Innovations in Smart Cities Applications Edition 3: The Proceedings of the 4th International Conference on Smart City Applications (Lecture Notes in Intelligent Transportation and Infrastructure) 3030376281, 9783030376284
This book highlights original research and recent advances in various fields related to smart cities and their applicati
144 79 140MB
English Pages 1301 [1284] Year 2020
Preface
Committee
Honor Committee
Conference General Chairs
Conference Local Chairs
Conference Technical Program Committee Chairs
Workshop Co-chairs
Publications Co-chairs
Tutorials Co-chairs
Web Chair
Publicity and Social Media Co-chairs
Sponsorship and Exhibit Chairs
Technical Program Committee
Keynote Speakers
Smart City Strategic Planning: Trends and Challenges
Citizens’ Engagement on Smart Cities
Smart Building Sustainability Assessment
Smart Governance: Leadership, Priorities and Indicators for Sustainable Cities
Traffic Flow Theory and Its Applications
Smart Airport: A Major Component of a Smart City
Contents
Smart Citizenship
General Smart City Experts’ Perceptions of Citizen Participation: A Questionnaire Survey
Abstract
1 Introduction
2 Citizen Participation Theory
3 The Smart City Concept
4 Citizen Participation in Smart Cities
5 Survey About Citizen Participation in Smart Cities
5.1 Methodology
5.2 Survey Highlights
5.3 Summary
6 Conclusion
Acknowledgments
References
Citizen Sentiment Analysis in Social Media Moroccan Dialect as Case Study
Abstract
1 Introduction
2 Related Work
2.1 Smart City
2.2 Sentiment Analysis
3 Sentiment Analysis Methodology
3.1 Data Collection and Annotation
3.2 Data Pre-processing and Features Selection
3.3 Opinion Classification
4 Validation and Analysis of Results
5 Illustrative Scenario
6 Conclusions and Future Work
References
Social Networks Fake Profiles Detection Using Machine Learning Algorithms
Abstract
1 Introduction
2 State of the Art
3 Using Naive Bayes Classifiers
4 Using Decision Trees Classifiers
5 Using Random Forest Classifiers
6 Our Approach
6.1 Initial Features
6.2 Features Selection
7 Results and Discussion
7.1 Splitting the Dataset
7.2 Evaluation Metrics
7.3 Experimental Results
8 Conclusion and Future Work
References
Smart City Mobile Apps Exploratory Data Analysis: Mauritius Case
Abstract
1 Introduction
1.1 Rationale Behind the Citizen as a Major Stakeholder in Smart Cities
1.2 Chapter Description
2 Technology Acceptance Model
3 Methodology
3.1 Sample Size and Hypotheses
3.1.1 Perceived Usefulness (PU)
3.1.2 Perceived Ease of Use (PEOU)
3.1.3 User Satisfaction (US) and Perceived Enjoyment (PE)
3.1.4 Behavioral Intention to Use (BI)
3.1.5 Quality of Internet (QI)
3.1.6 Citizen Experience and Readiness
4 Exploratory Data Analysis
4.1 Cronbach Alpha
4.2 Correlation
4.3 Anova
5 Conclusion
Acknowledgment
References
Gamification of Civic Engagement in Smart Cities (New Russian Practices)
Abstract
1 Introduction
2 Literature Review
3 Solidarity and Cooperation Through Gamification
4 Interactivity and Immersiveness Through Gamification
5 Conclusion
References
Machine Learning for Sentiment Analysis: A Survey
Abstract
1 Introduction
2 Background
2.1 Main Concepts
2.2 The Preprocessing Task in Sentiment Analysis
3 Related Works Review
4 A Comparative Study
4.1 Criteria
4.2 Analysis and Discussion
4.3 Statistical Review
5 Conclusion and Future Work
References
A Survey on Hand Modalities and Hand Multibiometric Systems
Abstract
1 Introduction
2 Hand Biometric Systems
2.1 General Biometrics Concepts
2.2 Hand Recognition Systems
3 Hand Modalities
3.1 Geometry
3.2 Contour (or Shape)
3.3 Surface (or Texture)
3.4 Vein
4 Multibiometrics
4.1 Comparison Study of Multibiometric Systems
5 Conclusion
References
A Social Media Ontology-Based Sentiment Analysis and Community Detection Framework: Brexit Case Study
Abstract
1 Introduction
2 Related Work
3 Main Concepts
3.1 Sentiment Analysis
3.2 Community Detection
3.3 Ontologies
4 Proposed Framework
4.1 Social Media Extractor
4.2 Text Preprocessing
4.3 Sentiment Analysis
4.4 Community Detection
4.5 Analytics Console
5 Brexit Analysis Case Study
5.1 Sentiment Analysis – Overall Pie Charts
5.2 Sentiment Analysis – Results on Map
5.3 Sentiment Analysis – Overtime Stacked Bars
5.4 Community Detection Results
6 Conclusion
References
Fuzzy Questions for Relational Systems
Abstract
1 Introduction
2 Fuzzy Systems
2.1 The Operation of a Fuzzy System
3 Fuzzy Querying System Architectures
3.1 Fuzzy Query to Database
3.2 Fuzzy Interrogation System Architectures
4 Comparison of Most Relevant Characteristic in Fuzzy Query Systems
5 Conclusion
References
Smart Education
Using Machine Learning Algorithms to Predict the E-orientation Systems Acceptancy
Abstract
1 Introduction
2 School Guidance and TIC
3 Theoretical Background
3.1 Technology Acceptance Model
3.2 Using TAM in E-orientation Systems
4 Methodology
4.1 Waikato Environment for Knowledge Analysis (Weka)
4.2 Feature Selection
4.3 Classification Algorithms
4.4 Classifier Accuracy Measures
5 Results and Discussion
6 Conclusion
References
Smart University Services for Collaborative Learning
Abstract
1 Introduction
2 Research on Smart University (SU)
3 Smart University’s Components
3.1 Smart Learner
3.2 Smart Knowledge
3.3 Smart Learning
3.4 Smart Interaction
4 Smartness in University
4.1 Introduction
4.2 Smart University’s Services
4.3 Service-Oriented Architecture for Smart University
5 Collaboration in Smart University
5.1 Motivations
5.2 Smart Collaborative Learning
5.3 Collaborative Smart University Services
6 Discussion
7 Conclusion
References
Multi-Agent System of an Adaptive Learning Hypermedia Based on Incremental Hybrid Case-Based Reasoning
Abstract
1 Introduction
2 Adaptive Learning System
3 Case Based Reasoning
4 Multi-agent Adaptive Learning System
4.1 Agent
4.2 Multi-Agent System
5 Our Proposed IHCBR Architecture
5.1 Architecture Presentation
5.2 Description of the Architecture
6 Conclusion and Perspectives
References
Augmented Reality Application in Laboratories and Learning Procedures
Abstract
1 Introduction
2 Related Work
2.1 Augmented Reality in Education
2.2 Augmented Reality in Laboratories
3 Method
3.1 Working Methodology
3.1.1 Tracking
3.1.2 Alignment
4 Conclusion
References
Pedagogy in the Digital Age: Making Learning Effective and Engaging for Students
Abstract
1 Introduction
2 Blended Learning
3 Interactive Learning Sites
3.1 Kahoot!
3.2 Quizlet
3.3 Quizalize
3.4 Wooclap
3.5 Jeopardylabs.com
3.6 PowToon
3.7 Interactive Whiteboards
4 Methodology
4.1 Sample
4.2 Instrument
4.3 Data Collection and Analysis
4.4 Findings
5 Conclusion
6 Limitations
References
Towards Recommendation Using Learners’ Interest in Social Learning Environment
Abstract
1 Introduction
2 Related Works
3 Discussion
4 System Framework
4.1 The Pedagogical Resources
4.2 The Learner History
4.3 The System Analyzer
4.3.1 Group Construction
5 Experiment
6 Conclusion and Future Work
References
Toward a Generic Student Profile Model
Abstract
1 Introduction
2 Student Profile Modeling
2.1 Definitions
2.2 Student Modeling Approaches
3 Related Works
4 Comparative Study
4.1 Criteria
4.2 Analysis and Discussion
4.3 Global Student Features
5 Conclusion and Future Work
References
Digital Business and Smart Governance
Digital Business Models: Doing Business in the Digital Era
Abstract
1 Introduction
2 Fundamental Concepts
2.1 Business Models: The Concept’s Definition
2.2 Business Models: The Origins
2.3 Digital Business Models
3 Methodology
4 Finding and Discussions
4.1 Customer Experience in the Digital Age
4.2 Data: The Fuel of the Digital Age
4.3 Platforms: The New Trend
4.4 Preparing to Compete in the Digital Age
4.5 Digital Strategy: The New Way of Doing Business
4.6 Content in the Digital Age
4.7 Open Innovation: The New Imperative for Creating Value
5 Digital Business Model Canvas
6 Research Directions
7 Conclusions
References
Social Media Made Me Buy It: The Impact of Social Media on Consumer Purchasing Behavior and on the Purchase Decision-Making Process
Abstract
1 Introduction
2 Research Aims and Objectives
3 History and Classification of Social Media
4 Literature Review
5 Social Media: Online Word of Mouth and Marketing Special
6 The Efficiency of Brands’ Marketing Strategies in Social Media Platforms
7 The Impact of Social Media on the Different Stages of Decision Making Process (Analysis of the Classical or Traditional Purchase Model)
8 The Impact of Social Media on Consumer Purchasing Behavior
9 Methodology
10 Data Analysis and Results
11 Conclusion
Appendix
References
Literature
Electronic Sources
Machine Learning as an Efficient Tool to Support Marketing Decision-Making
Abstract
1 Introduction
2 Related Works
3 Method
3.1 Collection the Campaign Features
3.2 Recipient Profile Features Collection
3.3 Predicting Models Establishment
4 Results and Discussion
5 Conclusion and Perspectives
References
Big Data Science and Analytics for Tackling Smart Sustainable Urbanism Complexities
Abstract
1 Introduction
2 Conceptual Background and Theoretical Constructs
2.1 Smart Sustainable Urbanism
2.2 Data-Intensive Science
3 Wicked Problems and Sustainable Urbanism
4 The Essence and Dilemma of Wicked Problems in Urbanism
5 Big Data Uses Opportunities in Sustainable Urbanism
6 Urban Science and Data-Intensive Science for Transforming Urban Sustainability Science
7 Conclusion
References
Towards an Eco Responsible Corporate Governance
Abstract
1 Introduction
2 The Model of Eco-Responsible Decision for Companies
2.1 The Decision-Making Dimensions of a Companies
2.2 The Dimensions of a Company’s Ecological Decision
3 Corporate Governance
3.1 Comparative Study Between Shareholder and Stakeholder Theory of Corporate Governance
3.2 The Aspect of Reasonable Corporate Governance System
4 The Impact of the Eco Responsible Decision on the Governance System: The Case of Moroccan Companies
4.1 The Relationship Between Eco Responsible Decision and Corporate Governance
5 Methodological Structure
5.1 The Use of Neural Networks as a Method of Measurement of the Relationship Between Eco Responsible Decision and Corporate Governance System: The Case of Moroccan Companies
6 Smart Ecological Mode of Corporate Governance
7 Conclusion
References
Comparative Study Using Neural Networks Techniques for Credit Card Fraud Detection
Abstract
1 Introduction
2 Related Works
2.1 Artificial Neural Network (ANN)
2.2 Multilayer Preceptron (MPL)
2.3 Long Short-Term Memory (LSTM)
2.4 Recurrent Neural Network (RNN)
2.5 Gated Recurrent Unit (GRU)
2.6 Self-organization Map (SOM)
3 Method of Comparison
3.1 NN
3.2 MPL
3.3 CNN
4 Data Description
4.1 Dataset Generation
4.2 Feature Engineer
5 Conclusion
References
Smart Healthcare
Application of Unsupervised Machine Learning to Cluster the Population Covered by Health Insurance
Abstract
1 Introduction
2 Research Methodology
2.1 Unsupervised Learning: Clustering Analysis
2.2 K-Means Clustering Algorithm
2.3 Multiple Correspondence Analysis (MCA)
3 Results and Discussion
3.1 Use of MCA
3.2 Application of K-Means Algorithm
3.3 Validation Measure: Silhouette Coefficient
3.4 Empirical Validation
3.5 Comparison Between K-Means and K-Nearest Neighbors
4 Conclusion
References
Secure Communication in Ehealth Care Based IoT
Abstract
1 Introduction
2 Related Work
3 Ehealth System
3.1 E-Health Architecture
3.2 E-health Procedure
4 Security Requirements
5 Secure Communication
5.1 Secure Layers
5.1.1 Physical and Media Access Layers
5.1.2 From Physical to Application Layers
5.1.3 Application Layers and Transport
5.1.4 Secure Routing
6 Discussion
7 Conclusion
References
Internet of Things Ehealth Ecosystem: Solution
Abstract
1 Introduction
2 Internet of Things (IoT)
2.1 Deftnition of IoT
2.2 Architecture of IoT
2.3 WSN in the Perception Layer of the IoT Architecture
3 Application of IoT on the Ehealth Ecosystem
3.1 EHealth Goals
3.2 Types of Applications and Portable Objects Used in EHealth
3.3 Social Interest of EHealth
3.4 Application and Data
3.5 Priority Use Cases
3.6 Technical Tools: The Micro Computer and the Shield Used
4 Related Work
5 Simulation of the IoT Ehealth Ecosystem
5.1 The WIMAX Protocol in the Physical and Datalink Layers
5.2 Using an Application Base on the HTTP Protocol
5.3 Using an Application Base on the FTP Protocol
5.4 Using an Application Base on the Email Protocol
5.5 Using a Voice Application
5.6 Using a Remote Login Application
6 Conclusion
References
Risks of Radiation Exposure During Abdominopelvic CT Procedures at Mohamed VI University Hospital of Oujda, Morocco
Abstract
1 Introduction
2 Materials and Methods
2.1 Patients Data
2.2 Effective Dose Calculation
2.3 Estimates of Cancer Risk
2.4 Estimates of Hereditary Risks
2.5 Statistical Data
3 Results
4 Discussion
5 Conclusion
References
Glucose Sensing for Diabetes Monitoring: From Invasive to Wearable Device
Abstract
1 Introduction
2 Glucose Monitoring Sensors
2.1 Blood-Based Diagnostics (The Gold Standard)
2.2 Tear-Based Diagnostics
2.3 Sweat-Based Diagnostics
2.4 Saliva-Based Diagnostics
3 Comparative Study
4 Proposed Glucometer
4.1 Materials and Methods
5 Results
5.1 Measurement Stage (Electrode Electrochemical Cell)
5.2 Electronic Glucometer Circuit
5.3 Amplification Stage
5.4 Filtering Stage
6 Discussion
7 Conclusion
References
Framework on Mobile Technology Utilization for Assisted Healthcare Service Request and Delivery of Aged Person: A Case of Ghana
Abstract
1 Introduction
2 Background
3 Related Work
3.1 Growth of Mobile Phone
3.2 Challenge to mHealth Implementation from a Developer Perspective
3.3 Use of Mobile Phones by Frontline Health Workers
3.4 Growing Aged Population
4 Methodology
5 Proposed Conceptual Framework
6 Evaluation of Conceptual Framework
6.1 Analysis of Data from Health Professionals
6.2 Analysis of Data from Patients
7 Schema to Assess Data Flow
8 Business and Economic Benefits
9 Conclusion
References
Segmentation and Classification of Microcalcifications Using Digital Mammograms
Abstract
1 Introduction
1.1 Mammography Images
2 Breast Cancer Detection
2.1 Related Work
3 Proposed Approach
3.1 Preprocessing Step
3.2 Segmentation Step
3.3 Features Extraction Step
3.4 Classification Step
3.5 Experimental Result
3.6 Pre-processing Evaluation
3.7 Segmentation Evaluation
4 Conclusion
References
Rough Set Based Supervised Machine Learning Approaches: Survey and Application
1 Introduction
2 Rough Set Theory: Key Notions
3 Involved Methods
3.1 Discretization Methods
3.2 Attribute Reduction Methods
3.3 Rule Induction Methods
