock market news dataset consisting of 62,478 news reviews is used for the evaluation of the built model. A semi-supervised technique is used to build a polarity dictionary. Bootstrapping approach is used here* which differs from supervised and unsupervised approaches. This bootstrapping is based on simple calculations in statistics. As this method contains a small amount of labeled and unlabeled data semi-supervised technique is used.
5 Evaluation Parameters Evaluation of the model is done with various performance measure metrics. The widely used evaluation metrics are accuracy, precision, recall, and F1-score. • Accuracy: Accuracy is measured by taking the number of corrected predictions done by the system over the total number of data instances [3].
A Comprehensive Survey on Multilingual Opinion Mining
Accuracy =
TP + TN TP + TN + FP + FN
53
(1)
• Precision: Precision is the ratio of the number of truly positive predicted results to the total positive predicted results [3]. Precision =
TP TP + FP
(2)
• Recall: Recall is the ratio of the number of truly predicted results to the total number of actual positive results [3]. Recall =
TP TP + FN
(3)
• F1 Score: The weighted average of precision and recall is termed as F1 score [3]. F1 - Score =
2*(Recall × Precision) (Recall + Precision)
(4)
where, TP TN FP FN
Truly positives predicted, Truly negatives predicted, Falsely positives predicted, Falsely negatives predicted.
6 Conclusion The motive of the review paper is to get knowledge about trending techniques, algorithms in multilingual opinion mining. Reviewed several papers and observed that currently, the researchers are working on machine learning, deep learning, hybrid as well as advanced deep learning models, only a few analyzers work on lexiconbased approaches. Support Vector Machine (SVM) and Logistic Regression (LR) was the perfect machine learning algorithms. Convolutional Neural Network (CNN) is the best deep learning algorithm and BERT is the researcher’s choice nowadays in advanced deep learning algorithms. There is a lot of work on English and Chinese language and only a few researchers are working on other languages. Most of the work is already done on the sentence level. There is a need to work on the document and aspect-level sentiment analysis. The future work will be hybridization and advanced deep learning algorithms.
54
A. K. Shahade et al.
References 1. Medagoda, N., Shanmuganathan, S., Whalley, J.: A comparative analysis of opinion mining and sentiment classification in non-english languages, pp. 144–148. IEEE (2013) 2. Kaushik, A., Naithani, S.: A study on sentiment analysis: methods and tools. Int. J. Sci. Res. 4, 287–291 (2015) 3. Hinrich, S., Christopher, D.M., Prabhakar R.: Introduction to information retrieval. In: Proceedings of the International Communication of Association for Computing Machinery Conference, p. 260 (2008) 4. Shi, H.X., Li, X.J.: A sentiment analysis model for hotel reviews based on supervised learning. In: 2011 International Conference on Machine Learning and Cybernetics (ICMLC), pp. 950–54. IEEE (2011) 5. Pang, B., Lee, L., Vaithyanathan, S.: Thumbs up? sentiment classification using machine learning techniques. In: Proceedings of the ACL-02 Conference on Empirical Methods in Natural Language Processing, vol. 10, pp. 79–86. Association for Computational Linguistics (2002) 6. Tong, S., Koller, D.: Support vector machine active learning with applications to text classification. J. Mach. Learn. Res.. 2, 45–66 (2002) 7. Xia, Y., Cambria, E., Hussain, A., Zhao, H.: Word polarity disambiguation using Bayesian model and opinion-level features. Cogn. Comput. 7(3), 369–380 (2015) 8. Boiy, E., Moens, M.-F.: A machine learning approach to sentiment analysis in multilingual Web texts. Inf. Retr. 12, 526–558 (2009) 9. Berger, A.L., Pietra, V.J.D., Pietra, S.A.D.: A maximum entropy approach to natural language processing. Comput. Linguist. 22, 39–71 (1996) 10. Habernal, I., Ptacek, T., Steinberger, J.: Sentiment analysis in Czech social media using supervised machine learning, 65–74 (2013) 11. Do, H.H., Prasad, P.W.C., Maag, A., Alsadoon, A.: Deep learning for aspect-based sentiment analysis: a comparative review. Expert Syst. Appl. 118, 272–299 (2019) 12. Dang, N.C., Moreno-García, M.N., De la Prieta, F.: Sentiment analysis based on deep learning: a comparative study. Electronics 9(3), 483 (2020) 13. Yadav, A., Vishwakarma, D.K.: Sentiment analysis using deep learning architectures: a review. Artif. Intell. Rev. 53(6), 4335–4385 (2020) 14. Zhang, L., Wang, S., Liu, B.: Deep learning for sentiment analysis: a survey. Wiley Interdiscip. Rev. Data Min. Knowl. Discov. 8(4), e1253 (2018) 15. Singh, V.K., Piryani, R., Uddin, A., Waila, P., et al.: Sentiment analysis of textual reviews; evaluating machine learning, unsupervised and SentiWordNet approaches. In: 2013 5th international conference on knowledge and smart technology (KST), p. 122–27. IEEE (2013) 16. Cambria, E., Speer, R., Havasi, C., Hussain, A.: SenticNet: a publicly available semantic resource for opinion mining. In: AAAI Fall Symposium: Commonsense Knowledge, p. 02 (2010) 17. Bhaskar, J., Sruthi, K., Nedungadi, P.: Enhanced sentiment analysis of informal textual communication in social media by considering objective words and intensifiers. In: Recent Advances and Innovations in Engineering (ICRAIE), p. 1–6. IEEE (2014) 18. Cambria, E, Li Y, Xing FZ, Poria S, Kwok K (2020) SenticNet 6: ensemble application of symbolic and subsymbolic AI for sentiment analysis. In: Proceedings of the 29th ACM International conference on ınformation & knowledge management. pp 105–114.
A Comprehensive Survey on Multilingual Opinion Mining
55
19. Minaee, S., Kalchbrenner, N., Cambria, E., Nikzad, N., Chenaghlu, M., Gao, J.: Deep learning based text classification: a comprehensive review 1(1):1–43 (2020). http://arxiv.org/abs/arXiv: 2004.03705 20. Mizumoto, K., Yanagimoto, H., Yoshioka, M.: Sentiment analysis of stock market news with semi-supervised learning. In: 2012 IEEE/ACIS 11th International Conference on Computer and Information Science (ICIS), p. 325–28. IEEE (2012)
Development Features and Principles of Blockchain Technologies and Real Options as the Main Components of the Digital Economy Radostin Vazov, Gennady Shvachych, Boris Moroz, Leonid Kabak, Vladyslava Kozenkova, Tetiana Karpova, and Volodymyr Busygin Abstract The paper shows that the digital economy reveals a huge range of opportunities for various enterprises. It noted its strengths: costs reduction, increasing level of transactions’ security and transparency, and close focus on various sectors of the economy. In this regard, for a clear and definite understanding of the problems under consideration, the authors introduced the definition of the digital economy, digital technologies in the economy, and “end-to-end” digital technologies in the economy. The authors’ proposed approach allowed concluding that the digital economy term is distinguished by several subtleties associated with insufficient knowledge, understanding of technical implementation, and flexibility. The research aims at revealing the development features and principles of the main components of the digital economy: distributed ledger technology (blockchain) and real option technologies. The paper shows that blockchain technology, as a decentralized data ledger, is the most discussed and relevant topic in the development of the digital economy. The paper analyzed its strengths, such as cost reduction, increased security, and transaction transparency affecting various sectors of the economy. The conducted research reveals the essence of the main provisions of tactics and strategies when solving the problem of real options pricing. At the same time, paper presented a new classification of options contracts allowing determining the ways of their application and development. Whereas, the analysis of the problem of options contracts pricing demonstrated the relevance of new mathematical methods developed for their reliable and accurate evaluation. The paper demonstrates that, at present, interest in the concept and technique of real options application has significantly increased as R. Vazov · V. Busygin (B) VUZF University (Higher School of Insurance and Finance), Sofia, Bulgaria e-mail: [email protected] G. Shvachych · T. Karpova Ukrainian State University of Science and Technology, Dnipro, Ukraine B. Moroz · L. Kabak Dnipro University of Technology, Dnipro, Ukraine V. Kozenkova State Agrarian and Economic University, Dnipro, Ukraine © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 S. Shakya et al. (eds.), Mobile Computing and Sustainable Informatics, Lecture Notes on Data Engineering and Communications Technologies 126, https://doi.org/10.1007/978-981-19-2069-1_5