4 Experiments: Description, Analysis and Discussion
4.1 Used Database Description
4.2 Test Environment
4.3 Realized Experiments
4.4 Results and Discussion
5 Conclusion
References
Knee Functional Telerehabilitation System for Inclusive Smart Cities Based on Assistive IoT Technologies
Abstract
1 Introduction
2 Related Work
2.1 IoT Functions and Architectures
2.2 IoT Platform and Framework
3 Material and Method
3.1 Knee Angle Measurement
3.2 Processing and Transmission Unit
3.3 Cloud Data Logging and Visualization
4 Experimental Test and Results
4.1 Prototype Implementation
4.2 Test of Measurement Accuracy
4.3 Results of Evaluation Test
5 Conclusion
References
Configuring MediBoard HIS for Usability in Hospital Procedures
Abstract
1 Introduction
2 Materials
2.1 MediBoard
2.2 OpenMRS
2.3 OpenEMR
2.4 HospitalOS
2.5 Care2x
2.6 OpenHospital
2.7 MedinTux
2.8 HOSxP
2.9 PatientOS
2.10 Linux
2.11 Apache
2.12 PHP
2.13 MySQL
2.14 IEEE 12207
2.15 ISO 14764
3 Methods
3.1 The DeLone & McLean Model
3.2 The SQALE Method
3.3 Comparative Study Implementations
3.4 Maintenance of MediBoard
3.4.1 Identifying the Problem
3.4.2 Analysis and Design
3.4.3 Implementing Change
4 Results and Discussions
4.1 Comparative Study Results
4.2 MediBoard Maintenance Results
5 Conclusion
References
Type 2 Diabetes Mellitus Prediction Model Based on Machine Learning Approach
Abstract
1 Introduction
2 Some Related Work
3 Materials and Methods
3.1 Diabetics Database Description
3.2 Machine Learning Algorithms for Prediction
3.3 Proposed Predictive Model
3.4 Performance Evaluation Methods
4 Results and Discussion
4.1 Results Evaluation and Comparison
4.2 Discussion
5 Conclusion and Perspectives
References
Hybrid Method for Breast Cancer Diagnosis Using Voting Technique and Three Classifiers
Abstract
1 Introduction
2 Breast Cancer
2.1 Types of Breast Cancer
2.2 Diagnosis of Breast Cancer
3 Related Works
4 Machine Learning Algorithms
4.1 Bayesian Network (BN)
4.2 Support Vector Machines (SVM)
4.3 K Nearest Neighbors Algorithm (KNN)
5 Voting Classifier Technique
6 Proposed Method
7 Experimentation and Results
7.1 Description of the Dataset
7.2 WEKA Tool
7.3 K-Fold Cross-validation
7.4 Confusion Matrix
7.5 Bayesian Network (BN)
7.6 Support Vector Machines (SVM)
7.7 BN-SM
7.8 K Nearest Neighbors Algorithm (KNN)
7.9 BN-SVM-KNN
8 Conclusion
Acknowledgement
References
Visual Question Answering System for Identifying Medical Images Attributes
Abstract
1 Introduction
2 Related Works
3 The Proposed Model
3.1 Image Representation
3.2 Question Representation
3.3 Merge
3.4 Encoder Decoder Model
3.5 Answer Generation
4 Results and Discussion
4.1 Evaluation and Results
4.2 Discussion
5 Conclusion
References
New Generation of Networks and Systems for Smart Cities
Enhanced Mobile Network Stability Using Average Spatial Dependency
1 Introduction
2 Related Work
2.1 Mobility vs Predictive MPR Selection
2.2 Mobility Quantification for MultiPoint Relays Selection Algorithm
2.3 Mobility Models Selection
3 Impact of the Mobility on the Network
3.1 Results and Discussions
4 The Proposed Model
4.1 Proposed Solution
4.2 Implementation
5 Result Discussion
5.1 Simulation Setups
5.2 Performance Results
6 Conclusion
References
A Review on 3D Reconstruction Techniques from 2D Images
Abstract
1 Introduction
2 Overview on 3D Reconstruction from Images
2.1 What is 3D Reconstruction from Images
2.2 Fields of Application of 3D Reconstruction from Images
2.3 3D Reconstruction from Images Requirements
3 Active and Passive Reconstruction Methods
3.1 Active Methods
3.2 Passive Methods
4 2D to 3D Conversion Algorithms
4.1 Binocular Disparity
4.2 Motion Parallax
4.3 Image Blur
4.4 Silhouette
4.5 Linear Perspective
4.6 Atmosphere Scattering
4.7 Shape from Shading
4.8 Structure from Motion
5 Comparison
5.1 Relative or Absolute Depth
5.2 Depth Range
5.3 Real Time Processing
5.4 Image Acquisition
5.5 Image Content
5.6 Motion Presence
5.7 Dense or Sparse Depth Map
5.8 State of Each Depth Cue
6 Conclusion
References
Analytical Approaches and Use Case on Network Interactions
Abstract
1 Introduction
2 Biological Network Analysis
2.1 Biological Networks Taxonomy
2.2 Network Measures
3 Cytoscape
3.1 Use Case Protein Interactions and Analysis on a Biogrid Database Network
4 Discussion
5 Conclusion
References
Performance Analysis of Cooperative MIMO Systems for Mobile Coverage in Smart City Application
Abstract
1 Introduction
2 MIMO-OFDMA System
3 Analysis of Cooperative MIMO System in Smart City Network
4 Simulation Analysis
4.1 Impact of Cooperative MIMO System on Spectral Efficiency
4.2 Cooperative MIMO for Smart Cities
5 Conclusion
References
Embedded Systems Hardware Software Partitioning Approach Based on Game Theory
1 Introduction
2 Related Works
3 Proplem Definition
4 Proposed Approach
4.1 Game Strategy
5 Implementation
5.1 MiniMax Algorithm
5.2 Proposed Algorithm
6 HSP with Three Metrics
7 Experiments
8 Conclusions
References
Comparative Study and Improvement of the Routing Protocols Used in the Vehicular Networks and v2v Communications
Abstract
1 Introduction
2 Routing Protocols
3 Topology-Based Routing Protocol
3.1 Proactive (Table-Driven) Routing Protocols Routing
3.1.1 DSDV (Destination-Sequenced Distance Vector)
3.1.2 OLSR (Optimized Link State Routing Protocol)
3.2 Reactive (On-Demand) Routing Protocols
3.2.1 AODV (Ad Hoc On-Demand Distance Vector)
3.2.2 Dynamic Source Routing DSR
3.3 Hybrid Routing Protocols
3.3.1 ZRP (Zone Routing Protocol)
4 Position-Based Routing Protocols
4.1 VADD (Vehicle-Assisted Data Delivery)
4.2 GPSR (Greedy Perimeter Stateless Routing)
4.3 GPCR (The Greedy Perimeter Coordinator Routing)
4.4 GyTAR (Improved Greedy Traffic Aware)
5 Comparative Analysis
5.1 Comparison Between Routing Protocols
6 Improvement of Routing Protocols
6.1 Improvement of the AODV Protocol
6.2 Simulation Results
6.2.1 Throughput
6.2.2 End to End Delay
6.2.3 Packet Delivery Fraction (PDF)
6.3 Proposed Improvement of the GPSR Protocol
7 Discussion
8 Conclusion and Future Work
References
Performance Study of Position-Based Zoning Techniques in MANETs
1 Introduction
2 Related Work
3 Problem Formulation
4 OLSR Model
5 Proposed Work
5.1 Geographic Forwarding Rules
5.2 Zone MPR Computation
5.3 Hybrid Zoning Algorithm : Z-GFR
6 Simulation and Results
7 Conclusion
References
Logical Structure of an IPv6 Network that Perfectly Uses the Summarization Technique
1 Introduction
2 Historical Overview
2.1 Step 1: Schematization of the Summarization Technique
2.2 Step 2: Finding an Aggregation Process
3 The Proposed Method
3.1 The Generating Process of the Logical Model
3.2 Illustration
Initiation.
Generating.
Outputting.
4 Discussion
4.1 The Basic Address Plan vs the Proposed Model
4.2 Subdivising and Minimizing the Routing Tables
4.3 The Value of the Aggregation Parameter (n)
5 Conclusion
References
A Simulation Analyses of MANET’s Attacks Against OLSR Protocol with ns-3
Abstract
1 Introduction
2 Background
2.1 Optimized Link State Routing Protocol
2.2 Link-Spoofing Attack in MANETs
2.3 Replay Attack in MANETs
2.4 Data-Flooding Attack in MANETs
3 Experimental Environment Setup
4 Performance Metrics
4.1 Packet Delivery Rate
4.2 Normalized Routing Load
4.3 Average End-to-End Delay
5 Simulation Result Analysis of Link-Spoofing Attack
5.1 Packet Delivery Rate
5.2 Normalized Routing Load
5.3 Average End-to-End Delay
6 Simulation Result Analysis of Replay Attack
6.1 Packet Delivery Rate
6.2 Normalized Routing Load
6.3 Average End-to-End Delay
7 Simulation Result Analysis of Data-Flooding Attack
7.1 Packet Delivery Rate
7.2 Normalized Routing Load
7.3 Average End-to-End Delay
8 Conclusions
References
Functional Modeling of IoT Protocols
Abstract
1 Introduction
2 Internet of Things
3 Applications Protocols
3.1 Protocol CoAP
3.2 Protocol MQTT
3.3 Protocol XMPP
3.4 AMQP
3.5 Protocols DDS
3.6 Protocol WebSocket
3.7 Protocol Rest API
4 Comparative Study of Protocols
5 Operating Models
5.1 Operating Model CoAP
5.2 Operating Model MQTT
5.3 Operating Model AMQP
5.4 Operating Model XMPP
5.5 Operating Model DDS
5.6 Operating Model WebSocket
5.7 Operating Model Rest API
6 Conclusion
References
New Tool for 5G Radio Network Planning and Deployment
Abstract
1 Introduction
1.1 NYUSIM Properties
1.2 NYUSIM Limits
2 2TISIM Software
2.1 2TISIM Conception
2.2 2TISIM Properties
2.2.1 Channel Model
2.2.2 Received Power and Coverage Limit
2.2.3 Path Loss Exponent
2.2.4 DoD and DoA Estimation
2.3 2TISIM Assets
3 Simulation Results
3.1 Simulation Parameters
3.2 Path Loss (PL) for Different Frequencies
3.3 Received Power for Different Frequencies
3.4 DoD and DoA Representations
4 Conclusion
References
Transmit-Power and Interference Control Algorithm in Cognitive Radio Network Based on Non-cooperative Game Theory
Abstract
1 Introduction
2 Related Work
3 System Model
4 Game Models for Interference and Power Control
4.1 Game Formulation
4.2 Algorithm Implementation
4.3 Static Non-cooperative Game
5 Simulation and Analysis
5.1 Effect of the Distance Between the Pairs on NE
5.2 Effect of the Distance Between (Sender, Receiver) Pair
5.3 Effect of the Path-Loss Exponent
6 Conclusion
References
Network Coding for Energy Optimization of SWIMAC in Smart Cities Using WSN Based on IR-UWB
Abstract
1 Introduction
2 Smart Cities Using WSN
3 MAC Protocols for WSN Based on IR-UWB
3.1 ALOHA
3.2 SWIMAC
4 XOR Network Coding
5 NC Implementation for SWIMAC
6 Simulation and Results
6.1 Simulation Parameters
6.2 Results
7 Conclusion
References
5G Energy Efficiency for Smart Cities: A Call Admission Control Proposition
Abstract
1 Introduction
2 Call Admission Control (CAC) in Mobile Networks
2.1 A CAC Evolution in Mobile Networks
2.2 A New RRC State Is Introduced in 5G
3 Power Consumed in the Macrocell in UDN
4 Call Admission Control with Lower Energy Consumption in 5G
5 Algorithm of Call Admission Control with Lower Energy Consumption in 5G
6 5G Frequency Spectrum and Propagation Models
6.1 5G Frequency Spectrum
6.2 5G Propagation Models
7 Some Applications for Smart Cities
8 Conclusion
References
Watermarking Image Scheme Based on Image Content and Corners Decomposition Method
Abstract
1 Introduction
2 Spatial Domain Based Methods
2.1 Feature Points Detectors
2.2 Image Decomposition Method
3 The Proposed Scheme
3.1 Watermark Embedding Scheme
3.2 Watermark Detection Scheme
4 Simulation, Results and Discussion
4.1 Simulation
4.2 Experimental Results
4.3 Discussion
5 Conclusion
References
Smart Grids and Electrical Engineering
A Combined Source and Demand-Side Energy Management System for a Grid-Connected PV-Wind Hybrid System
Abstract
1 Introduction
2 Hybrid Renewable Energy
2.1 Photovoltaic System
2.2 Wind Turbine Generation
2.3 Storage System
2.4 Proposed Hybrid System Interconnection Strategy
3 Hybrid System Energy Optimization: Our Proposed
4 Simulation and Results
5 Conclusions
References
New Hybrid Approach Multi-agents System and Case Based Reasoning for Management of Common Renewable Resources
Abstract
1 Introduction
2 Related Work
3 Problematic
4 State of the Art
4.1 Dynamic Case-Based Reasoning
5 The MDA Approach
6 Our Proposed Approach
7 Design and Modeling
7.1 Description of the AUML Class Diagram
7.2 Sequence Diagram
8 Implementation of Our Approach on JADE Platform
8.1 General Architecture
8.2 JADE Platform
8.3 Implementation of Our Approach
9 Conclusion and Perspectives
References
Robust Control of Induction Motor Drive Facing a Large Scale Rotor Resistance Variation
1 Introduction
2 IM Drive Modeling
3 Indirect Field Oriented Control
4 Sliding Mode Approach for Robust Control
4.1 Sliding Mode Control Design
4.2 Application to the Induction Machine
5 Extended Kalman Filter for Rotor Flux and Speed Observation
6 Simulation Results
7 Conclusion
References
Optimization of the Feasibility Study of an Energy Production System Based on the Wind and the Marine Current Turbine in Morocco
1 Introduction
1.1 Motivation
1.2 Related Work
1.3 Contribution
2 Ksar Sghir Load and Renewable Energy Potential
2.1 Site Description
2.2 Wind Energy
2.3 Marine Current Potential
2.4 Load Profile
3 Methodology
4 Ksar Sghir Hybrid System Modeling
4.1 System Modeling on HOMER
4.2 System Modeling on GAMS
5 Results and Discussion
5.1 HOMER Results
5.2 GAMS Results
6 Conclusion
References
Identification of Relevant Input Variables for Prediction of Output PV Power Using Artificial Neural Network Models
Abstract
1 Introduction
2 System and Database Description
2.1 System Description
2.2 Data Base Description
3 ANN
4 Methodology
5 Results and Discussion
5.1 Choice of the Best Model
5.2 Prediction of PV Output Power
6 Conclusion
References
Smart Mobility
On the Communication Strategies in Heterogeneous Internet of Vehicles
Abstract
1 Introduction
2 Background
3 Related Works
4 Proposed Protocol (ERTP-IOV)
4.1 Interface Selection Steps
5 Performance Evaluation in Vanet
6 Conclusion
References
A Comparison of Random Forest Methods for Solving the Problem of Pulsar Search
1 Introduction
2 Proposed Random Forest Methods
2.1 Random Forest (method = `rf')
2.2 Random Forest Satges
2.3 Parallel Random Forest (method = `parRF')
2.4 Random Forest by Randomization (Method = `Extra-Trees')
2.5 Regularized Random Forest (method = `RRF')
2.6 Conditional Inference Random Forest (method = `cforest')
2.7 Random Forest (method = `ranger')
3 Results, Analysis and Discussion
3.1 Random Forest Methods Parameters
3.2 Calcul of Evaluation Metrics
3.3 Experimental Results
3.4 Analysis and Discussion
4 Conclusion
References
Digital Service Supply Chain Management: Current Realities and Prospective Visions
Abstract
1 Introduction
2 Supply Chain Management: Realities and Innovations for the Service Company
3 Digitalization of the Company, a Popular Concept
4 The Digital Supply Chain Management of the Service Company
4.1 The Current Digitization of the Service Company’s SCM
4.2 Digitalization of the Service Company’s SCM: A Prospective Approach
5 Conclusion
References
Providing Context Awareness in the Smart Car Environment: State of the Art
Abstract
1 Introduction
2 Background
2.1 Smart Car
2.2 Context-Awareness
3 Related Work Review
4 Comparative Study
4.1 Criteria
4.2 Comparison and Discussion
5 Smart Car Context-Awareness Model
6 Conclusion
References
Applying External Guidance Commands to Deep Reinforcement Learning for Autonomous Driving
Abstract
1 Introduction
2 Related Work
3 Deep Reinforcement Learning with External Commands
3.1 Actor-Critic DRL for Continuous Action Space
3.2 DRL with External Control Commands
4 Application on Self-driving Car
4.1 CARLA Self-driving Simulator
4.2 DDPG for Self Driving Car
4.3 A3C for Self Driving Car
4.4 DDPG and A3C with External Commands
4.5 DDPGG and A3CG Model Training
4.6 DRL with Guidance (A3CG and DDPG): Test Results
4.7 Results Summary
5 Conclusion
References
Image Correlation Based Smart Throttle-Brake Control System for Disability Vehicles
Abstract
1 Introduction
2 Related Work
3 Intelligent Brake System
3.1 Collision Avoidance System
3.2 Anti-lock Braking System (ABS)
4 Proposed Intelligent Throttle and Brake Control System
4.1 Correlation Based Image Processing
5 Implementation and Results
6 Conclusion
References
A Multi-layer System for Maritime Container Terminal Management Using Internet of Things and Big Data Technologies
Abstract
1 Introduction
2 Related Works
3 Description of Container Terminal Management
4 Description and Comparison of Different Technologies Related to IoT and Big Data
4.1 Internet of Things Technologies
4.2 Big Data Technologies
5 The Architecture of the Proposed System
5.1 Data Reception
5.2 Data Management
5.3 Data Interpretation for Decision Making
6 Conclusion
References
Meteorological Parameters Prediction Along Roads Between Two Cities for the Safest Itinerary Selection
Abstract
1 Introduction
2 Related Works
3 Time Series Modeling
4 Autoregressive Integrated Moving Average (ARIMA) Models
5 The Proposed System
5.1 Collection and Pretreatment of Data
5.2 Prediction of the Time Series by the ARIMA Method
5.3 Classification
6 Results and Discussion
6.1 Route1: Highway A5: In the Middle of the Road Between Tangier and Tetouan (Latitude: 35.77214 and Longitude: −5.61594)
6.2 Route2: The Mediterranean Road RN16: In the Middle of the Road Between Tangier and Tetouan (Latitude: 35.83552 and Longitude: −5.57255)
6.3 Route 3: The RN2 National Road: In the Middle of the Road Between Tangier and Tetouan (Latitude: 35.56479 and Longitude: −5.59341)
7 Case of Study
8 Conclusion
References
Vehicle Traffic Management with the Help of Big Data Technologies
Abstract
1 Introduction
2 Architecture of VANET
3 State of Art
4 Methodology and Experiment
5 Conclusion
References
Smartwatch-Based Wearable and Usable System for Driver Drowsiness Detection
Abstract
1 Introduction
2 Related Work
2.1 Physiological-Based Drowsiness Detection
2.2 Wearable Drowsiness Detection Solutions
3 Solution Architecture
4 Measurement Algorithms
4.1 Galvanic Skin Response
4.2 Training and Monitoring
5 Evaluation
5.1 Methodology
5.2 Metrics
5.3 Results
6 Usability Evaluation
6.1 Usability Task Test
6.2 System Usability Survey
6.3 Societal Impact of a Usable Solution
7 Conclusions and Future Work
Acknowledgement
References
Smart Life Saving Navigation System for Emergency Vehicles
Abstract
1 Introduction
2 Convolutional Neural Network (CNN)
3 Fast Region-Based Convolutional Network (Fast R-CNN)
4 You Only Look Once (YOLO) Algorithm
5 Related Work
5.1 Machine Learning Based Methods
5.2 Lane Detection Schemes
5.3 Lane Planning Schemes
5.4 Intersection and Lateral Control Schemes
5.5 Co-ordinated Driving Schemes
6 System Architecture
6.1 Training
7 Implementation
7.1 Hardware Required
7.2 Dataset
7.3 Software Required
8 Experimental Analysis
9 Conclusion
References
Smart-Logistics for Smart-Cities: A Literature Review
Abstract
1 Introduction
2 Review of Literature
2.1 Logistics and Smart-Logistics
2.2 Smart-Transportation for Smart-Cities
2.3 Internets of Things (IoT) and Green-IoT
2.4 Big-Data Analysis
3 Big Data and IoT Application for «Smart-Logistics»
4 Conclusion
References
Smart Security
Towards a Holistic Privacy Preserving Approach in a Smart City Environment
Abstract
1 Introduction
2 Privacy in Smart Cities
2.1 Privacy Concept
2.2 Privacy in Smart Cities: Major Risks and Stakeholders
2.3 Privacy Languages, Formalization and Conflicts Issues
2.4 Current Privacy Preserving Approaches and Limitations
3 Presentation of Some Privacy Laws/Regulations
3.1 The United States Case
3.2 Canada Case
3.3 European Union Case
3.4 Moroccan Case
3.5 Case Study: Privacy Laws and Conflicting Policies and Preferences
4 Related Work and Discussion
5 Conclusion and Future Work
References
A Hybrid Intrusion Detection System Against Egoistic and Malicious Nodes in VANET
Abstract
1 Introduction
2 Architecture of Vehicular Ad-Hoc Network
3 Different Attacks in Vanet
3.1 Attacks on Identification and Authentication
3.2 Attacks on Availability
3.3 Attacks on Confidentiality
3.4 Attacks on Integrity
3.5 Attacks on Non-repudiation
4 Intrusion Detection Systems
4.1 Definition and Different Types of IDS
4.2 IDS Techniques Used in VANET
5 Our Contribution
6 Simulation and Results
6.1 Attacks Detected
6.2 Corroborated Attacks
6.3 Average Time of Corroboration
6.4 Alert Packets Generated
7 Conclusion
References
A New Secure Schema to Enhance Service Availability in Urban IoT
Abstract
1 Introduction
2 Related Work
2.1 Black Hole Attack
2.2 Countermeasures Against Black Hole Attack
2.3 Gray Hole Attack
2.4 Countermeasures Against Gray Hole Attack
3 Our Contribution
3.1 RAID Technology
4 Simulation and Results
4.1 Packet Delivery Ratio
4.2 End-to-End Delay
5 Conclusion
References
Evaluation of Deep Learning Approaches for Intrusion Detection System in MANET
Abstract
1 Introduction
2 Security Attacks in MANET
2.1 Denial of Service Attacks (DoS)
2.2 Remote to User Attacks (R2L)
2.3 User to Root Attacks (U2R)
2.4 Probing
3 Intrusion Detection System (IDS) in MANET
3.1 The Standalone Architecture
3.2 The Distributed and Collaborative Architecture
3.3 The Hierarchical Architecture
4 Deep Learning-Based Intrusion Detection System
4.1 Deep Learning in Network Security
4.2 Deep Learning Models-Based Intrusion Detection System
5 Experiment and Results
5.1 Data Preprocessing
5.2 Dataset
5.3 Evaluation Metrics
5.4 Results Analysis
6 Conclusion
References
Recommendation Enhancement Using Traceability and Machine Learning: A Review
Abstract
1 Introduction
2 Traceability Overview
2.1 Software Engineering
2.2 Web Engineering
2.3 Industry
3 Machine Learning
4 Related Works
4.1 Traceability-Based Recommendation
4.2 ML-Based Recommendation
5 Discussion
5.1 Changing User Preferences
5.2 Sparsity
5.3 Scalability
5.4 Synonymy
5.5 Privacy
6 Conclusion
References
Smart Incident Management, Prediction Engine and Performance Enhancement
Abstract
1 Introduction
2 Related Works
3 Prediction Model on Machine Learning
3.1 Unsupervised Learning
3.2 Supervised Learning
3.3 Classification
3.4 K-Nearest Neighbors (KNN)
3.5 Naïve Bayes Algorithm
3.6 Decision Tree Algorithm J48
4 The Proposed System
5 Results and Discussion
6 Conclusion
References
A Proposed Architecture Based on CNN for Feature Selection and Classification of Android Malwares
Abstract
1 Introduction
2 Feature Selection
2.1 Features Types
2.2 Feature Selection Methods
3 Deep Learning Models
3.1 CNN (Convolutional Neural Network)
3.2 RNN (Reccurent Neural Network)
3.3 RBM (Restricted Boltzman Machines)
3.4 DBN (Deep Belief Network)
3.5 DAE (Deep Autoencoder)
4 Related Work
4.1 Machine Learning
4.2 Deep Learning
5 Methodology
6 Conclusion
Acknowledgement
References
Classification of Grayscale Malware Images Using the K-Nearest Neighbor Algorithm
Abstract
1 Introduction
2 Related Works
3 Literature Review
3.1 Image Processing
3.2 K-Nearest Neighbor Algorithm
4 Proposed Solution
5 Results and Discussion
5.1 KNN Parameters
5.2 Performance Analysis
6 Conclusion
Acknowledgments
References
Improving Recommendations Using Traceability and Machine Learning
Abstract
1 Introduction
2 Traceability Overview
2.1 Software Engineering
2.2 Web Engineering
2.3 Industry
3 Machine Learning
4 Related Works
4.1 Traceability-Based Recommendation
4.2 ML-Based Recommendation
5 Discussion
5.1 Changing User Preferences
5.2 Sparsity
5.3 Scalability
5.4 Synonymy
5.5 Privacy
6 Conclusion
References
Sustainable Building
Bioclimatic Approach and Appropriate Financial Governance to Promote Sustainable Building Design
Abstract
1 Introduction
2 Energy and Comfort Concern in the Bioclimatic Approach
3 Experiment in Morocco
3.1 Measures Arising from Sustainable Vision
3.2 Demonstration Projects
4 Discussion
5 Conclusion
References
Contribution to a Smart Home Design for Medical Surveillance
Abstract
1 Introduction
2 Related Work
2.1 Smart Healthcare
2.2 Smart Home
2.3 IoT
2.4 Wearables
2.5 Multi Agent Systems
2.6 Artificial Intelligence
3 Proposed Smart Home Model for Medical Surveillance
3.1 The Proposed Smart Home Architecture
3.2 The Proposed Smart Healthcare Platform
3.3 The Proposed Smart Phone Timetable Application
4 Conclusion
References
Shading Devices’ Benefits on Thermal Comfort and Energy Performance of a Residential Building in Different Climates in Morocco
Abstract
1 Introduction
2 Methodology
2.1 Data Collection and Processing Approach
2.2 Thermal Comfort
2.3 Comfort Measurement
2.4 Air Conditioning Period
2.5 Unified Day Degree
2.6 Overhangs and Wingwall Shading
2.7 Reference Building
3 Results and Discussion
3.1 Energy Performance of the Reference Building
3.2 Overhangs and Wings Impact
4 Conclusion
References
Conjugate Natural Convection-Surface Radiation in a Square Cavity with an Inner Elliptic Body
Abstract
1 Introduction
2 Studied Configuration
3 Numerical Procedure and Validation
4 Results and Discussion
4.1 Streamlines and Isotherms
4.2 Eccentricity Effect
4.3 Inclination Angle \phi Effect
4.4 Average Heat Transfer
5 Conclusion
References
Well-Being Observing Framework in Smart Home
Abstract
1 Introduction
2 Towards the Observation of Health in the Smart Home
3 Related Work
4 Smart Home Configuration
5 Dataset
5.1 Sensors
5.2 Activities
5.3 Daily Activities
5.4 Objectives
6 Application Machine Learning
6.1 Linear Regression
6.2 KNN (K Nearest Neighbors)
6.3 Learning Our Model
7 Conclusion
References
Sustainable Environment
How to Improve Wastewater Treatment in Smart City
Abstract
1 Introduction
2 Materiel and Methods
2.1 Description of the Treatment Unit
2.2 Operation of the Filter
2.3 Measurement of Water Quality Parameters
2.4 Data Analysis
3 Results and Discussion
3.1 Trend of Temperatures and Dissolved Oxygen Concentrations
3.2 Trend of Electrical Conductivity and PH
3.3 Reduction of COD, BOD5 and TSS
3.4 Ammonia Removal and Nitrate Transformation
3.5 Elimination of Orthophosphate
3.6 Spatial Analyzes of All Parameters
4 Conclusion
Acknowledgements
References
Performance of Aluminum and Iron-Based Coagulants for the Removal of Water Turbidity for Human Consumption in the Cities (Rabat and Casablanca) of Morocco and Dewaterability of Hydroxide Sludge
Abstract
1 Introduction
2 Materials and Methods
2.1 Study Area
2.2 Determination of Optimum Coagulant Dose
2.3 Description of the Sludge Collection, Thickening and Dewatering Unit
3 Results and Discussion
3.1 Raw Water Treatment
3.2 Treatment of Hydroxide Sludge
4 Conclusion
Acknowledgements
References
Improving Safe and Sustainable Gray Water Reuse: A New Solution to Curb Water Shortages in Moroccan Cities
Abstract
1 Introduction
2 Materiel and Methods
2.1 Description of the Treatment Unit
2.2 Measurement of Water Quality Parameters
2.3 Data Analysis
3 Results
3.1 Trend in Effluent Temperature and Dissolved Oxygen Concentrations
3.2 Reduction of BOD5, COD and TSS
3.3 Ammonia Removal
3.4 Elimination of Orthophosphate
3.5 Spatial Analyzes of All Parameters
4 Conclusion
Acknowledgements
References
Historical Weather Data Recovery and Estimation
Abstract
1 Introduction
2 Methodology
2.1 Dataset
2.2 Analysis Techniques
3 Experimental Results
3.1 System Settings
3.2 Results
4 Conclusion
Acknowledgements
References
The Contribution of Cartography in Risk Management of Vector-Borne Diseases: Cas of Leishmaniasis in the Fez-Meknes, Region of Morocco
Abstract
1 Introduction
2 Materials and Methods
2.1 Study Area
2.2 Acquisition of Data
2.3 Data Processing
3 Results and Discussion
3.1 Temporal Distribution of the CL at the Level of the Region Fes Meknes
3.2 Distribution of Leishmaniasis Cases in the Fes-Meknes Region
3.3 Spatiotemporal Distribution of Leishmaniasis by Province Between 2008 and 2016
3.4 The Average Cumulative Incidence of Leishmaniasis
3.5 Health Risk Map
4 Conclusion
References
Spatial Relation Among Incidence of Leishmaniasis and Altitude Factor of Different Communes of Sefrou Province: Contribution of Geographic Information Systems
Abstract
1 Introduction
2 Methods
2.1 Study Area
2.2 Source and Analysis of Data
2.3 Data Processing
3 Results and Discussion
3.1 Frequency of Leishmaniasis
3.2 Spatial Distribution of the Average Cumulative Incidence of Leishmaniasis
3.3 Geographical Distribution of the Average Cumulative Incidence of Leishmaniasis According to the Altitude
4 Conclusion
Acknowledgments
References
Seasonal Variations of the Microbiological Parameters of the Quality of Water in Urban Oued Bouishak of the City of Meknes (Morocco)
Abstract
1 Introduction
2 Materiel and Methods
2.1 Study Area
2.2 Bacteriological Analyzes
3 Results and Discussions
4 Conclusion
References
Seasonal Variation of Parasitic Content of Wastewater Discharging in Boufekrane River at the Collector of the Agdal District (City of Meknes, Morocco)
Abstract
1 Introduction
2 Materials and Methods
2.1 Site of the Study
2.2 Methodology
3 Results and Discussions
3.1 The Specific Wealth
3.2 Prevalence
3.3 Frequency
3.4 Seasonal Repartition
4 Conclusion
References
Typology of the Surface Water Quality of the Aguelmam Sidi Ali Wetland (Midelt-Morocco)
Abstract
1 Introduction
2 Material and Method
2.1 Study Area
2.2 Sampling Plan
2.3 Monitored Parameters
2.4 Development Process of DWQI
2.5 One-Way Analysis of Variance
3 Results and Discussion
4 Conclusion
References
Geoinformatics Approach to Water Allocation Planning and Prognostic Scenarios Sustainability: Case Study of Lower Benue River Basin, Nigeria
Abstract
1 Introduction
1.1 Scaling of Modeling and Emerging Processes
1.2 Study Area Description
2 Sustainability Framework
3 Materials and Methods
3.1 SWAT-WEAP Model Development
4 Results and Discussion
4.1 Resultant SWAT Input Simulation
4.2 SWAT Results Discussion
4.3 Dependable Water Allocation Assessment
4.4 WEAP Prognostic Scenarios Results
4.5 WEAP Model Basin-Wide Water Balance
5 Conclusion
Acknowledgments
References
Author Index
Recommend Papers
File loading please wait...
Citation preview
Lecture Notes in Intelligent Transportation and Infrastructure
Series Editor: Janusz Kacprzyk