57
58
R. Vazov et al.
they draw attention as a potentially essential tool for evaluation and improving an enterprise development strategy. Keywords Digital economy · Real options · Blockchain · Transaction transparency · Costs · Investments
1 Introduction Until recently, the world had been organized according to the centralization principles of management, resource allocation, money circulation, and supervisory and regulatory bodies. There are many inefficiency and fragility samples of the centralized government model. The point here is simply the ordinary centralization ineffectiveness for long-term stable development. Meanwhile, the centralization can be very effective in solving urgent, critical, and short-term problems. However, in the long run, it is ineffective, as it is not amendable to modernization processes. The stable and high economic growth of the USA is mostly connected with the decentralized system of government. Each state of the USA in terms of legislative and executive power differs little from the states of Europe, each of them has its own laws and, in fact, independent, but mutually integrated economies. At the same time, China, which is often mistakenly perceived as the top of centralization, is in an economic sense very similar to the USA, especially after the start of Deng Xiaoping’s reforms [1], which began precisely with the decentralization of economic management. There are many other examples in economics and history testifying to the centralization ineffectiveness and, conversely, the decentralization effectiveness. The widespread of modern information and communication technologies presupposes the economy transition to the digital area. That process is directly reflected in management methods both at the macro level and at the level of commercial entities particularly, the transition to a decentralized management system is underway. The spread of digital technologies for a long period defines the development of the economy and society and has more than once led to dramatic changes in people’s lives. The advent of the digital economy is one of the priority areas for most countries, including economic leaders. The advent of new generation of digital technologies, which, due to the scale and depth of their influence, were called “end-to-end”—artificial intelligence, robotics, the Internet, wireless technologies, etc. —have caused another rise in business and social models transformation recently. Their implementation can increase labor productivity in companies. Shortly, new digital technologies determine the international competitiveness of individual companies and entire countries that form the infrastructure and legal environment for digitalization. Note that in international practice, there is still no precise definition of the digital economy. In most sources, when describing the digital economy, the emphasis is on technologies and the changes about their use in the interaction among economic
Development Features and Principles of Blockchain …
59
agents. Here one can either mention specific types of technologies or certain forms of changes in economic processes. The digital economy definition is often substituted by listing the directions of its influence on the economy and social sphere. In this regard, the concept of the digital economy and digital technologies is shared. For clarity and concepts definiteness, there can be introduced authors’ understanding of the provisions. Definition 1. The digital economy is creating, distributing, and applying digital technologies and related products and services. Definition 2. Digital technologies in the economy are technologies for collecting, storing, processing, searching, transferring, and presenting data electronically. Definition 3. “End-to-end” digital technologies in the economy are those used to collect, store, process, search, transfer, and present data in digital form, which operation is based on software and hardware tools and systems that are in demand, creating new markets and changing business processes. Currently, the following components of the digital economy develop most intensively: 1.
2.
Distributed ledger technologies (blockchain) are the algorithms and protocols for decentralized storage and processing of transactions structured as a sequence of related blocks that cannot be changed subsequently. Option technologies are one of the most flexible and practical financial instruments on the world market, which is a kind of the equivalent of a contract that gives any buyer the right, but not the obligation, for purchasing or selling a specified asset at a clear cost or for a specified period.
Considering the above, the research highlights the analysis of the main mechanisms of contradictions, arrangement and implementation of blockchain technology, and revealing the principles of organizing real options in the digital economy.
2 Analysis of Recent Research and Publications In decentralization, an actively discussed hot topic is the possibility of applying blockchain at different managing levels. Blockchain technology is a continuous sequential chain of blocks with information built according to set rules. The blockchain application is based on the decentralized storage of the data chain. Herein, data on completed transactions are stored in a specific order and form an invariable sequence of related blocks. Consequently, the information in the block is replicated and copied to each node in the network. That algorithm provides the technology with resistance to data changes. Typically, the blockchain is managed by a peer-to-peer network. Once written, the data in any block cannot be changed without completely changing all subsequent blocks, which requires the majority consent of the network participants. M. Swan
60
R. Vazov et al.
in the “Blockchain: Scheme of a New Economy” book [2] identifies three types of blockchain: 1. 2.
3.
Blockchain 1.0 is a cryptocurrency including Bitcoin, Ethereum, Litecoin, etc. Blockchain 2.0 is smart contracts. It covers a wide class of financial applications that work with stocks, bonds, futures, mortgages, and many other financial assets. Blockchain 3.0 includes all other applications based on the technology, beyond financial sphere.
The fundamental feature of blockchain technology implies the processing of transactions without intermediaries [3]. Transactions are distributed by nodes that are linked to each other via hash numbers. The miners compute those hash numbers for the block and, based on consensus, the block is accepted into the blockchain network. These blocks contain a register of transactions or smart contracts. That blockchain evolution has changed the way the Internet is viewed as a source of some economic value [4]. Experts in the digital economy claim that by 2027, 10% of global GDP will have been stored in the blockchain [5]. Despite the serious focus on the technology, the blockchain still requires appropriate research and clarifications to solve, e.g., problems related to the transactions’ processing time [6], etc. Blockchain as a decentralized data ledger has a great potential. The blockchain allows simplifying the work of registration chambers, notaries, and medical institutions. Moreover, registration chambers will simply become unnecessary if blockchain technology develops. Likewise, there will be no need for registrars of various securities, which result in reducing the transaction costs. The second most important and relevant component of decentralization of management is option technologies. According to some authors [7, 8], the real options theory application to assess projects allows a greater number of key factors in the analysis than creating a discounted cash flow model. Thus, a disadvantage of the net present value method is eliminated—the lack of flexibility and the impossibility of a full analysis of the available scenarios in the implementation of most investment projects. Recently, some experts have paid great attention to the real options as a new toolkit that has come to the realm of real investments from financial markets. In the classical sense, an option is an instrument of a financial or commodity derivatives market, which is a right to buy (call option) or sell (put option) an underlying asset (usually a standard number of shares or an exchange commodity) at a specified price (strike price) for a certain fixed date in the future (European option) or at any time before a certain fixed date in the future (American option), if the option holder finds it profitable to do so; otherwise, the holder has the right not to carry out the transaction. Thus, in the general case, an option is a prepaid opportunity (but not an obligation) to take any action if conditions are favorable in the future. The situation, essentially similar to buying and exercising/not exercising an option, often occurs not only in financial markets but also in other areas of economic life, in particular in corporate finance and real investments. So, e.g., the company
Development Features and Principles of Blockchain …
61
shareholders attracting debt financing become the owners of the CALL option for the right to own the company, and they will abandon the company if its value is insufficient, transferring rights to creditors. An example is a land plot, which is a CALL option for its owner to develop that plot if the forecast market conditions are favorable enough and the project’s net present value reaches the desired level. A real option is the ability to make flexible decisions under conditions of uncertainty [9]. At first, there was little interest, mainly of an academic nature, but since the mid-90s of the twentieth century, interest in the concept and technique of real options application has increased significantly; as it was a potentially essential tool for assessing and developing strategies, first in the oil and gas sector, and then in other areas related to corporate investments. The first conference, devoted to theoretical and practical issues related to applying the theory of real options, was held in 1996 and has been held annually ever since in the USA and other countries [10]. Entire books devoted to this issue began to appear, with many academic papers. There was a transition from an average, exclusively academic interest to significant, active scientific and practical attention. Objectives Based on the literature review and the above analysis results of the current state of digital economy problems development, there was decided to analyze the main mechanisms for the blockchain technology arrangement, identifying the fundamental problems of its implementation, performing a systematic analysis of the contradictions of blockchain technology, and proposing ways to eliminate them; study features of the real options application in the digital economy, analyzing the basic requirements of digitalization when using real options, and identifying the main provisions of tactics and strategy in solving the problem of options’ pricing.