Mohamed Ben Ahmed Anouar Abdelhakim Boudhir Domingos Santos Mohamed El Aroussi İsmail Rakıp Karas Editors
Innovations in Smart Cities Applications Edition 3 The Proceedings of the 4th International Conference on Smart City Applications
Lecture Notes in Intelligent Transportation and Infrastructure Series Editor Janusz Kacprzyk, Systems Research Institute, Polish Academy of Sciences, Warszawa, Poland
The series “Lecture Notes in Intelligent Transportation and Infrastructure” (LNITI) publishes new developments and advances in the various areas of intelligent transportation and infrastructure. The intent is to cover the theory, applications, and perspectives on the state-of-the-art and future developments relevant to topics such as intelligent transportation systems, smart mobility, urban logistics, smart grids, critical infrastructure, smart architecture, smart citizens, intelligent governance, smart architecture and construction design, as well as green and sustainable urban structures. The series contains monographs, conference proceedings, edited volumes, lecture notes and textbooks. Of particular value to both the contributors and the readership are the short publication timeframe and the world-wide distribution, which enable wide and rapid dissemination of high-quality research output.
More information about this series at http://www.springer.com/series/15991
Mohamed Ben Ahmed Anouar Abdelhakim Boudhir Domingos Santos Mohamed El Aroussi İsmail Rakıp Karas •
•
•
•
Editors
Innovations in Smart Cities Applications Edition 3 The Proceedings of the 4th International Conference on Smart City Applications
123
Editors Mohamed Ben Ahmed Faculty of Sciences and Techniques of Tangier Mediterranean Association of Sciences and Technologies Tangier, Morocco Domingos Santos Polytechnic Institute of Castelo Branco Castelo Branco, Portugal
Anouar Abdelhakim Boudhir Faculty of Sciences and Techniques of Tangier Abdelmalek Essaadi University Tangier, Morocco Mohamed El Aroussi Hassania School of Public Works Casablanca, Morocco
İsmail Rakıp Karas Computer Engineering Department, Faculty of Engineering Karabuk University Karabük, Turkey
ISSN 2523-3440 ISSN 2523-3459 (electronic) Lecture Notes in Intelligent Transportation and Infrastructure ISBN 978-3-030-37628-4 ISBN 978-3-030-37629-1 (eBook) https://doi.org/10.1007/978-3-030-37629-1 © Springer Nature Switzerland AG 2020 This work is subject to copyright. All rights are reserved 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 Switzerland AG The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Preface
This book is one of the important series which encompasses recent and advanced research on Smart City Applications. It is an interesting manuscript that will help newer and advanced researchers, industrialists and policy-makers to understand, to attract and conclude new ideas, new solutions and applications in this area. It is also an opportunity to learn about the exiting scientific contributions to literature in order to develop and think about new ones. The paradigm of the future life of humanity is of great concern to scientists, researchers, sociologists and governments around the world. As a general scope of SCA19, the concept of smart cities is one of the most attractive models because in principle it brings together all the frameworks of human life: health, transport, education, energy, water, agriculture, pollution and environment, etc. Several researchers in various disciplines related to this field continue to delve extensively and through new information and communication technologies and using the latest advances in the field of applied artificial intelligence. In its third edition, this book lists original research in new directions and advances focused on multidisciplinary fields and closely related to the fields of smart cities and their applications. This edition is the result of a reviewed, evaluated and presented work in more than fifteen sessions opened and listed in SCA19 conference as followed: Smart E-Business and Governance–Smart Vehicles–Smart healthcare–Smart Education–Smart Citizenship–Smart Logistics and Mobility–Sustainable Building and Smart Earth–Smart Security Management–Smart Water Management–Smart Energy Management and Electrical Engineering–5G Technologies–Modeling and Algorithms–Networks and Wireless Communications–Image Processing. The present book contains selected and extended papers of the Fourth International Conference on Smart City Applications SCA19 co-organized conjointly by Medi-Ast Association and EHTP school of Casablanca and held on October 2–4, 2019, in Casablanca-Morocco. The SCA19 conference scope also discusses how smart cities are currently being conceptualized and implemented, examining the theoretical underpinnings and technologies that connect theory with tangible practice achievements. Using v
vi
Preface
numerous examples from different city contexts and countries, this book, thus, constitutes a precious contribution to the ongoing discussion of this urban phenomenon. We thank all authors from across the globe for choosing SCA19 to submit their manuscripts. A sincere gratitude to all keynote speakers for offering their valuable time and sharing their knowledge with the conference attendees. A special thanks go out to all the organizing committee members, to local chairs and the local committee in EHTP school, to all program committee members, to all chairs of sessions for their efforts and the time spent in order to make this event a success. Many thanks to Springer staff for their support and guidance. In particular, our special thanks to Dr. Thomas Ditzinger and Ms. Varsha Prabakaran for their help, support and guidance. Mohamed Ben Ahmed İsmail Rakıp Karas Anouar Abdelhakim Boudhir Laaroussi Mohamed Domingos Santos SCA19 Chairmen
Committee
Honor Committee M. Noureddine Maana
Conference General Chairs Ben Ahmed Mohamed Boudhur Anouar Abdelhakim
FSTT, Abdelmalek Essaadi University, Morocco FSTT, Abdelmalek Essaadi University, Morocco
Conference Local Chairs Mohamed Wahbi Rachid Saadane
EHTP, Casablanca, Morocco EHTP, Casablanca, Morocco
Conference Technical Program Committee Chairs Mohammed Karim Guennoun Boudhir Anouar Abdelhakim Wassila Mtalaa
EHTP, Casablanca, Morocco FST Tangier, Morocco Luxembourg Institute of Science and Technology, Luxembourg
Workshop Co-chairs El Rharras Abdessamad Said Agoujil Domingos Santos
EHTP, Casablanca, Morocco FST, Errachidia, Morocco Polytechnic Institute, Castelo Branco, Portugal
Publications Co-chairs Mohamed El Aroussi Ben Ahmed Mohamed İsmail Rakıp Karașo
EHTP, Casablanca, Morocco FST Tangier, Morocco Karabuk University, Turkey vii
viii
Committee
Tutorials Co-chairs Khalil Amine Mustapha El Moudden Senthil Kumar
EMI, Rabat, Morocco FS, MI University, Meknes Hindustan College of Arts and Science, India
Web Chair Ben Ahmed Mohamed
FSTT, Abdelmalek Essaadi University, Morocco
Publicity and Social Media Co-chairs Abdellaoui Alaoui El Arbi El Hafid Yassine Amine El Haddadi Mohamed Lahby
EIGSI, Casablanca, Morocco EHTP, Casablanca, Morocco Paul Sabatier University, Toulouse, France ENS, University Hassan II, Morocco
Sponsorship and Exhibit Chairs El Rharras Abdessamad Mustpha Harmouzi Adil Mehdary
EHTP, Casablanca, Morocco FS Tetouan, Morocco EHTP, Casablanca, Morocco
Technical Program Committee Ahmad S. Almogren Abdel-Badeeh M. Salem Alabdulkarim Lamya Accorsi Riccardo Alghamdi Jarallah Ahmed Kadhim Hussein Anabtawi Mahasen Arioua Mounir Astitou Abdelali Assaghir Zainab Bessai-Mechmach Fatma Zohra Benaouicha Said Ben Yahya Sadok Boulmalf Mohammed Boutejdar Ahmed Chadli Lala Saadia Damir Žarko Dousset Bernard
King Saud University, Saudi Arabia Ain shams university, Egypt King Saud University, Saudi Arabia Bologna University, Italy Prince Sultan University, Saudi Arabia Babylon University, Iraq Al-Quds University, Palestine UAE, Morocco UAE, Morocco Lebanese University, Lebanon CERIST, Algeria UAE, Morocco Faculty of Sciences of Tunis, Tunisia UIR, Morocco German Research Foundation, Bonn, Germany University Sultan Moulay Slimane, Beni Mellal, Morocco Zagreb University, Croatia UPS, Toulouse, France
Committee
Dominique Groux Elhaddadi Anass El-Hebeary Mohamed Rashad El Kafhali Said El Yassini Khalid El Yadari Mourad El Mhouti Abderrahim Ensari Tolga Enrique Arias En-Naimi El Mokhtar Haddadi Kamel Hazim Tawfik Jaime Lioret Mauri Jus Kocijan Khoudeir Majdi IUT Labib Arafeh Lalam Mustapha Loncaric Sven Mademlis Christos Miranda Serge Mousannif Hajar Ouederni Meriem Sagahyroon Assim Senthil Kumar Senan Sibel Slimani Yahya Sonja Grgić Tebibel Bouabana Thouraya Vo Ngoc Phu’s Abderrahim Ghadi Mohamed El Ghami Bataev Vladimir Hanane Reddad Ehlem Zigh Otmae Yazidi Alaoui Hossain El Ouarghi Lotfi El Achaak Elyusufi Yasyn My Lahcen Hasnaoui Nafil Khalid Abderrahmane Janyene Khadija Abbassi
ix
UPJV, France Mohammed Premier University, Morocco Cairo University, Egypt Hassan 1st University, Settat, Morocco Moulay Ismail University, Morocco FP, Errachidia, Morocco FST, Al-Hoceima, Morocco Istanbul University, Turkey Castilla-La Mancha University, Spain UAE, Morocco IEMN Lille University, France Cairo University, Egypt Polytechnic University of Valencia, Spain Nova Gorica University, Slovenia Poitiers University, France Al-Quds University, Palestine Mouloud Mammeri University of Tizi-Ouzou, Algeria Zagreb University, Croatia Aristotle University of Thessaloniki, Greece Nice University, France Cadi Ayyad University, Morocco INP-ENSEEIHT Toulouse, France American University of Sharjah, United Arab Emirates Hindustan College of Arts and Science, India Istanbul University, Turkey Manouba University, Tunisia Zagreb University, Croatia ESI, Alger, Algeria Duy Tan University, Vietnam FSTT UAE, Morocco University of Bergen, Norway Zaz Ventures, Switzerland USMS University, Morocco National Institute of Telecommunications and ICT of Oran, Algeria FSTT UAE, Morocco ENSAH UAE, Morocco FSTT UAE, Morocco FSTT UAE, Morocco Moulay Ismail University, Morocco UM5, Morocco Hassania School of Public Works, Morocco Hassania School of Public Works, Morocco
x
Horia Benomar Mohammed Tachafine Boutaher Jaouad Souhiel El Ghazi Omar Bachir Alami Kamal Jetto Hasna Chaibi Samira El Margae Rachid Ilmen Hassania Messaoud Nabil Benamar Mounir Deri Sami El Moukhlis Hassan El Brirchi Hassan Oulidi Jerrar Sinan Mohamed Addou Malika Abdelhamid Fadil Younes Adnani Brahim El Bhiri Mohamed Tabbaa Smail Tigani Darif Anouar Ishak Hbiak Bbadr El Hatimi Haidi Touria Asmae El Kohen Saghir Merouane Debbah Mehdi Bennis Yamna Ghabbar Salim Nafiri Nourddine Semane Mouloud Mouzzoun Essaid Saber Mohamed Kamili Samir Mbarki Hatim Kharraz El Aroussi Abdelatif Kobban Fakhri Youssef Mohammed Malaainine Fatima Zahra Barramou Zineb Rachik Mohamed Yatim
Committee
Hassania School of Public Works, Morocco Hassania School of Public Works, Morocco Hassania School of Public Works, Morocco Hassania School of Public Works, Morocco Hassania School of Public Works, Morocco Hassania School of Public Works, Morocco ENSIAS, Morocco Iben Toufiel University, Morocco Hassania School of Public Works, Morocco Hassania School of Public Works, Morocco EST Meknes, Morocco Hassania School of Public Works, Morocco AIAC, Morocco Hassania School of Public Works, Morocco Hassania School of Public Works, Morocco Hassania School of Public Works, Morocco Hassania School of Public Works, Morocco Hassania School of Public Works, Morocco EST, Ibn Tofail University, Morocco EMSI, Morocco EMSI, Morocco Euro-Medeteranienne University, Morocco FST Beni Mellal, Morocco Hassania School of Public Works, Morocco Hassania School of Public Works, Morocco Hassania School of Public Works, Morocco EHTP, Morocco Supelec, France Oulu University, Finland Hassania School of Public Works, Morocco Hassania School of Public Works, Morocco Hassania School of Public Works, Morocco Hassania School of Public Works, Morocco ENSEM-Morocco Sidi Mohammed Ben Abdellah University, Morocco Iben Toufiel University, Morocco Iben Toufiel University, Morocco ENSIAS, Morocco Iben Toufiel University, Morocco EHTP, Casablanca, Morocco EHTP, Casablanca, Morocco EHTP, Casablanca, Morocco EHTP, Casablanca, Morocco
Keynote Speakers
Smart City Strategic Planning: Trends and Challenges Domingos Santos Polytechnic Institute, Castelo Branco, Portugal [email protected]
Abstract. The aim of this paper is threefold. On the one hand, the first part of the paper aims to contextualize and clarify the concept of smart city that has been gaining a growing importance in terms of urban development, planning and policy. Despite its theoretical evolution, the smart city approach is still somehow fuzzy, not fully understood and lacking a proper conceptualization. On the other hand, the paper will also discuss three pathways (technological, institutional, human) that are usually suggested as alternatives to operationalize the smart city concept. We suggest that the smart city model is transitioning from a mainly technology-focused to a more socially oriented stage—searching for a more adequate equilibrium between the efficiency of processes and the effectiveness of outputs. The third part of the paper is dedicated to the understanding of the main trends and challenges that are marking the issue of smart cities, arguing that it needs a fertile environment guided by a clear vision, the participation of relevant actors and the efficient and effective organization of its processes, on the ambit of a city-wide smart city strategy, pointing out to some instruments that can be used to enhance urban competitiveness and sustainability. Additionally, stronger than ever effort and support toward interdisciplinary research are needed. One of the things that so far has been lacking is a multi-perspective and collaborative methodology toward a more integrated smart city making. Biography. Domingos Santos holds a degree in Environmental Engineering (New University of Lisbon), a master’s degree in Regional and Urban Planning (Technical University of Lisbon) and a PhD in Environmental Applied Sciences (University of Aveiro). He is a professor of the Polytechnic Institute of Castelo Branco (IPCB) where he has taught curricular units in the field of Social Development Planning, Development Programs and Projects, Innovation and Entrepreneurship, as well as Sustainable Development. xiii
xiv
D. Santos
He has developed teaching activities in the framework of cooperation and mobility programs, in: University of Valladolid-Faculty of Economic and Business Sciences and University of Extremadura, both in Spain; Lithuania Business University of Applied Sciences; and, in Brasil, at Dom Bosco Catholic University and University of Santa Cruz do Sul.
Citizens’ Engagement on Smart Cities Sehl Mellouli Laval University, Laval, Canada [email protected]
Abstract. Citizens’ participation (CP) aims to reinforce citizens’ involvement in decision-making processes of issues that will affect their communities. It is an interactive process that takes place between citizens themselves and between citizens and government officials in order to significantly contribute, among others, in public policy decisions in a transparent and responsible way. Several researchers and practitioners mentioned the important added value that participation can bring to both governors and governed. This added value consists of the capabilities to develop better understanding of communities’ problems and needs as well as to bring innovative ideas to handle these problems. In this context, people are more and more using social media to express themselves about different services that their governments are delivering. This is part of citizens’ participation. They express themselves by posting different types of information: texts, images or videos. In these posts, they can either provide positive or negative comments on government services. Consequently, it is important for cities to analyze these posts to extract this valuable knowledge in a comprehensive way. This analysis may bring new information to policy-makers that they may consider in their decision-making processes. To this end, cities require the development of new natural language processing tools to extract this valuable knowledge. Biography. Sehl Mellouli is a professor at the Department of Management Information Systems at Université Laval, Quebec City, Canada, since June 2005. Professor Mellouli’s research interests are related to smart cities, smart government, artificial intelligence, business intelligence and natural language processing techniques. He has publications in highly ranked journals and conferences. Professor Mellouli was involved in the last years in research projects with a total amount of more than 10 million dollars. Professor Mellouli is serving as an advisor for different
xv
xvi
S. Mellouli
governments and institutions. He is currently the president of the Digital Government Society. Professor Mellouli has a PhD in Computer Science from Université Laval, a MBA in Management Information Systems from Université Laval and an engineer degree in computer science from the Ecole Nationale des Sciences Informatiques, Tunisia.
Smart Building Sustainability Assessment Luís Bragança University Of Minho, Portugal [email protected]
Abstract. During the past years, there has been an increasing concern about the building’s negative impacts on the environment. Due to the need of providing proper answers to the growing demand for sustainable buildings, several processes and technologies have been developed to support designer’s decision making and to manage all the project data and complexity efficiently. Among those, stands out the Building Information Modeling (BIM), which is a working methodology that turns possible to manage all the project data and design in a virtual model, considering the entire life cycle of a building. By allowing multidisciplinary information to be overlaid and grouped into a single model, BIM creates an excellent opportunity to incorporate sustainable measures throughout the different stages of a project. Thus, by analyzing BIM software capabilities for building sustainability and published papers about the topic, this study investigates how the BIM method can enhance building’s sustainability. BIM practical applications are presented and listed, as well as their benefits and challenges discussed. Additionally, a couple of successful projects are analyzed, where BIM played a key role to achieve designers’ sustainable targets. BIM is still not oriented to building sustainability, but it has great potential to improve the actual context. This paper shows that it will become an essential tool for developing more sustainable and high-performance buildings, by effectively allowing to improve several aspects related to the main dimensions of sustainable construction. Biography. Luís Bragança is a PhD, MSc and Licentiate in Civil Engineering and is Professor at the Civil Engineering Department of the School of Engineering of University of Minho, Portugal. He is the coordinator of the international post-graduation programs (PhD and MSc) in Sustainable Built Environment, coordinator of the MSc in Sustainable Construction and Rehabilitation, director of the Building Physics and Construction Technology
xvii
xviii
L. Bragança
Laboratory and director of the Energy and Sustainable Building Laboratory. Professor Luís Bragança is the president of the International Initiative for a Sustainable Built Environment (iiSBE) since 2011.
Smart Governance: Leadership, Priorities and Indicators for Sustainable Cities M. Michel Raimondo Matriciel Consulting, Casablanca, Morocco [email protected]
Abstract. Governance is about conformance and performance. In order to manage risks, smart cities need to ensure that they are compliant with regulatory requirements about the data they acquire, store and share, on the one hand. When it comes to data protection, concerns are security, confidentiality and privacy. On the other hand, smart cities need to ensure who are the decision makers, how decisions are made, what are the decision-making processes and define plans, priorities, business rules and performance indicators. A smart sustainable city is an innovative city that uses technologies and other means to improve quality of life, efficiency of urban operations and services, while ensuring that it meets the needs of present and future generations with respect to economic, social and environmental aspects. Smart Cities need to define themes they want to manage such as infrastructure, energy, healthcare, mobility, water and waste. Examples of indicators are time, health, safety, environment, social activities, jobs and costs of living. Smart Cities need to implement an integrated approach for their data architecture and their application architecture. Strategic development plans should be based on some foundational principles such as citizen-centric, innovative funding, inclusive, data-driven, fit for purpose, services management, both top-down and bottom-up. Because of constant evolution of usages and needs, the main goal and challenge is bridging the digital divide between citizens and cities. Biography. Michel Raimondo is a Belgian citizen and an international consultant (Africa, Canada, France, UK, Belgium); he has been working as operational consultant for over 30 years, covering business transformation, change management programs, IT governance and strategic planning, portfolio/program management, business process re-engineering, business architecture/analysis. He specializes in linking strategies and operations, aligning business/IT goals and
xix
xx
M. Michel Raimondo
strategies, processes and operations management systems. As strategic business partner, he has a sound understanding of how technology can be used to advance business/IT strategies. He has carried out assignments in a wide variety of industries, including oil and gas, manufacturing, utilities, engineering, mining, public transportation, healthcare, and insurance. In addition to his consulting roles, he has held several leadership positions in information systems departments as CIO (15 years). He presents at conferences and facilitates workshops. Always keeping the big picture in mind, he assesses situations quickly and set forth the best recommended solution. He provides clients with strategic plans and road maps, overseeing implementations of sustainable, successful, and repeatable business process improvement programs, business transformation and organizational technology-driven change programs. He recognizes and serves the best interest of my clients providing professional advices and/or engagements with independence, objectivity, and uncompromised intellectual integrity.
Traffic Flow Theory and Its Applications Hamid Ez-Zahraouy Faculty of Science, Mohammed V University, Rabat, Morocco
Abstract. Traffic flow theory plays an interesting role in the comprehension and the development of several phenomena observed in different domains. Recently, it attracts the attention of many theoretical and empirical studies from different disciplines, especially the behavior of traffic in complex networks, such as vehicular traffic, packet of information in computer networks, airplane traffic in airports and so on. Although the great number of these studies, there exists still other serious problems to solve; in particular, the congestion in Internet communication and vehicular urban areas around the world, which have increased to the point where improving infrastructure is no longer an effective and economical solution to this problem. In this context, my talk focused on the modeling and simulation of vehicular traffic in road networks, traffic of packet information and virus spreading in computer networks, using numerical cellular automata methods. In the case of vehicular traffic as well as pedestrians, we have shown that the implementation of Intelligent Transportation Systems (ITS) based on the integrated use of all current technology to manage road traffic in a more efficient and effective way could be a good solution to reduce traffic congestion, increase the road safety rate and make maximum use of existing infrastructure for more efficient, more economical and more profitable traffic. In the case of packet information, we showed that fluidity and the virus spreading depend strongly on the network topology and/or the routing protocol governing the emission and reception of the information between adjacent nodes of the network. One of the most known and the nature of many networks is scale-free, such as the Barabási and Albert. However, the most important problem lies at the level to find the efficient routing protocol for reducing the congestion and enhance the performance of the traffic without losing sight of technical and monetary costs and also to achieve a high safety level to protect network and information spreading across it without affecting the network capacity and traffic fluidity. Beside we studied the effectiveness of local routing protocols and their additional algorithms; next-nearest neighbors with restrictive queue-length algorithm in terms of robustness in computer virus spreading. It is found that the local
xxi
xxii
H. Ez-Zahraouy
routing protocols become highly secured and overcome surprisingly the efficient path routing protocol. These results could be very helpful for network routing protocols designers to give more attention from a side to prioritization traffic models to deal more with real network protocols. Furthermore, we have studied information traffic on complex networks (such as the internet) and mainly routing rules that allow traffic to be redirected and balanced in the network. In particular, our “Self Avoiding Paths Routing” (SAPR) routing algorithm constructs the most possible disjoint paths in order to convey the traffic to and with a minimum of delay. The results obtained show an improvement over other known static algorithms. We have also generalized this same algorithm to take into account cases where destinations are more stressed than others and source nodes are more “demanding” from others. On the other hand, we have proposed a hybrid algorithm that benefits at the same time in terms of the cost of implementation of a standard shortest path algorithm and the advantages in terms of efficiency of a dynamic algorithm (Global Dynamics). We have found that we need to implement the latter only on a fraction of the routers to obtain an efficiency equivalent to that, which would have been obtained with a 100% dynamic algorithm. For the virus spreading aspect, we have shown that nodes with a large number of links are very susceptible to infection compared to other nodes in a network that evolves like the scale-free network of Barabasi-Albert but with a ceiling on the maximum degree that can have a given node called “Restricted Scale-Free.” This kind of network exhibits the existence of a maximum K_c threshold below which the network shows great resistance to the spread of the infection. This phenomenon is all the more obvious if the network is not very broad. Beside, we have shown that high-performance routing algorithms that are beneficial to traffic flow have a high sensitivity to virus propagation compared to less efficient algorithms such as the shortest path and a selective vaccination at the same time could reduce the spread of the infection without affecting the efficiency of the routing procedure in reducing congestion. Biography. Prof. Hamid Ez-Zahraouy received his Doctor of state (Statistical Physics); 1994 Faculty of Sciences, Mohammed V University Rabat. He is a researcher-professor in the Faculty of Sciences Rabat, since September 1995; the founder and coordinator of the Master Computational Physics; the founder member and coordinator of the Bachelor Computational Physics “Physique Informatique;” the founder member of the Laboratory of Magnetism and High Energy Physics (LMPHE); the founder and actually the director of the Laboratory of Condensed Matter and Interdisciplinary Sciences (LaMCScI); and actually, the head of the Physics Department at the Faculty of
Traffic Flow Theory and Its Applications
xxiii
Sciences, University Mohammed V in Rabat. Regular Associate Member to the International Center For Theoretical Physics (ICTP), Trieste, Italy, during the period 1996 to 2011; Associate researcher with CNRS-Grenoble university (2005); Expert evaluator with the CNRST-Morocco during the period 2011– 2014; Expert evaluator with more than 20 scientific journals; Guest Editor of the Special Issue “Material Science for green Energy” with superlattices and Microstructure Journal; Editor with Moroccan Journal of Condensed Matter; Advisory Board Editor with the International journal “Open Surface Science Journal.”