3 Statement of the Main Research Material 3.1 Main Mechanisms for the Blockchain Technology Implementation The main task for which blockchain technology is an appropriate solution is coordinating the actions of system participants united by a single goal but lacking trust in each other. Among cryptologists, it is known as a classic “task of the Byzantine generals,” with the following formulation: “The Byzantine army besieges the city.” Generals need to develop a single strategy leading to victory, even if there are traitors among them with deliberately distorting information about the number of their troops and the time of the offensive. “ Blockchain solves that problem with the consensus mechanisms.
62
R. Vazov et al.
This technology has great potential for the systems, in which participants have no mutual trust since it provides reliable storage of personal data, making changes in them inaccessible for fraudulent purposes [11]. The most valuable link in blockchain technology is the algorithms for reaching consensus, as those provide it with reliability. There are three main mechanisms for reaching consensus. (a)
Proof-of-work is a system security protocol. Anyone wishing to write a block to a database must perform a certain hard-to-compute task based on the principle of a one-way function. The computation takes a long time, while the receiving party quickly checks the result. Before sending the message, some mark is added to the header, where validity can only be confirmed by brute force. Computations verification on the receiving side is fast—due to a single SHA-1 computation with a preprepared label.
At the moment, the proof-of-work algorithm is the most popular among other mechanisms for creating reliable systems. The matter is that the one who can withstand the “ Sibyl attacks,” which essentially means that the attacker creates many fake participants and thus tilts the consensus in its direction. Running such an attack complicates the proof-of-work algorithm since the defrauder must spend enormous computing power to complete it. Also, most blockchains charge a fee for participating in the consensus; hence, the “Sibyl attack” becomes a very expensive operation. Often, the proof-of-work algorithm is criticized due to its excessive energy consumption, but so far, this is the only means of resisting such interventions in the system. (b)
Proof-of-stake is an alternative protection protocol to proof-of-work that requires confirming the storage of a certain amount in the account as proof. With a higher probability, when forming the next block, the system chooses a miner with big funds on the account, while this choice probability does not depend on the processors’ power. In order to undermine the system’s reliability, one of the participants must collect over 50% of all system funds, which is very costly.
Proof-of-stake has more advantages over proof-of-work. The main issue is lower time costs (there is no need for lengthy computations), but this does not eliminate possible problems. There is also no evidence of effectiveness in protecting against risks arising in cryptocurrencies. Two significant advantages of this protocol are that an attack on a system is very expensive, and the participant who runs it will suffer significantly for violating the system’s stability. Arguments against are that the method motivates accumulating funds in separate accounts, which calls into question decentralization. In the formation of a few participants with the most concentrated funds can own conditions for the system functioning. (c)
Delegated-proof-of-stake is an improved version of the proof-of-stake protection protocol, where specificity is that blocks are generated by a predefined set
Development Features and Principles of Blockchain …
63
of system users (101 delegates) who are rewarded for the duty and punished for misconduct (such as double spending). The list of users eligible for block signing changes periodically according to certain rules, e.g., in Slasher, delegates are selected based on the stake and blockchain history. Delegates can receive votes from all users; the strength of the vote depends on the share of the voter’s currency. Delegated-proof-of-stake has the same advantages and disadvantages as those of proof-of-stake ones.
4 Blockchain Technology and Problems of its Implementation Blockchain is surely an attractive and most promising technology but not suitable for every system. Several prerequisites indicate blockchain implementation: (a) (b) (c) (d)
shared database; no trust between the participants; need for the absence of intermediaries; interdependence of operations; the need to create chains.
Nevertheless, it is worth noting that even for those systems where blockchain technology is applicable, its implementation has several obstacles caused by structure and technology principles [11]. Let us consider a few of them. (a)
(b)
(c)
Security and privacy issues. Despite the security solutions using sophisticated encryption algorithms, cybersecurity issues remain one of the most discussed. Any software is written by a person and therefore imperfect. The more it gets complicated, the faster the number of vulnerabilities grows. In addition, the integrity of the software and the network is vital for the blockchain transformation into an infrastructure technology. If the blockchain gets intertwined with all the world’s major financial systems, the powerful attacks can lead to disastrous consequences. Implementation and integration issues. When an organization adopts technology to modernize its business processes, it challenges migrating its old data to a new format. Here, the blockchain implementation is no simpler than other similar tasks, which means that the issue of planning the transition from current systems to blockchain still stands. The cost savings that blockchain implementation promises are encouraging, but implementation requires high upfront costs that cannot be ignored. Understanding technology. One of the biggest operational risks is that relatively few people understand how it works. If it is planned to introduce blockchain into a system whose users are wide sections of people, it can lead to unpleasant consequences. The thing is that the blockchain does not protect against the most popular type of fraud—phishing, which essence is to steal confidential user’s data. Key compromise can result in the permanent loss of cryptographically protected funds. Unfortunately, today not every ordinary user can boast of
64
(d)
R. Vazov et al.
knowing the basic rules for protecting personal data. A possible solution to identity theft is to associate public keys with an individual or a legal entity, but this mechanism requires additional costs. The operations speed issue. In order to protect against a 51% attack, the block size (e.g., Bitcoin) remains no more than 1 megabyte, which allows maintaining decentralization. However, it significantly limits the transaction speed—3.3 per second, while the Visa conducts 22 thousand per second. Expanding the throughput to at least ten transactions per second requires an increase in the block size to 1.6 gigabytes, which, firstly, causes problems for low-power miners and, secondly, complicates block distribution across nodes. Today, the Bitcoin blockchain “takes” about 38 GB [12]. Suppose subsequently blockchain systems appear that store information about transactions and other, more voluminous data. In that case, they are likely to fail since forcing miners to store other people’s data for free; the developer deprives them of the incentive to maintain the network, i.e., miners’ costs will exceed revenues.
5 System Analysis of Blockchain Technology Contradictions Blockchain, like any new innovative technology, has many problems to be solved for full-scale implementation. One of the most important is the decentralized data storage problem: Fig. 1 depicts a feature of the problem of decentralized data storage with the main contradictions for the technology under consideration wherein complete data register is stored in each node of the network; it allows restoring the network until the last node of its network is destroyed. However, it should be borne in mind that during the operation, the network is constantly growing, which leads to uncontrolled amounts of data. In addition, to enter the network of a new member, one needs to sync a large amount of data [12]. As an alternative to solve the problem, there could be a standard database that stores exact data in encrypted form and enters only their hash into the blockchain. The obsolete blocks are then archived. However, it is worth mentioning that this is some local solution to the problem. Nevertheless, the following circumstance must be kept in mind. When using blockchain technology to store data, it is important to remember that modern technologies do not allow storing large amounts of information on the blockchain. Thus, in essence, blockchain technology in this industry here is practically used as an intermediary and a ledger that enforces the terms of a transaction to provide storage from one person to another (see Fig. 1). The circumstance means that neither blockchain technology, nor smart contracts, nor cryptography protect information in a decentralized storage [13]. Moreover, it can be argued that information has the same protection in such circumstances as in traditional repositories.