Smart Airport: A Major Component of a Smart City Sami El Moukhlis Mohammed Vi International Academy of Civil Aviation, Casablanca, Morocco
Abstract. Airport is an important key player in the development of the city. Adopting the smart approach will enhance this role and promote the region by enabling the connectivity and the access. In this context, my presentation is about the smart airport where we introduce the definition and the historical evolution from a simple airfield used to landing and take-off to a complex building where interacting different activities and stakeholders. I also explain the involvement of the new technologies such as IoT, cloud computing, block chain, 5G in the three major activities of the airport: passenger’s activities, airport operation and retail activities. Then I describe the passenger experience inside the terminal and the journey from the gate to gate is described and I show that the smart airport could make this experience easy and enjoyable. The airport operations are described and all the stakeholders involved in this process are introduced, the smart aspect in this activity consists in sharing real-time data, enhancing security and optimizing the time processing. ACDM (Airport Collaborative Decision Making) is a EUROCONTROL tool which use to manage all these aspects. Finally, I speak about the Research and Development and I show to the scientific community the different topics in this domain. Biography. Prof Sami El Moukhliss received his PhD on Data processing, he started his career in 2002 in ONDA (Moroccan Air Navigation Service Provider) as an ATC (Air traffic controller) where he has been promoted to an ATC instructor, then he joint the INTERNATIONAL ACADEMY OF CIVIL AVIATION (AIAC) head of ATC training in 2014, and two years later he was nominated head of the Management and Quality Division.
xxv
Contents
Smart Citizenship General Smart City Experts’ Perceptions of Citizen Participation: A Questionnaire Survey . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Jihane Tadili and Hakima Fasly
3
Citizen Sentiment Analysis in Social Media Moroccan Dialect as Case Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Monir Dahbi, Rachid Saadane, and Samir Mbarki
16
Social Networks Fake Profiles Detection Using Machine Learning Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Yasyn Elyusufi, Zakaria Elyusufi, and M’hamed Ait Kbir
30
Smart City Mobile Apps Exploratory Data Analysis: Mauritius Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Nawaz Mohamudally and Sandhya Armoogum
41
Gamification of Civic Engagement in Smart Cities (New Russian Practices) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Olga Sergeyeva, Elena Bogomiagkova, Ekaterina Orekh, and Natalia Kolesnik
53
Machine Learning for Sentiment Analysis: A Survey . . . . . . . . . . . . . . . Zineb Nassr, Nawal Sael, and Faouzia Benabbou
63
A Survey on Hand Modalities and Hand Multibiometric Systems . . . . . Farah Bahmed and Madani Ould Mammar
73
A Social Media Ontology-Based Sentiment Analysis and Community Detection Framework: Brexit Case Study . . . . . . . . . . . . . . . . . . . . . . . Moudhich Ihab, Loukili Soumaya, Bahra Mohamed, Hmami Haytam, and Fennan Abdelhadi
89
xxvii
xxviii
Contents
Fuzzy Questions for Relational Systems . . . . . . . . . . . . . . . . . . . . . . . . . 104 Rachid Mama and Mustapha Machkour Smart Education Using Machine Learning Algorithms to Predict the E-orientation Systems Acceptancy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117 Rachida Ihya, Mohammed Aitdaoud, Abdelwahed Namir, Fatima Zahra Guerss, and Hajar Haddani Smart University Services for Collaborative Learning . . . . . . . . . . . . . . 131 Ouidad Akhrif, Chaymae Benfares, Younés El Bouzekri El Idrissi, and Nabil Hmina Multi-Agent System of an Adaptive Learning Hypermedia Based on Incremental Hybrid Case-Based Reasoning . . . . . . . . . . . . . . 143 Nihad El Ghouch, Mohamed Kouissi, and El Mokhtar En-Naimi Augmented Reality Application in Laboratories and Learning Procedures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157 Anasse Hanafi, Lotfi Elaachak, and Mohamed Bouhorma Pedagogy in the Digital Age: Making Learning Effective and Engaging for Students . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168 Soukaina Oulaich Towards Recommendation Using Learners’ Interest in Social Learning Environment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183 Mahnane Lamia, Mohamed Hafidi, and Samira Aouidi Toward a Generic Student Profile Model . . . . . . . . . . . . . . . . . . . . . . . . 200 Touria Hamim, Faouzia Benabbou, and Nawal Sael Digital Business and Smart Governance Digital Business Models: Doing Business in the Digital Era . . . . . . . . . . 217 Wail El Hilali and Abdellah El Manouar Social Media Made Me Buy It: The Impact of Social Media on Consumer Purchasing Behavior and on the Purchase Decision-Making Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230 Bedraoui Oumayma Machine Learning as an Efficient Tool to Support Marketing Decision-Making . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244 Redouan Abakouy, El Mokhtar En-Naimi, Anass El Haddadi, and Lotfi Elaachak
Contents
xxix
Big Data Science and Analytics for Tackling Smart Sustainable Urbanism Complexities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259 Simon Elias Bibri, John Krogstie, and Nesrine Gouttaya Towards an Eco Responsible Corporate Governance . . . . . . . . . . . . . . . 275 Oumaima Riad, Sahar Saoud, Lamia Boukaya, and Khalid Azami Comparative Study Using Neural Networks Techniques for Credit Card Fraud Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287 Imane Sadgali, Nawal Sael, and Faouzia Benabbou Smart Healthcare Application of Unsupervised Machine Learning to Cluster the Population Covered by Health Insurance . . . . . . . . . . . . . . . . . . . . . 299 Sara Zahi and Boujemâa Achchab Secure Communication in Ehealth Care Based IoT . . . . . . . . . . . . . . . . 311 Somaya Haiba and Tomadar Mazri Internet of Things Ehealth Ecosystem: Solution . . . . . . . . . . . . . . . . . . . 324 Hamza Zemrane, Youssef Baddi, and Abderrahim Hasbi Risks of Radiation Exposure During Abdominopelvic CT Procedures at Mohamed VI University Hospital of Oujda, Morocco . . . . . . . . . . . . 339 Mohammed Aabid, Slimane Semghouli, Oum Keltoum Hakam, and Abdelmajid Choukri Glucose Sensing for Diabetes Monitoring: From Invasive to Wearable Device . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 350 Loubna Chhiba, Basma Zaher, Mustapha Sidqui, and Abdelaziz Marzak Framework on Mobile Technology Utilization for Assisted Healthcare Service Request and Delivery of Aged Person: A Case of Ghana . . . . . . 365 Israel Edem Agbehadji, Richard C. Millham, Abdultaofeek Abayomi, Ekua Andowa Biney, and Kwabena Obiri Yeboah Segmentation and Classification of Microcalcifications Using Digital Mammograms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 384 Ichrak Khoulqi and Najlae Idrissi Rough Set Based Supervised Machine Learning Approaches: Survey and Application . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 404 Abdelkhalek Hadrani, Karim Guennoun, Rachid Saadane, and Mohammed Wahbi
xxx
Contents
Knee Functional Telerehabilitation System for Inclusive Smart Cities Based on Assistive IoT Technologies . . . . . . . . . . . . . . . . . . . . . . . . . . . 425 Mohamed El Fezazi, Mounaim Aqil, Atman Jbari, and Abdelilah Jilbab Configuring MediBoard HIS for Usability in Hospital Procedures . . . . . 440 Youssef Bouidi, Mostafa Azzouzi Idrissi, and Noureddine Rais Type 2 Diabetes Mellitus Prediction Model Based on Machine Learning Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 454 Othmane Daanouni, Bouchaib Cherradi, and Amal Tmiri Hybrid Method for Breast Cancer Diagnosis Using Voting Technique and Three Classifiers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 470 Hajar Saoud, Abderrahim Ghadi, and Mohamed Ghailani Visual Question Answering System for Identifying Medical Images Attributes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 483 Afrae Bghiel, Yousra Dahdouh, Imane Allaouzi, Mohamed Ben Ahmed, and Abdelhakim Anouar Boudhir New Generation of Networks and Systems for Smart Cities Enhanced Mobile Network Stability Using Average Spatial Dependency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 495 Halim Berradi, Ahmed Habbani, Chaimae Benjbara, Nada Mouchfiq, and Mohammed Souidi A Review on 3D Reconstruction Techniques from 2D Images . . . . . . . . 510 M. Aharchi and M. Ait Kbir Analytical Approaches and Use Case on Network Interactions . . . . . . . 523 Hamza Hanafi, Badr Dine Rossi Hassani, and M’hamed Aït Kbir Performance Analysis of Cooperative MIMO Systems for Mobile Coverage in Smart City Application . . . . . . . . . . . . . . . . . . . . . . . . . . . 531 Richard Musabe, Said Rutabayiro Ngoga, Emmanuel Manizabayo, Vienna N. Katambire, and Yaramba Hadelin Embedded Systems Hardware Software Partitioning Approach Based on Game Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 542 Adil Iguider, Kaouthar Bousselam, Oussama Elissati, Mouhcine Chami, and Abdeslam En-Nouaary Comparative Study and Improvement of the Routing Protocols Used in the Vehicular Networks and v2v Communications . . . . . . . . . . . . . . . 556 Kawtar Jellid and Tomader Mazri
Contents
xxxi
Performance Study of Position-Based Zoning Techniques in MANETs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 572 Mohammed Souidi, Ahmed Habbani, Chaimae Benjbara, and Halim Berradi Logical Structure of an IPv6 Network that Perfectly Uses the Summarization Technique . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 590 Izem Acia, Wakrim Mohamed, and Ghadi Abderrahim A Simulation Analyses of MANET’s Attacks Against OLSR Protocol with ns-3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 605 Oussama Sbai and Mohamed Elboukhari Functional Modeling of IoT Protocols . . . . . . . . . . . . . . . . . . . . . . . . . . 619 Sakina Elhadi, Abdelaziz Marzak, and Nawal Sael New Tool for 5G Radio Network Planning and Deployment . . . . . . . . . 634 Lamiae Squali, Jacob Chabi Aloh, and Fatima Riouch Transmit-Power and Interference Control Algorithm in Cognitive Radio Network Based on Non-cooperative Game Theory . . . . . . . . . . . . 647 Mohammed Saber, Abdessamad El Rharras, Rachid Saadane, Hatim Kharraz Aroussi, and Mohammed Wahbi Network Coding for Energy Optimization of SWIMAC in Smart Cities Using WSN Based on IR-UWB . . . . . . . . . . . . . . . . . . . . . . . . . . 663 Anouar Darif, Hasna Chaibi, and Rachid Saadane 5G Energy Efficiency for Smart Cities: A Call Admission Control Proposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 675 Ahmed Slalmi, Rachid Saadane, H. Chaibi, and Hatim Kharraz Aroussi Watermarking Image Scheme Based on Image Content and Corners Decomposition Method . . . . . . . . . . . . . . . . . . . . . . . . . . . 693 Abdelhay Hassani Allaf and M’hamed Aït Kbir Smart Grids and Electrical Engineering A Combined Source and Demand-Side Energy Management System for a Grid-Connected PV-Wind Hybrid System . . . . . . . . . . . . . . . . . . . 707 Asmae Chakir, Mohamed Tabaa, Fouad Moutaouakkil, Hicham Medromi, and Karim Alami New Hybrid Approach Multi-agents System and Case Based Reasoning for Management of Common Renewable Resources . . . . . . . 722 Mohamed Kouissi, Nihad El Ghouch, and El Mokhtar En-Naimi
xxxii
Contents
Robust Control of Induction Motor Drive Facing a Large Scale Rotor Resistance Variation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 736 Yassine Zahraoui, Mohamed Akherraz, Chaymae Fahassa, and Sara Elbadaoui Optimization of the Feasibility Study of an Energy Production System Based on the Wind and the Marine Current Turbine in Morocco . . . . . 750 Rajae Gaamouche, Prince Acouetey, Abdelbari Redouane, and Abdennebi El Hasnaoui Identification of Relevant Input Variables for Prediction of Output PV Power Using Artificial Neural Network Models . . . . . . . . . . . . . . . . . . . 766 Elmehdi Karami, Mohamed Rafi, and Abderraouf Ridah Smart Mobility On the Communication Strategies in Heterogeneous Internet of Vehicles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 783 Nadjet Azzaoui, Ahmed Korichi, Bouziane Brik, Med el amine Fekair, and Chaker Abdelaziz Kerrache A Comparison of Random Forest Methods for Solving the Problem of Pulsar Search . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 796 Mourad Azhari, Altaf Alaoui, Abdallah Abarda, Badia Ettaki, and Jamal Zerouaoui Digital Service Supply Chain Management: Current Realities and Prospective Visions . . . . . . . . . . . . . . . . . . . . . . . 808 Badr Bentalha, Aziz Hmioui, and Lhoussaine Alla Providing Context Awareness in the Smart Car Environment: State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 823 Abdelfettah Soultana, Faouzia Benabbou, and Nawal Sael Applying External Guidance Commands to Deep Reinforcement Learning for Autonomous Driving . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 837 Fenjiro Youssef and Benbrahim Houda Image Correlation Based Smart Throttle-Brake Control System for Disability Vehicles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 853 G. Lavanya, M. Deva Priya, A. Christy Jeba Malar, T. Sangeetha, and A. Saravanan A Multi-layer System for Maritime Container Terminal Management Using Internet of Things and Big Data Technologies . . . . . . . . . . . . . . . 865 Farah Al Kaderi, Rim Koulali, and Mohamed Rida
Contents
xxxiii
Meteorological Parameters Prediction Along Roads Between Two Cities for the Safest Itinerary Selection . . . . . . . . . . . . . . . . . . . . . . . . . 878 Samir Allach, Badr Benamrou, Mohamed Ben Ahmed, Anouar Abdelhakim Boudhir, and Mustapha Ouardouz Vehicle Traffic Management with the Help of Big Data Technologies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 894 Mouad Tantaoui, My Driss Laanaoui, and Mustapha Kabil Smartwatch-Based Wearable and Usable System for Driver Drowsiness Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 906 Mohammed Misbhauddin Smart Life Saving Navigation System for Emergency Vehicles . . . . . . . 921 M. Deva Priya, A. Christy Jeba Malar, G. Lavanya, L. R. Vishnu Varthan, and A. Balamurugan Smart-Logistics for Smart-Cities: A Literature Review . . . . . . . . . . . . . 933 Chouar Abdelssamad, Tetouani Samir, Lmariouh Jamal, Soulhi Aziz, and Elalami Jamila Smart Security Towards a Holistic Privacy Preserving Approach in a Smart City Environment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 947 Driss El Majdoubi and Hanan El Bakkali A Hybrid Intrusion Detection System Against Egoistic and Malicious Nodes in VANET . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 961 Meriem Houmer and Moulay Lahcen Hasnaoui A New Secure Schema to Enhance Service Availability in Urban IoT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 974 Fatna El Mahdi, Halim Berradi, Ahmed Habbani, and Bachir Bouamoud Evaluation of Deep Learning Approaches for Intrusion Detection System in MANET . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 986 Safaa Laqtib, Khalid El Yassini, and Moulay Lahcen Hasnaoui Recommendation Enhancement Using Traceability and Machine Learning: A Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 999 Kamal Souali, Othmane Rahmaoui, and Mohammed Ouzzif Smart Incident Management, Prediction Engine and Performance Enhancement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1011 Jamal El Abdelkhalki and Mohamed Ben Ahmed A Proposed Architecture Based on CNN for Feature Selection and Classification of Android Malwares . . . . . . . . . . . . . . . . . . . . . . . . . 1026 Soussi Ilham, Ghadi Abderrahim, and Boudhir Anouar Abdelhakim
xxxiv
Contents