Development Features and Principles of Blockchain …
65
Fig. 1 Contradiction scheme for decentralized data storage in the blockchain system (Source Authors’ elaboration)
In connection with the above, it can be assumed that one can apply methods from such technologies as “big data” [14]. The approach was developed in various architectures such as Map Reduce, Shared Memory, Shared Nothing, Shared Disk, etc. It is also focused on working with large amounts of data storage and processing. The main obstacle to integrating big data tools into blockchain is the type of system: blockchain is a decentralized distributed system, which means that computations are distributed among multiple nodes, and there are no nodes that control operations of other nodes. However, another approach can be considered here. It should be borne in mind that, practically, the blockchain is a simple database with significant scalability and no-query languages. However, decentralization, immutability, transparency, and universal data exchange more than compensate for its shortcomings. Referring to the above, Bigchain DB and IPDB technologies are currently underway, which are becoming global databases with decentralized management. Another critical task is to ensure trust in the system; as the system must be anonymous and transparent for its participants. Figure 2 shows a diagram of trust contradictions in the system. However, it should be noted that the concept of a distributed ledger is an attempt to create universal tools for solving the problem of trust in
66
R. Vazov et al.
Fig. 2 Scheme of trust contradictions in the blockchain system (Source Authors’ elaboration)
the remote implementation of business relations using information and telecommunication systems. The idea of a distributed ledger is embodied in several open, proprietary, and hybrid software platforms, most of which are generic, but some are specialized. The platforms enable the applications’ development for many areas of business relations. It is necessary to consider the following basic aspect. Users want to see the data movement on the web, without knowing what they are doing. For this reason, it was decided to apply asymmetric encryption algorithms—thus, each user has a pair of keys: private and public one. In this case, Fig. 3 demonstrates the relationship of users’ data. The private key is used to sign the blocks that are sent by the user. The user’s network address is displayed using the public key. Here are some of the main
Fig. 3 Interrelation of user keys in the blockchain system (Source Authors’ elaboration)
Development Features and Principles of Blockchain …
67
features of using custom keys. Before a user can conduct any transaction, one needs keys: a public and a private key. It is a string of characters—in the case of a public key, there can be from 26 to 35. Public and private keys are related to each other (Fig. 3)—and the user needs them both to send and receive data via the network. Interestingly, the network always knows that the user’s public and private keys are related, without even seeing the private key (Fig. 3). Thus, a custom public key is communicated to senders and recipients. It can be passed on to anyone. The private key, however, is a key that is extremely important to keep it completely safe. It is linked to the user’s public key by a cryptographic cipher and acts as a digital signature to authorize the transaction. In addition, access to the information sent to the user requires both keys (Fig. 3). Figuratively, the relationship between public and private keys can be formulated as follows. There is some box, and only one branch can open the public key. Someone puts money in this box and closes it. After closing that compartment, money is transferred to an adjacent compartment, which can only be opened with a closed key. Moreover, while someone has only one private key, others will never reach the box. If it is lost, then the money will forever remain in the box; however, some cryptocurrancy wallets provide a private key backup feature. One of the blockchain’s main problems is data reliability, which facilitates practical encryption algorithms [15]. Here the unique algorithms for concurrent access and conflict resolution in the network are considered. Those must guarantee sufficient cryptographic strength of information on the network and allow the implementation of the digital signature [16].
6 Analysis of Real Options as the Most Flexible Financial Tool of Digital Economy Options are one of the most flexible and practical financial tools in the global market. Experienced traders often use them in trading strategies. So, an option is a contract equivalent that gives any buyer the right, but not the obligation for purchasing or selling a specified asset at a clear cost or for a specified period. In this regard, options are a replacement for a standard contract, where the main subject of bargaining is the very possibility of its preferential exercise. Along with bonds, those can serve as an analog of securities, and it is now a legally binding agreement with strict terms and conditions. The decision to trade options contracts is one of the most sought-after methods of financial activity. That derivative financial tool has a fixed profit and loss, making it the most efficient option against the backdrop of alternative online trading methods. Note that over the past five years, investors have often used binary options in practice. The average income from such contracts is over 80%. Adjacent, one can conclude deals with a minute expiration, which is also beneficial.
68
R. Vazov et al.
7 Classification of Option Contracts There are several classifications of options contracts. The most advanced is based on the period for executing the basic requirements. According to that classification, options are divided into the following two types: European and American one. Figure 4 depicts such options classification in more detail. Analysis of Fig. 4 shows that if the option is trading according to the European scheme, a derivative provides the right to clear action with an asset at a specified value for a set period, and any attempt to execute the contract before the date leads to penalties. The American option has less stringent conditions, and it is permitted to be executed by the holder before the expiration period. Hence, the repayment is often made the entire time before the set day. Note that both schemes are still in use, so one can argue that those are equally in demand. However, their area application is radically different. So, during the conclusion of exchanges, the American style is often used: it opens up more opportunities for trading participants and does not interfere with implementing investment strategies planned for the European option. A strict standard is observed only for the premium volume. Moreover, the exchange adjusts parameters such as price and execution period after trading and subsequent clearing. However, the European format also became widespread when concluding transactions outside the exchange, i.e., directly between investors. It justifies itself by tough conditions, which are negotiated in advance for the sake of the interest of both parties, and are not subject to daily revision, like contracts from the exchange. Meanwhile, there is no need for constant reassessment of contract and further trading in it. So, when it comes to American contracts, the expiration dates are determined arbitrarily within the chosen strategy with the necessity to pay extra, and brokers set high commissions.
Fig. 4 Classification of options by the length of the exercise period (Source Authors’ elaboration)
Development Features and Principles of Blockchain …
69
Fig. 5 Classification of options according to the pricing method (Source Authors’ elaboration)
American style is allowed to close any day prior to expiration. Furthermore, the European one can be redeemed only on a specified date (expiration date, settlement, and maturity). According to the option pricing method, those can be classified as Asian and exotic. Figure 5 illustrates such a classification of options in more detail. An Asian option means an option that is executed at the weighted average cost for the entire duration of the option throughout the entire time from the date of purchase. That is also an independent derivative, where the final execution price is determined based on the average value of the investment asset over a fixed time. In the economic environment, those contracts are also called average price options. They are most often used in markets with the high volatility of investment assets that have to be traded [17, 18]. That mainly applies to commodities such as oil, exchange rates, and stock indices. A distinctive feature of such a scheme is that the strike or the expiration option price at the time of its conclusion is unknown—sometimes, only the method of its determination is indicated. In most cases, traders who trade futures contracts deal with two options: European (cannot be exercised until expiration) or American (with such an option). Asian options are much less common than the two varieties mentioned above. However, this does not prevent them from stable demand in Asia and worldwide. Let us note the main advantages of Asian options: • less risk as the trader can accept a more rational offer; • as a rule, Asian options are cheaper than European and American ones. Exotic options take a special place in the options market. Unfortunately, the exotic option still has no strict definition. There are different points of view explaining the term origin, and here are some of them: • exotic options are options that a one-factor model cannot evaluate; Exotic options have two main criteria: the complexity of payments and the rarity of signing. Exotic options that appeared in the 90s quickly developed and gained great popularity in Western markets as they surpassed classic options in hedging efficiency. Exotic options provide investors with the opportunity to receive guaranteed returns
70
R. Vazov et al.
in volatile conditions and additional income at low-interest rates. In most cases, they are very flexible, relatively cheap compared to a combination of simple options.