Classification of Grayscale Malware Images Using the K-Nearest Neighbor Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1038 Ikram Ben Abdel Ouahab, Mohammed Bouhorma, Anouar Abdelhakim Boudhir, and Lotfi El Aachak Improving Recommendations Using Traceability and Machine Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1051 Kamal Souali, Othmane Rahmaoui, and Mohammed Ouzzif Sustainable Building Bioclimatic Approach and Appropriate Financial Governance to Promote Sustainable Building Design . . . . . . . . . . . . . . . . . . . . . . . . . 1065 Najoua Loudyi and Khalid EL Harrouni Contribution to a Smart Home Design for Medical Surveillance . . . . . . 1080 Eloutouate Lamiae, Elouaai Fatiha, Bouhorma Mohammed, and Gibet Tani Hicham Shading Devices’ Benefits on Thermal Comfort and Energy Performance of a Residential Building in Different Climates in Morocco . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1094 Badr Chegari, Mohamed Tabaa, Fouad Moutaouakkil, Emmanuel Simeu, and Hicham Medromi Conjugate Natural Convection-Surface Radiation in a Square Cavity with an Inner Elliptic Body . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1112 Lahcen El Moutaouakil, Mohammed Boukendil, Zaki Zrikem, and Abdelhalim Abdelbaki Well-Being Observing Framework in Smart Home . . . . . . . . . . . . . . . . 1128 Naoufal Ainane, Mohamed Ouzzif, and Khalid Bouragba Sustainable Environment How to Improve Wastewater Treatment in Smart City . . . . . . . . . . . . . 1141 Aziz Taouraout, Abdelkader Chahlaoui, Driss Belghyti, Mohamed Najy, and Rachid Sammoudi Performance of Aluminum and Iron-Based Coagulants for the Removal of Water Turbidity for Human Consumption in the Cities (Rabat and Casablanca) of Morocco and Dewaterability of Hydroxide Sludge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1154 Mohamed Najy, Mohamed Lachhab, Aziz Taouraout, Mohamed El Qryefy, and Driss Belghyti
Contents
xxxv
Improving Safe and Sustainable Gray Water Reuse: A New Solution to Curb Water Shortages in Moroccan Cities . . . . . . . . . . . . . . . . . . . . 1167 Aziz Taouraout, Abdelkader Chahlaoui, Driss Belghyti, Imane Taha, and Khadija Ouarrak Historical Weather Data Recovery and Estimation . . . . . . . . . . . . . . . . 1179 Fadoua Rafii and Tahar Kechadi The Contribution of Cartography in Risk Management of Vector-Borne Diseases: Cas of Leishmaniasis in the Fez-Meknes, Region of Morocco . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1192 H. El Omari, A. Chahlaoui, F. Talbi, K. Ouarrak, and A. El Ouali Lalami Spatial Relation Among Incidence of Leishmaniasis and Altitude Factor of Different Communes of Sefrou Province: Contribution of Geographic Information Systems . . . . . . . . . . . . . . . . . 1202 Fatima Zahra Talbi, Amal Sbai, Hajar El Omari, Mohamed Najy, and Abdelhakim El Ouali Lalami Seasonal Variations of the Microbiological Parameters of the Quality of Water in Urban Oued Bouishak of the City of Meknes (Morocco) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1211 Khadija Ouarrak, Addelkader Chahlaoui, Imane Taha, Aziz Taouraout, and Adel Kharroubi Seasonal Variation of Parasitic Content of Wastewater Discharging in Boufekrane River at the Collector of the Agdal District (City of Meknes, Morocco) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1222 Imane Taha, Abdelkader Chahlaoui, Aziz Taouraout, Khadija Ouarrak, and Rachid Sammoudi Typology of the Surface Water Quality of the Aguelmam Sidi Ali Wetland (Midelt-Morocco) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1235 Rachid Sammoudi, Abdelkader Chahlaoui, Adel Kharoubi, Imane Taha, and Aziz Taouraout Geoinformatics Approach to Water Allocation Planning and Prognostic Scenarios Sustainability: Case Study of Lower Benue River Basin, Nigeria . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1249 Zainab Abdulmalik, Adebayo Wahab Salami, Solomon Olakunle Bilewu, Ayanniyi Mufutau Ayanshola, Oseni Taiwo Amoo, Abayomi Abdultaofeek, and Israel Edem Agbehadji Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1263
Smart Citizenship
General Smart City Experts’ Perceptions of Citizen Participation: A Questionnaire Survey Jihane Tadili(&) and Hakima Fasly Strategic Intelligence Research Laboratory, Hassan II University of Casablanca, Casablanca, Morocco [email protected], [email protected]
Abstract. Smart cities gained the support of scientists, urban planners, and governments all over the world because they suggest innovative solutions for all urban development problems using Information and communication technologies (ICT). Citizen participation is the key challenge to develop a smart city project since the main objective of a smart city is to improve the quality of life of citizens. Thus, decision-makers should cooperate with citizens and stakeholders. In this article, we will explain the current state of the art in the process of empowering citizens within smart cities and detail the results of a survey conducted in the frame of the Smart City Expo event held in Casablanca in April 2018. We administered the survey to key stakeholders in smart cities spread all over the world like city council representatives, technology developers and scientists (n = 20 respondents) in order to evaluate citizen participation in a smart city in practice. Keywords: Smart city
Citizen participation Literature review Survey
1 Introduction Urbanization is a factor of economic growth; cities produce 80% of the Gross National Product (GNP) worldwide [2]. According to the United Nations Population Fund, half of the world’s population lives in cities, this figure will reach about 5 billion by 2030 [2]. Moreover, cities consume about 75% of the global primary energy and generate between 50 and 60% of global greenhouse gas emissions [3]. In this difficult context, smart cities have reached unprecedented progress and gained the support of scientists, urban planners, and governments all over the world, because they suggest innovative solutions for all urban development problems using Information and communication technologies (ICT) [1]. Citizen participation is the key challenge to develop a smart city project since the main objective of a smart city is to improve the quality of life of citizens. However, there are no recent studies about practices, barriers and new ways to increase civic participation in smart cities. In order to contribute to reducing this research gap, we will © Springer Nature Switzerland AG 2020 M. Ben Ahmed et al. (Eds.): SCA 2019, LNITI, pp. 3–15, 2020. https://doi.org/10.1007/978-3-030-37629-1_1
4
J. Tadili and H. Fasly
try to answer the following research question: “How to enhance citizen participation in a smart city project?” To this end, we will represent in this article the citizen participation theory and a literature review of smart city concept and then discuss the role of citizen participation in smart cities, we will also report the results of a survey administered to smart city experts from different cities in the world (n = 20 respondents).
2 Citizen Participation Theory Top-down decision-making approach has shown its ineffectiveness in most democracies, citizens become more demanding and aware of their rights [4, 5]. Actually, there is a big gap between citizens’ aspirations and government politics, which has led to social movements and declining trust in public institutions [6]. In this democratic deficit context as labeled by Hindess [7], citizen participation has appeared as an alternative to break through this deadlock and give an active role to citizens. Those citizens, who are often excluded from the government’s decision-making process and are only considered as elections voters, can act also as experts and information providers for their communities [8]. First scientific researches on citizen participation in public decision making were started in the late 1960s in a period marked by urban struggles and students’ protest movement, from which came the first demands for “participatory democracy” [9]. Thus, the famous article «A Ladder of Citizen Participation» of Sherry Arnstein come out in 1960. Arnstein 1960 suggested a ladder for participation levels starting from manipulation to citizen control with each rung corresponding with the extent to which citizens can determine the product. As it is argued, citizen participation is a power redistribution that allows the disadvantaged citizen to be deliberately involved and consulted in urban planning, information diffusion, political programs, and resource allocation. However, it can take a “tokenism” form being just a consultation directed by public administrations to allow citizens to hear and to have a voice [10]. Scientists and experts have also studied the objectives, functions, advantages, and inconveniences of participation, and developed many models and frameworks to categorize and understand citizen participation. Bishop and Davis define participation as “the expectation that citizens have a voice in policy choices” and categorize it according to policy objectives (Table 1). In concrete terms, the authors provide a map of policy participation [6].
General Smart City Experts’ Perceptions of Citizen Participation
5
Table 1. Adapted from Bishop and Davis, 2002
Maximum participation
Type Control Consumer choice Standing Partnership Consultation Information
Minimum participation
Objective Giving the control of an issue to the electorate Customer preferences shaping a product through the choice of products Involving third parties in the review process Involving citizens in some aspects of decision making Evaluate community reactions and feedbacks Giving some information by the decision makers
According to Irvin and Stansbury, citizen participation in the decision-making process has many advantages like educational benefits for citizens and a better understanding of community expectations by administrators, political suasion, citizen empowerment and avoiding litigation costs. However, involving citizens is not easy and might be costly and time-consuming. Thereby, technology can be the best solution to reduce cost and help along with the process implementation. In this case, the smart city concept can be a suitable ground for citizen participation [11].
3 The Smart City Concept Ecological, demographical, economical or spatial problems need smarter approaches to be solved. With the increase in population and rapid urbanization, we need smarter solutions that help us to create sustainable cities. The term “smart city” has not been defined recently. The term has been generated from the “smart growth” movement in the nineties, which supports community-driven solutions to solve urban problems [12]. Besides, the protocol of Kyoto in 1997 signed by 192 parties to reduce CO2 emissions has also generated interest in Smart City as an innovative solution to achieve the protocol’s goal [21]. Therefore, several institutions adopted this concept (the European Commission, Setis-EU, OECD, and the California Institute for Smart Communities) and private firms (IBM, Cisco, Siemens). Indeed, IBM was the first firm to identify cities as a potential market by combining them with information and communication technologies (ICT) and thereafter promoted the Smart City concept to cities, enabling this one to gain popularity [30]. There is no consensus among researchers about a smart city definition, key elements, and boundaries. As the concept is used all over the world with different names
6
J. Tadili and H. Fasly
and different contexts, there are numerous terms similar to “smart cities” such as intelligent city, knowledge city, wired city, digital city, and so on [12–15]. Over the last decade, the term “Smart City” has grown more than its analogues including “Intelligent City” and “Digital City” for the following reasons: • From a Marketing perspective, the word “Smart” is used more than the adjective “Intelligent” because it focuses on the needs of users. In order to attract a wider base of community members, “Smart” serves better than “intelligent” who is more elitist. “Smart” is also more user-friendly than “intelligent”. Smart is needed to adapt to user needs and provide custom interfaces [12]. Thus, a city can become smart if it adapts to the needs of its inhabitants. • [31] focused on the different ways of using the term “Smart”. More concretely, if the word “Intelligent” clearly implies a kind of positive technological innovation based on the city and the change via ICT, the adjective “Smart” has also been used in relation to governance, communities, social learning and to address issues of urban growth and social and environmental sustainability. • In 2007, the Apple Company launched and marketed the smartphone “iPhone” which democratized the daily use of Smart devices. The success of the term Smart in mobile telephony has influenced the adoption of this term in urban context [13]. In his literature review, A. Cocchia distinguished smart from the digital city and tried to define the “smart city” according to two different approaches: • Their contents: the digital city regards the use of ICT in urban areas; the smart city regards the attention to be paid to the environmental quality in cities; • Their nature and relationship with the government: the digital city is a free trend emerging from the daily use of smart and digital devices by citizens. It incites the local governments to supply e-services, that is, to transform gradually the city into a digital city; smart city is a political trend, driven by international institutions, to implement adequate initiatives to improve the environmental quality in cities [13]. Giffinger et al. identified economy, mobility, people, governance, environment and living as the main factors of the smart city: “A city well performing in a forwardlooking way in economy, people, governance, mobility, environment, and living, built on the smart combination of endowments and activities of self-decisive, independent and aware citizens. Smart city generally refers to the search and identification of intelligent solutions which allow modern cities to enhance the quality of the services provided to citizens.” [16]. The Smart City Council highlighted the smart city in the use of technology across all city functions [17]. Some definitions argue the use of technologies, the private firm IBM was the first to introduce the term “Smart city” and defined three main characteristics for it: instrumented, intelligent and interconnected. Other definitions stressed different aspects like natural resources management [18, 19]. Other scientists claimed that Smart cities must undertake open access and inclusive strategies to diminish the digital dividend [20] and consider that a successful smart city should be inclusive and must concern all citizens including people with disabilities. Therefore, the new term “Inclusive Smart City” is defined as “A city that uses digital assistive technology in the urban spaces in order to enhance the experience that people
General Smart City Experts’ Perceptions of Citizen Participation
7
with disabilities have in these spaces, extending to a considerable number of citizens the gain envisaged by SC initiatives” [20]. Nam & Pardo organized the definitions according to three dimensions depending on some recurrent characteristics: • Technology dimension: The use of urban information and communication technology (ICT) infrastructure to improve quality of life is the main driver for the smart city. The concepts included in this dimension are Virtual City; Digital City, Wired City, Information City, Intelligent City, and Ubiquitous City; • Human dimension: It includes human capital and it binds smart city to education, citizens, learning and knowledge. This dimension includes the concepts about Learning City and Knowledge City; • Institutional dimension: The combination of technological and human dimensions is very important but without cooperation between stakeholders and institutional governments, it will be difficult to develop a smart city project. This dimension concerns the concepts about Green City, Smart Community and Sustainable City [12].