8 Option Pricing Analysis Note that options are split into buy and sell options. CALL options enable the owner to purchase the asset in the future at a pre-agreed rate. Moreover, the PUT format gives the holder a chance to sell the asset on similar terms. There are also dual schemes: their distinguishing feature is that they are available simultaneously for both operations. The European option allowed importers and exporters to ensure their work, making payments in different currencies. Until the contract expiration, the investor can only passively observe what is happening without influencing the situation. Nevertheless, one can use a hedging strategy that automatically compensates for losses when the agreement suddenly becomes unprofitable. Figure 6 shows a detailed description of the options classification by the types.
Fig. 6 Classification of options by type of trade (Source Authors’ elaboration)
Development Features and Principles of Blockchain …
71
Fig. 7 Call option strategy
The CALL option provides the owner with a good opportunity to buy the financial asset in the future at a pre-agreed cost. To do this, a trader only needs to pay a fixed transaction fee. Figure 7 shows the CALL option strategy. So the investor automatically becomes the option holder. That is acquired when there is confidence in the growth of quotations of a financial asset. Such options are often used in the case of short market action. The transaction is similar to a long-term position in stocks: the CALL buyer hopes that the price increases strongly before the expiration date. Also, that scheme variant is called the BUY option. The PUT option is a similar tool. However, the main difference is that it is primarily focused on the sale of an asset. To make a profit, it is necessary to follow the price policy dynamics and, if necessary, use the hedging method in parallel. It turns out that in this way, the investor makes money by forecasting a decrease in the underlying asset value without errors. Figure 8 shows the PUT option strategy. Sometimes, even a drop of a single point is enough for dramatic changes. It is often advisable to purchase them when the securities owner wishes to protect the finances from a coming fall. PUT itself gives the option buyer the right to sell the asset at an indicated price within a set time. In its structure, this scheme is similar to Fig. 8 Put option strategy
72
R. Vazov et al.
a short position; in this way, the buyer hopes that the share price declines before the allocated period ends. Sometimes that option is called a PUT option. In rare circumstances, the most profitable solution would be to buy PUT and CALL options simultaneously. Such trading operations have related assets, but in fact, have a different focus. This strategy is also called a double play. However, it makes sense to use the scheme only if there is a well-pronounced cyclicality of the asset’s volatility in practice.
9 Conclusions This paper demonstrates that the digital economy reveals a huge scope of opportunities, ranging from the transfer of money to the transfer of music, from the approval of large government projects to innovations in land regulation, from transparent monitoring of the spending of public funds to the regulation of the salaries receipt. The application of this approach is multifaceted, and in general, it is difficult to predict where humanity will find the application of digital technology in the foreseeable future. It was noted that its strengths including the reducing costs, increasing the level of security, and transparency of transactions drew the attention of various economic sectors. In this regard, for a clear and definite understanding of the problems under consideration, the authors’ definition of the digital economy, digital technologies in the economy, and “end-to-end” digital technologies in the economy were introduced. The approach proposed by the authors allowed concluding that the digital economy is distinguished by several subtleties associated with insufficient knowledge, understanding of technical implementation, and flexibility. Therefore, it is too early to state a complete change in the current appearance of enterprises under its influence. However, the indisputable fact is that the digital approach can transform its internal structure. However, for the digital economy to become widespread in various fields, it is necessary to address legal legitimacy, regulation, technical viability, standardization, and widespread adoption. The paper highlights the most intensively developing main components of the digital economy: distributed ledger technologies (blockchain) and real option technologies. Research has shown that blockchain technology is one of the main directions for the digital economy development. The paper shows that blockchain technology, as a decentralized data ledger, is the most discussed and relevant topic in the development of the digital economy. Its strengths are analyzed, such as cost reduction, increased security, and transparency of the transaction applicable for various sectors of the economy. The main mechanisms of contradictions, arrangement, and implementation of blockchain technology are highlighted in detail. The author’s approach to eliminate the identified contradictions is presented.
Development Features and Principles of Blockchain …
73
Besides, the paper highlights the fundamental features of the option technologies development as one of the essential components of the digital economy. So, the options analysis is the most flexible and practical financial tool of the digital economy. In addition, the above study highlights the main provisions of tactics and strategies when solving the problem of options’ pricing. At the same time, a new authors’ classification of options contracts is presented, allowing determining the ways of their application, use, and development. At the same time, evaluating the price of option contracts showed the relevance of developing new mathematical methods for their reliable and accurate evaluation [19]. The studies in this paper show that one of the main problems of the studied technologies is in the features of the modeling process, both machine and mathematical. For instance, both for servicing and solving problems of security of those technologies, it is necessary to use not just powerful computing equipment, but highperformance ones. On the other hand, the problem of determining the price of real options can be solved only via up-to-date complex mathematical apparatus. The authors attribute those problems to further research. It should be emphasized that the computing tools have always remained the main factor in the progress development of information technology. The problem of increasing of the computing resources efficiency has long been beyond any doubt and is currently relevant. Moreover, it is known that parallel computing is the most promising approach to increasing the speed and productivity of computing facilities. At the same time, modern practice shows that distributed (parallel) computer modeling can be implemented by the entire spectrum of modern computing technology: supercomputers, cluster computing systems, local and global networks, etc. In addition, distributed modeling allows processing the problems, which solution requires significant processing time, integrate mathematical models to work on different (including geographically distant) computing systems.
References 1. Sullivan, M.J.: Book Reviews : David Shambaugh (Ed.), Deng Xiaoping: Portrait of a Chinese Statesman. Oxford: Oxford University Press 1995. 172 pp., with index. ISBN: 0–19–828933–2. (Originally published as a special issue of China Quarterly (No. 135, September 1993)). Chin. Inf. 10(3–4), 201–203 (1995). https://doi.org/10.1177/0920203x9501000325 2. Swan, M.: Blockchain: blueprint for a new economy. O’Reilly Media (2015) 3. Zavorotny, A.: Analysis of practice of blockchain technology application in financial management. Politechnical Student J. 27 (2018). https://doi.org/10.18698/2541-8009-2018-10-391 4. Box, S., West, J.K.: Economic and social benefits of internet openness. SSRN Electron. J. (2016). https://doi.org/10.2139/ssrn.2800227 5. Morton, H.: Blockchain technology: an emerging public policy issue. NCSL Nat. Conf. State Lagislatures LegisBrief 25(44) (2017). https://www.ncsl.org/documents/legisbriefs/2017/lb_ 2544.pdf 6. Carton, F., Adam, F., Brézillon, P.: Why real-time transaction processing fails to capture the context required for decision support. Supporting Real Time Decision-Making, pp. 221–236. Springer, Berlin (2010). https://doi.org/10.1007/978-1-4419-7406-8_11
74
R. Vazov et al.
7. Hengels, A.: Creating a practical model using real options to evaluate large-scale real estate development projects. Massachusetts Institute of Technology, Cambridge (2005). https://core. ac.uk/download/pdf/4398593.pdf 8. Samis, M.R., Laughton, D., Poulin, R.: Risk discounting: the fundamental difference between the real option and discounted cash flow project valuation methods. SSRN Electron. J. (2003). https://doi.org/10.2139/ssrn.413940 9. Hamdan, Y.B.: Faultless decision making for false information in online: a systematic approach. J. Soft Comput. Paradigm (JSCP) 2(04), 226–235 (2020) 10. Annual International Real Options Conference. (n.d.). Realoptions.Org. http://www.realoptio ns.org (2021). Accessed 18 Aug 2021 11. Allison, I.: Guardtime secures over a million Estonian healthcare records on the blockchain. International Business Times UK. https://www.ibtimes.co.uk/guardtime-secures-over-millionestonian-healthcare-records-blockchain-1547367 (2016) 12. ENISA.: Distributed ledger technology & cybersecurity—improving information security in the financial sector. ENISA (European Network and Information Security Agency). http://www. the-blockchain.com/docs/European%20Union%20Agency%20for%20Network%20and%20I nformation%20Security%20-%20Distributed%20Ledger%20Technology%20And%20Cybe rsecurity.pdf (2016) 13. ISO.: ISO/FDIS 23257 ISO/WD blockchain and distributed ledger technologies—Reference architecture: technical committee document TC307. https://www.iso.org/standard/75093. html?browse=tc (2017) 14. Handanga, S., Bernardino, J., Pedrosa, I.: Big data analytics on the supply chain management: a significant impact. In: 2021 16th Iberian Conference on Information Systems and Technologies (CISTI) (2021). https://doi.org/10.23919/cisti52073.2021.9476482 15. Cheng, R.: Ekiden: a platform for confidentiality-preserving, trustworthy. ArXiv.Org. https:// arxiv.org/abs/1804.05141 (2018) 16. Rivest, R.L., Shamir, A., Adleman, L.M.: A method for obtaining digital signatures and publickey cryptosystems. Massachusetts Institute of Technology, Laboratory for Computer Science (1977) 17. Pavlov, R., Pavlova, T., Lemberg, A., Levkovich, O., Kurinna, I.: Influence of non-monetary information signals of the USA on the Ukrainian stock market volatility. Investment Manage. Financ. Innov. 16(1), 319–333 (2019). http://dx.doi.org/https://doi.org/10.21511/imfi.16(1). 2019.25 18. Pavlov, R., Grynko, T., Pavlova, T., Levkovich, O, Pawliszczy, D.: Influence of monetary information signals of the USA on the Ukrainian stock market. Investment Manage. Financ. Innov. 17(4), 327–340 (2020). http://dx.doi.org/https://doi.org/10.21511/imfi.17(4).2020.28 19. Andi, H.K.: An accurate bitcoin price prediction using logistic regression with lstm machine learning model. J. Soft Comput. Paradigm 3(3), 205–217 (2021)
Technical Efficiency Analysis of China’s Telecommunication Infrastructure: A Copula-Based Meta-Stochastic Frontier Model Anuphak Saosaovaphak, Chukiat Chaiboonsri, and Fushuili Liu
Abstract This research aims to quantify the technical efficiency of economic growth based on export promotion in China’s various economic regions using infrastructure construction data from 2001 to 2019. The copula-based meta-stochastic frontier model (CMSFM) is used to process it. According to CMSFM’s findings, the average efficiency of overall economic areas is 0.1335. The Eastern region’s efficiency is 0.0190, the Central region’s efficiency is 0.0353, the Western region’s efficiency is 0.0893, and the Northeast region’s efficiency is 0.3906. According to the model’s findings, the Northeast region has the highest technical efficiency. Furthermore, the outcome of estimation using the CMSFM model indicates that China’s telecommunication infrastructure needs to be improved. Keywords China · Telecommunication infrastructure · Technical efficiency · Copula-based meta-stochastic frontier
1 Introduction Infrastructure improvements are seen as critical to achieving economic growth and improving people’s lives [6]. China has been the world’s fastest-growing country for several decades. And, investment-led growth is one of China’s defining characteristics. Meanwhile, China’s fast economic growth has been supported by significant improvements in physical infrastructure [14]. This research aims to assess the technical efficiency of economic development in China’s various economic areas based on infrastructure construction and provide useful information to the government in order to decrease regional disparities in economic development. According to data from China’s National Bureau of Statistics, China’s infrastructure investment in 2019, which included transportation, telecommunications, the Internet, water conservation, the environment, and public facilities, totaled 17,000 billion yuan, ranking first in the world (National Bureau of Statistics of China [NBSC]). However, in the early days A. Saosaovaphak · C. Chaiboonsri (B) · F. Liu Faculty of Economics, Chiang Mai University, Chiang Mai 51000, China e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 S. Shakya et al. (eds.), Mobile Computing and Sustainable Informatics, Lecture Notes on Data Engineering and Communications Technologies 126, https://doi.org/10.1007/978-981-19-2069-1_6
75
76
A. Saosaovaphak et al.
of the People’s Republic of China’s creation, China’s infrastructure development in the areas of transportation, post, and telecommunications was severely lacking. It was progressing slowly. China’s infrastructure capital stock grew at an annual pace of 7.1% on average between 1953 and 1978. China’s infrastructure capital stock grew at an annual pace of 6.9% on average between 1979 and 1989. China began to implement certain short-term infrastructure development initiatives during this time, utilizing limited money to enhance investment in critical infrastructure and replace aging infrastructure [16]. Between 1990 and 2008, China’s infrastructure capital stock increased by 16 times in 18 years, from 1.19 trillion yuan to 1.92 trillion yuan. On average, it grew at a rate of 16.7% every year. Better infrastructure has reduced manufacturing industry production and delivery costs, allowing Chinese products to enter new markets and compete more effectively. The importance of China’s infrastructure in encouraging economic growth has grown in the twenty-first century. ‘Build roads first if you want to grow rich.’ This is a well-known Chinese proverb emphasizing the need of transportation infrastructure development. Allow all people and all regions to get wealth through equity. Some areas develop more quickly than others, which aids in the progress of most areas. This is a quick way to speed up progress and establish shared wealth. China’s former Chairman, Deng Xiaoping, advocated it in the 1980s [12]. This policy has set the groundwork for various economic zones to develop at their own pace. China divided the country into four major economic regions based on economic growth across the country in the 1980s to scientifically reflect different regions’ social and economic development. The Eastern, Central, Western, and Northeast regions are currently known as these [11]. Figure 1 and Table 1 show that the eastern section of China’s mainland (excluding Macao, Hong Kong, and Taiwan) is divided into eight provinces and two municipalities, with a total area of 930,000 square kilometers, accounting for 9.7% of the country’s total area. The Central region is made up of six provinces that cover a combined land area of 1,028,000 square kilometers, or about 10.7% of the country’s total land area. The Western region covers 6,870,000 km2 , accounting for 71.4% of Four Economic Regions' Area (Unit: Square kilometer) 7,88,500 9,30,000 10,28,000
68,70,000
Eastern region
Central region
Western region
Northeast region
Fig. 1 Four economic regions’ area. Source Provincial statistical yearbooks of China 2019
Technical Efficiency Analysis of China’s Telecommunication Infrastructure …
77
Table 1 Provincial-level administrative region in China Economic regions
Provincial-level administrative region
Eastern region
Beijing, Tianjin, Hebei, Shandong, Jiangsu, Shanghai, Zhejiang, Fujian, Guangdong, Hainan, Macao, Hong Kong, and Taiwan
Central region
Shanxi, Henan, Anhui, Hubei, Hunan, and Jiangxi
Western region
Chongqing, Sichuan, Guizhou, Guangxi, Yunnan, Tibet, Shaanxi, Gansu, Qinghai, Ningxia, Xinjiang, and Inner Mongolia
Northeast region
Liaoning, Jilin, and Heilongjiang
Source National Bureau of Statistics of China 2020
the country. It is divided into twelve provinces, one municipality and five autonomous territories. The Northeastern region covers 788,500 km2 , accounting for 8.2% of the country’s total size (China’s Provincial Statistical Bureaus [13]). Figure 2 shows that from 1978 to 2019 China’s Eastern area was the most economically developed, with a GDP of 51 billion RMB in 2019, accounting for 42.1% of the total 2020 (CEIC). This is due in part to one of China’s growth goals, which states that the Eastern area should be a development leader, with the Central and Western regions providing support. Following the development of the Eastern area, the Central and Western regions should be supported. In 2019, the Central region’s GDP came in second with 22 billion RMB. It is situated in the middle of the Eastern and Western hemispheres. The Central region’s economic development is at a medium level. It is also in China’s second tier of economic growth. It is also well-known in China as a major population center, transportation hub, economic zone, and crucial market, and it plays an important part in China’s labor division. The Western region, which covers the most ground, is rich in natural resources, has a great market potential, and is strategically important. However, the Western region’s economic development trailed substantially behind the other three economic zones due to natural, topographical, historical, and other 60000.000
(in billion yuan)
50000.000 40000.000 30000.000 20000.000 10000.000 0.000
Axis Title Eastern region
Central region
Western region
Fig. 2 GDP in four economic regions (1978–2019). Source CEIC data
Northeast region
78
A. Saosaovaphak et al.
factors. Figure 2 shows that economic growth in the Western region of twelve provinces is very similar to that in the Central region of six provinces. (Source: CEIC 2020). The Northeast region was previously China’s historic industrial base and most economically developed region, and it played a critical role in the country’s economy. As the industrial core relocated to the east and center; however, the northeast region’s development slowed and fluctuated over the decades. (Source: CEIC 2020). Based on the information above, this research article tries to improve and develop the physical infrastructure plan, especially China’s telecommunication infrastructure, by measuring the technical efficiency and comparing each region in China. The research results from this article may provide helpful information for the government to improve and develop the physical infrastructure more efficiently in order to benefit from the allocation of the budget and time management.
2 The Conceptual Framework and Methodology 2.1 The Conceptual Framework The conceptual framework of this study proceeds to Fig. 3 to measure the technical efficiency of each region’s economy for China based on infrastructure investment,
Infrastructure of Transport - Fixed asset investment in highway industry
Infrastructure of Telecommunication - Length of optical cable
Infrastructure of
Infrastructure of
Electricity
Water
-Electricity production
-Population with access
Export
Technical Efficiency
Four Economic Regions of China
Fig. 3 Conceptual framework. Source Authors
Technical Efficiency Analysis of China’s Telecommunication Infrastructure …
79
Table 2 Displays the variables that were employed to measure the technical efficiency in the production function Key factors
Indicators
Abbr.
Unit
Data source
State of economics
Export
EX
USD mn
CEIC database
Infrastructure of transportation
Fixed asset investment in highway industry
FAIHI
RMB mn
CEIC database
Infrastructure of water system
Population with access to water
PAW
Person th
CEIC database
Infrastructure of Electricity production
Electricity production
EP
kWh bn
CEIC database
Infrastructure of Telecommunication
Length of optical cable
LOC
km th
CEIC database
Source Authors
especially telecommunication infrastructure investment. This conceptual framework still comprehensive the production function based on the Cobb–Douglas function1 [see Eq. (15)]. This conceptual framework still comes from the production function based on the Cobb–Douglas function [see Eq. (15)]. This function generally has two main types of variables in it. The first variable is the output variable (Y: Export), and the second variables are the input factors (X: Transport infrastructure, Telecommunication infrastructure, Electricity infrastructure, Water infrastructure), respectively. The analysis examined annual data from China’s infrastructure in four economic categories from 2001 to 2019. The eastern region’s data is obtained in great detail from ten provinces or municipalities. The data from the core region is supplemented by six provinces. The data for the Western and Northeast areas is also obtained from their subordinate provinces, resulting in a total of twelve and three provinces for each region. Railways, roads, and other forms of transportation infrastructure are examples. Water infrastructure includes dams, water pipes, and other structures, and it is reflected in this study by the population with access to water. Energy infrastructure is made up of a network of wires, towers, dams, and turbines, and it is measured in the study by the amount of electricity produced. Telephone wires, cables (including undersea cables), satellites, microwaves, and mobile technology are all part of the telecommunication infrastructure. Here, the length of an optical cable will be investigated. As a result, export, which is a statistical indicator of the status of the economy, is the dependent variable. Transport infrastructure proxy is defined as fixed asset investment in the highway industry, water infrastructure proxy is defined as population with access to water, electricity infrastructure proxy is defined as electricity production, and telecommunication infrastructure proxy is defined as optical cable length (see more details in Table 2).
1
Cobb, C. W.; Douglas, P. H. (1928). ‘A Theory of Production’ (PDF). American Economic Review. 18 (Supplement): 139–165. JSTOR 1811556. Retrieved 26 September 2016.
80
A. Saosaovaphak et al.
2.2 Copulas and Dependence The copulas model is very useful for measuring the random variables in accordance with the real dependency structure. The copulas model’s substantial property is still important in computing the tails of the real joint distribution dependency [1, 3, 5, 8, 10] because the estimation of the old version for stochastic frontier production, especially the computation of technical efficiency using the joint distribution among random error terms, is based on only assumptions [2] without simulation study. To deal with this problem, this research article attempts to compute the technical efficiency of China’s infrastructure for economic development, especially telecommunication infrastructure improvement, based on the copula-based meta-stochastic frontier model estimation. The marginal probability distribution of each variable is uniform in a copula, which is a multivariate cumulative distribution function. Copulas are used to express the interdependence of random variables. A copula connects a certain number of marginal distributions (one-dimensional distributions) to a multivariate joint distribution [7]. If the collective distribution H is given, the system function C (.,.) is denoted as: [see Eq. (1)] H = (X 1 , X 2 ) = H F1−1 (u 1 ), F2−1 (u 2 ) = Cα (F1 (x1 ), F2 (x2 )),
(1)
where u 1 = F1 (x1 ) and u 2 = F2 (x2 ), α is a parameter vector of the copula, generally referred to as the dependence parameter vector. F1 (…) and F2 (…) are uniform distributions, and bivariate copulas satisfy the features as below: 1. 2. 3.
C(u 1 , 0) = C(0, u 2 ); C(u 1 , 1) = u 1 and C(0, u 2 ) = u 2 ; If u, v, w, and z belong to [0, 1], such that u ≤ w and v ≤ z, then C(w, z) − C(w, v) − C(u, z) + C(u, v) ≥ 0.
(2)
Meanwhile, the Sklar theorem is necessary for connecting copula functions and distribution functions that rely on the three features indicated before [see Eq. (2)]. Any multivariate probability distribution function can be represented by a dependent structure and a marginal distribution, as demonstrated below: [see Eq. (3)] ∂ F(x1 , . . . xn ) ∂ x1 . . . ∂ xn ∂C(u 1 . . . u n ,) ∂ F xi = × ∂u 1 . . . ∂u n ∂ xi = c(u 1 . . . u n ,) × f i (xi )
f (x1 , . . . xn ) =
(3)
Therefore, a bivariate density function of x1 and x2 can be expressed by two parts of marginal densities and copula density: (see Eq. (4))
Technical Efficiency Analysis of China’s Telecommunication Infrastructure …
f (x1 , x2 ) = c(u 1 , u 2 ) × f 1 (x1 ) × f 2 (x2 )
81
(4)
Pearson’s linear correlation is the most widely used approach for determining dependency, although it lacks information for asymmetric distributions. As a result of Pearson’s linear correlation’s limitations, the USA began to adopt rank correlations such as Kendall’s tau and Spearman’s rho to measure nonlinear dependence. The following are the results of Kendall’s tau and Spearman’s rho calculations: [see Eq. (5)] τx1 x2 = 4
C(u 1 , u 2 ) − 1 = 4E[C(u 1 , u 2 ) − 1
[0,1]2
ρ(x1 x2 )
= 12
u 1 u 2 d[C(u 1 , u 2 )] − 3
[0,1]2
=
C(u 1 , u 2 )du 1 du 2 − 3 [0,1]2
= 12E[(U1 , U2 )]
(5)
Many copulas are capable of capturing dependency structures. Few copulas, on the other hand, can measure the entire range of dependent structures, from the lower to the upper boundaries. Some copulas, such as the Gumbel, Clayton, and Joe copulas, can still be used to measure negative or positive reliance [7, 15]. The Akaike information criterion (AIC) and the Bayesian information criterion (BIC) are two criteria for choosing a model in which all variables fit the model to the best of their ability [see Eqs. (6) and (7)]. For a large class of models fitted with maximum likelihood, the AIC is defined as follows: AIC = −2logL + 2 ∗ k
(6)
where L is the maximum likelihood function for the estimated model. The BIC value is well-defined as: BIC =
1 RSS + log(n)k σˆ 2 n
(7)
The lowest values of AIC and BIC are preferred, implying that this is the best model that was applied to compute in the copula-based meta-stochastic frontier model later on.
82
A. Saosaovaphak et al.
2.3 Copula-Based Meta-Stochastic Frontier Model (CMSFM) for Technical Efficiency Computational For technical efficiency, this research article uses the copula-based meta-stochastic frontier model (CMSFM) for the computation of China’s infrastructure efficiency. It can start from Eq. (8) to compute the technical efficiency for each of the regions in China, especially for the four regions covered by this model. The copula-based meta-stochastic frontier production function model is defined mathematically as: ∗ Yit∗ ≡ f xit , β ∗ = exit β , i = 1, 2, . . . , N
(8)
where β ∗ signifies the vector of parameters for the meta-stochastic frontier function:(see Eq. (9)) xit β ∗ ≥ xit βi ,
(9)
where βi is a vector of parameters estimated by a CMSFM (see Eq. (9)) and is defined as a deterministic parameter function whose value is not smaller than the deterministic component of the CMSF production function for different units and periods [15]. Yit = e−U × it
exit βi xit β ∗ +V it ∗ × e x β it e
(10)
The technical efficiency relative to CSFM is the first term on the right-hand side of Eq. (10), and it is examined as: (see Eq. (11)) TEit =
Yit x it e βi +Vit
= e−Uit
(11)
The technological gap ratio (TGR) for the observation of the sample region concerned (see Eq. (12) is the second term on the right-hand side of Eq. (10), which is designated as: TGRit =
exit βi exit β ∗
(12)
The ratio of the frontier production function’s output to the prospective output described by the CMSF function given the observed inputs is measured here. The ratio of the observed output to the sum of absolute deviation and squared deviation, which is the output in CMSFM, is given by Eq. (11). It could be tweaked to account for the random error, which is as follows: (see Eqs. (13) and (14)) TEit∗ =
Yit x it e β ∗ +Vit
(13)
Technical Efficiency Analysis of China’s Telecommunication Infrastructure …
83
Table 3 Mean values of variables Region
EX (USD mn) FAIHI (RMB mn) PAW (Person th) EP (kWh bn) LOC (km th)
Eastern
1,300,080
1,758,324
186,117
1566
192
Central
89,530
1,177,265
79,065
836
164
Western
86,101
2,440,956
76,908
1204
361
Northeast
49,856
216,830
42,585
245
83
Source CEIC 2020
TEit∗ = TEit × TGRit
(14)
Finally, the technical efficiency of CMSFM could be stated by converting Eqs. (10)–(14).
3 The Empirical Study and Simulation Study 3.1 Descriptive Statistics for All Variables in the Copula-Based Meta-Stochastic Frontier Model (CMSFM) Table 3 shows the average of various variables from 2001 to 2019. It is clear that the eastern region had the highest export yield, which is a statistical indicator of the economic situation here. Furthermore, the eastern region rated first in terms of population with access to water. The Western region made the biggest investments in the highway business during the last two decades and had the longest cable length. The Central region’s economic situation and infrastructure elements were all in the middle. The Northeast received the lowest mean values across all factors.
3.2 Copula-Based Stochastic Frontier Model (CSFM) For the estimation of the stochastic frontier model for the technical efficiency analysis of China’s infrastructure based on the copula-based meta-stochastic frontier approach, which follows the linear translog production model, and it can implement Eq. (15) for estimation of the Cobb–Douglas production function to find out the significant independent factor pushing China’s economy in each of the regions with more efficiency. ln(EX)it = lnβ0 + β1 ln(FAIHI)it + β2 ln(PAW)it + β3 ln(EP)it
84
A. Saosaovaphak et al.
Table 4 Results of AIC statistics and BIC statistics for model selection
Copulas model Normal copula T-copula
AIC
BIC
−5.93
−3.60
−15.90
−11.24
−5.93
−3.60
−11.48
−9.15
Gumbel copula
2.00
4.33
Joe copula
2.00
4.33
Clayton copula Frank copula
Source Computed The T-copula model is the best-fitted model
+ β4 ln(LOC )it + Vit − Uit
(15)
where i represents different individuals of four economic regions in China, t is the year from 2001 to 2019. EX, FAIHI, PAW, EP, and LOC refer to exports, fixed asset investment in highway industry, population with access to water, length of optical cable, respectively. And, there are two error terms, Vit being regarded as noise terms and Uit inefficiency terms. In the conventional stochastic frontier model, it is assumed that Vit and Uit are not correlated to each other, as is shown the same as an independent copula CIN = u 1 u 2 . To further understand the discrepancy, Normal (Gaussian) copula, t-copula, Clayton copula, Frank copula, Gumbel copula, and Joe copula are used to estimate parameters. These copulas are assumed that Vit and Uit are correlated with each other. The copula allows for the most efficient use of stochastic frontier models. The Akaike information criterion (AIC) and the Bayesian information criterion (BIC) are estimators used to assess the model’s quality. The AIC and BIC values with the lowest values are chosen for selecting the best model. It is just essential to determine the lowest AIC and BIC values. Table 4 clearly shows that t-copula has the lowest AIC and BIC values, with −15.90 and −11.24, respectively. As a result, the t-copula is the best-fitted copula model for the stochastic frontier model. However, the simulation study of the copulas model to invent Vit and Uit for the copula-based meta-stochastic frontier model to estimate the technical efficiency of China’s infrastructure, especially telecommunication infrastructure improvement, is already displayed the figures together with code of programming in both Appendices A.1 and A.2.
3.3 Copula-Based Meta-Stochastic Frontier Model (CMSFM) The copula-based meta-stochastic frontier model has very high competitiveness when compared with the old version or the classical version of the meta-stochastic frontier model [9]. Therefore, this research essay attempts to implement this model to an
Technical Efficiency Analysis of China’s Telecommunication Infrastructure …
85
analysis of China’s infrastructure improvement by technical efficacy calculation. Moreover, this model will still be able to go for computation of the technical efficiency of China’s telecommunication infrastructure to develop planning accurately in the future as well. The meta-stochastic frontier function from Fig. 4 is above the deterministic functions for the stochastic frontier models of the groups involved as a production function of specified functional form the MF (x; β ∗ ). For the meta-frontier, the meta-stochastic frontier model posits a different data generation mechanism than for the distinct group frontiers. In Table 5, the empirical estimation results of the copula-based meta-stochastic frontier model continues to show that both the fixed asset investment in highway industry (log (FAIHI) and the telecom infrastructure (log (LOC)) are not propelling output Metafrontier≡MF ;
∗
) 3 ( ))
Frontier 3 2 Frontier 2
( ))
1 Frontier 1
( ))
input
Fig. 4 Diagram of meta-frontier and individual frontiers
Table 5 Empirical estimation results for the copula-based meta-stochastic frontier model Independent variables
Estimate
Std. error
Z value
Pr(>|Z|)
Intercept
−72.55
1.002
−72.46