4 Citizen Participation in Smart Cities Citizen participation is the key challenge to develop a smart city project since the main objective of a smart city is to improve the quality of life of citizens. Thus, all actions should be done in cooperation with citizens and stakeholders. Otherwise, the smart city will not achieve its objectives. In fact, ICT technologies are certainly the main driver of any smart city project, but without considering people, technologies and strategic visions will not create public value for citizens [21]. Hollands 2008 criticized the smart city approach based on the use of ICT only, and argued that Smart cities should start with people and human capital of the city: “Progressive smart cities must seriously start with people and the human capital side of the equation, rather than blindly believing that IT itself can automatically transform and improve cities.” ICT should be adapted to citizen needs and used to empower and educate them in order to create a smart community in place of a smart city. Cities with more educated populations experience more rapid growth [22]. Actually, the most used and accepted definitions of smart city take into account this critique plus the other dimensions: “A city to be smart when investments in human and social capital and traditional (transport) and modern (ICT) communication infrastructure fuel sustainable economic growth and a high quality of life, with a wise management of natural resources, through participatory governance” [26]. Many scientists acknowledge citizen participation as the main component for smart city projects; decision-makers should innovate in interactions with citizens [23]. According to Berntzen & Johannessen (2015), Citizens can be involved as: • Democratic participants in the decision making process and build sustainable local communities where every inhabitant cares for the other.
8
J. Tadili and H. Fasly
• Main source of experience and competencies in order to develop better solutions and plans. Thanks to competent citizens, problems can be avoided early and the risk of failure decreases. • Data collectors; The citizens can help to collect data after the implementation of the smart city as data collectors by using mobile devices or other technologies, in this way, they will feel like an active and integral part of the smart city [24]. Other researches argue that the Quadruple Helix Model for innovation should be followed to create an alliance between the four pillars of innovation: university, government, industry, and civil society and involve them in the early innovation of the smart city process in order to develop new ideas that can meet social needs. The civil society is introduced as an important stakeholder, which has updated the classical triple helix model [25–27].
5 Survey About Citizen Participation in Smart Cities 5.1
Methodology
In order to compare cities initiatives for the citizen participation in smart cities projects and identify smart practices and new ways to increase civic participation, we have administered a survey to smart city experts from different cities in the world (n = 37 interviewed): Amsterdam, Chicago, Boston, Washington, Barcelona, Toronto, Mulhouse, London, Lethbridge, Coimbatore, Casablanca and Palo Alto. We got responses from 20 experts. We administered the survey during the Smart City Expo in April 2018 in Casablanca, which helped us to meet respondents that are participating in a smart city projects development in their countries and serving in important positions like Mayor, strategy advisor - international smart city ambassador; Future Cities Council President, Ph.D. students, Professors; Intelligent Community Program Manager; CIO and Consultants. 5.2
Survey Highlights
This survey gives insights about smart city projects and citizen participation in different cities. Key topics explored include city priorities for smart city and citizen participation, city inclusiveness, budget allowed to citizen participation, Practices used to enhance citizen participation, citizen participation barriers. Smart City Priorities: How important are each of the following objectives for smart city project in your country? (Check one box for each row.) (Fig. 1).
General Smart City Experts’ Perceptions of Citizen Participation
9
[Open data (within and across city responsibility areas)] [Smart payments and finance (web-based payment for services, etc.)] [Civic engagement] [Public safety (police body cameras, streetlights with gunshot detection, etc.)] [Health, education, and human services] [Waste management (sensors for waste containers, etc.)] [Water and wastewater (smart meters, automated leak detection, etc.)] [Transportation (mobility apps supporting multiple travel modes, electric vehicle charging stations, etc.)] [Telecommunications (public WiFi, interoperable systems, etc.)] [Energy (smart meters, renewable energy, etc.)] [Built environment (building management systems, streetlights with WiFi, or other services, etc.)] 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Very Important
Important
Moderately Important
Less Important
Not Important
Fig. 1. Smart city priorities
Respondents most frequently identified smart city technologies as a priority in civic engagement and transportation, with 71% of respondents identifying these initiatives as a top priority in this area. Health, education and human services (65%), telecommunications (59%), public safety (59%), built environment (53%) were also among the top five sectors in which smart-city technologies were identified as a top priority by respondents. Other respondents chose to add other priorities and qualify them as very important like cybersecurity, big data for urban informatics, embedded computing, and networks for smart cities, regulation, public facilities and an open-air ground for people, interoperable data rather than open data. Citizen Participation Priorities: How important are each of the following benefits in motivating your local government to implement or expand the citizen participation in a smart city project? (Check one box for each row.) (Fig. 2). Enhanced services for residents and open/sharing data was identified as the most important benefits from citizen participation in a smart city project by the respondents, Economic development and administrative efficiencies were also among the top five benefits.
10
J. Tadili and H. Fasly
[Sustainability benefits] [Safety and security benefits] [Cost savings] [Administrative efficiencies] [Economic development] [Open data/sharing data] [Enhanced services for residents ( health, social, education, services, etc)] 0%
20%
Very Important
Important
Less Important
Not Important
40%
60%
80%
100%
Moderately Important
Fig. 2. Citizen participation in Smart cities benefits
Budget Allocation to Citizen Participation: Does your local government typically allocate a certain amount of funding for civic engagement in a smart city project? (Fig. 3).
Budget Allocation
35% 53% 12% I dont know
no
yes
Fig. 3. Budget allocation
If yes, on average what percentage of the project budget is typically allocated for civic engagement?
General Smart City Experts’ Perceptions of Citizen Participation
11
Percentage of Budget Allocation
18% 36% 9% 0% 36%
0 to 1%
1 to 5%
5 to 10%
Over 10%
Other
Fig. 4. Percentage of budget allocation
Only half of the respondents think that their cities are allocating a certain amount of funding for civic engagement in a smart city project and 72% of them ranged the budget from zero to 5%, which means that even if decision-makers are aware of the importance of citizen participation, they do not give enough budget for it (Fig. 4). Practices Used to Enhance Citizen Participation: Which of these practices do you use in your city to enhance citizen’s participation in your city? Please rate it (Fig. 5).
Fig. 5. Practices used to enhance citizen participation
12
J. Tadili and H. Fasly
Among seventeen practices suggested to respondents, allowing access to city services via smartphones; reporting service problems via mobile app; being involved in the planning of the new public services; participation in political decisions concerning issues of citizenship’s interest; creation or expansion of public wireless networks, were the top five practices. This proves that smart city experts insist on the important role of citizens in the early stage of planning public services as co-creators, also as experiences providers by reporting problems. Free courses in computer literacy for youth and adults were also identified as an important practice by 47% of respondents, which highlight the smart city experts’ awareness about City inclusiveness. Citizen Participation Barriers: To what extent do each of the following issues represent barriers for your community to implement or expand the citizen engagement in a smart city project? (Check one box for each row.) (Fig. 6).
Citizen participation in smart cities barriers [Absence of public wireless network] [Limited uses of IOT and smartphones by citizens] [Digital illiteracy] [Need to gain community support] [Need more long-term vision or plan] [Difficulty of coordinating across departments] [Too reliant on legacy systems] [Need more technical expertise] [Budget limitations] [Smart city illiteracy] [Need more supportive policies] [Need better understanding of how to get started] 0% 10% 20% 30% 40% 50% 60% 70% 80% 90%100% Very Important
Important
Moderately Important
Less Important
Not Important
Fig. 6. Citizen participation in smart city barriers
Approximately 70% of responding smart city experts indicated that the need for more long-term vision or plan is the most significant citizen participation barrier. More than half of respondents (53%) agreed that smart city illiteracy and budget limitations are preventing decision-makers from involving citizens in their smart city projects. Needs for a good start, community supports, supportive policies, and coordination across departments are also very significant barriers according to 47% of respondents.
General Smart City Experts’ Perceptions of Citizen Participation
5.3
13
Summary
In summary, according to survey participants, smart city technologies are a priority in civic engagement and transportation, health, education, and human services. Enhanced services for residents and open/sharing data was identified as the most important benefits from citizen participation in a smart city project. Respondents highlighted also the need for a budget’s allocation for citizen participation in smart cities; they are more likely to involve citizens in the early stage of planning public services as co-creators, also as experiences providers by reporting problems. They are also willing to provide free courses in computer literacy for youth and adults in order to build an inclusive smart city. Based on survey results, the need for more long-term vision or plan; smart city illiteracy and Budget limitations are the most significant barriers for citizen participation in smart cities.
6 Conclusion In order to answer the research question: “How to enhance citizen participation in a smart city project?” we presented in this article a literature review about smart city, citizen participation theory and citizen participation in smart cities. Then, we evaluated citizen participation in smart city in practice among a survey administered to smart city experts from different cities in the world. The first key finding is that smart city experts are aware of the importance of citizen participation in smart cities and identified it as a priority. They are more likely to involve citizens in the early stage of planning public services as co-creators and after delivering services as experiences providers by reporting problems. We notice also an interest in the role of citizens as data collectors, indeed citizens can provide decisionmakers with important data. Thanks to this data, they can get valuable insights to provide services to many sectors in the smart city, as well as improve citizens’ experiences and create new business opportunities [28]. City Inclusiveness was also among the major concerns of respondents since they agreed that smart city illiteracy is preventing them from involving citizens in their smart city projects, they are also willing to give free courses in computer literacy for youth and adults in order to reduce the digital divide. Actually, decision-makers should create a good environment for developing a smart city for all and promoting city inclusiveness. Our study noted also a lack of long-term vision and specific budget allocation for citizen participation in smart city, which provides a significant barrier. Local government should not only act as the initiator of new policies but also the main organizer of citizen participation, knowing that citizen participation level increases when it comes to implementing a policy already agreed on by local government [29]. We acknowledge that our survey has a relatively reduced sample, however, respondents are participating in a smart city projects development in their countries and serving in important positions like Mayor, strategy advisor…etc., which can offer a first impression of what citizen participation in smart cities currently looks like. We also
14
J. Tadili and H. Fasly
believe that our findings can help both researchers and local governments further. Academically, researchers can try to address the research gap about citizen participation in smart cities and come up with smart practices to enhance citizen participation. Governments can take lessons from our survey highlights, namely, the citizen participation barriers in order to update their policies and support financially citizen participation. Acknowledgments. We would like to express our special thanks to all survey participants, and the organizers of the smart city expo in Casablanca who gave us the excellent opportunity to do this research.
References 1. Mori, K., Christodoulou, A.: Review of sustainability indices and indicators: towards a new City Sustainability Index (CSI). Environ. Impact Assess. Rev. 32(1), 94–106 (2012) 2. United Nations Population Fund (UNFPA) (2014), 10 January 2019. https://www.unfpa.org/ urbanization 3. United Nations (2018), 10 January 2019. https://www.un.org/development/desa/publications/ graphic/world-urbanization-prospects-2018-more-megacities-in-the-future 4. Grodzińska-Jurczak, M., Cent, J.: Can public participation increase nature conservation effectiveness? Innov. Eur. J. Soc. Sci. Res. 24(3), 371–378 (2011) 5. Carpini, M.X.D., Cook, F.L., Jacobs, L.R.: Public deliberation, discursive participation, and citizen engagement: a review of the empirical literature. Ann. Rev. Polit. Sci. 7, 315–344 (2004) 6. Bishop, P., Davis, G.: Mapping public participation in policy choices. Aust. J. Public Adm. 61(1), 14–29 (2002) 7. Hindess, B.: Politics and governmentality. Int. J. Hum. Resour. Manage. 26(2), 257–272 (1997) 8. Mansbridge, J.J.: Beyond Adversary Democracy. University of Chicago Press (1983) 9. Blondiaux, L., Fourniau, J.-M.: Un Bilan des Recherches sur la participation du public en démocratie: beaucoup de bruit pour rien? Participations 1, 8–35 (2011). https://doi.org/10. 3917/parti.001.0008 10. Arnstein, S.R.: A ladder of citizen participation. J. Am. Inst. Planners 35(4), 216–224 (1969) 11. Irvin, R.A., Stansbury, J.: Citizen participation in decision making: is it worth the effort? Publ. Adm. Rev. 64(1), 55–65 (2004) 12. Nam, T., Pardo, T.A.: Conceptualizing smart city with dimensions of technology, people, and institutions. In Proceedings of the 12th Annual International Digital Government Research Conference: Digital Government Innovation in Challenging Times, pp. 282–291. ACM, June 2011 13. Cocchia, A.: Smart and digital city: a systematic literature review. In: Smart City, pp. 13–43. Springer, Cham (2014) 14. Albino, V., Berardi, U., Dangelico, R.M.: Smart cities: definitions, dimensions, performance, and initiatives. J. Urban Technol. 22(1), 3–21 (2015) 15. Chourabi, H., Nam, T., Walker, S., Gil-Garcia, J.R., Mellouli, S., Nahon, K., Scholl, H.J.: Understanding smart cities: an integrative framework. In: 2012 45th Hawaii International Conference on System Science (HICSS), pp. 2289–2297. IEEE, January 2012 16. Giffinger, R., Fertner, C., Kramar, H., Meijers, E.: City-ranking of European medium-sized cities. Cent. Reg. Sci. Vienna UT, pp. 1–12 (2007)
General Smart City Experts’ Perceptions of Citizen Participation
15
17. Smart Cities Council (2014), 10 January 2019. https://smartcitiescouncil.com/smart-citiesinformation-center/definitions-and-overviews 18. Caragliu, A., Del Bo, C., Nijkamp, P.: Smart cities in Europe. J. Urban Technol. 18(2), 65–82 (2011) 19. Hall, R.E., Bowerman, B., Braverman, J., Taylor, J., Todosow, H., Von Wimmersperg, U.: The vision of a smart city (No. BNL-67902; 04042). Brookhaven National Lab., Upton, NY (US) (2000) 20. de Oliveira Neto, J.S.: Inclusive Smart Cities: theory and tools to improve the experience of people with disabilities in urban spaces. Other. Université Paris-Saclay (2018). English.
21. Dameri, R.P., Rosenthal-Sabroux, C.: Smart city and value creation. In: Smart City, pp. 1–12. Springer, Cham (2014) 22. Shapiro, J.: Smart cities: explaining the relationship between city growth and human capital (2003) 23. Sproull, L., Patterson, J.F.: Making information cities livable. Commun. ACM 47(2), 33–37 (2004). https://doi.org/10.1145/966389.966412 24. Berntzen, L., Johannessen, M.R.: The role of citizen participation in municipal Smart City projects: lessons learned from Norway. In: Smarter as the New Urban Agenda, pp. 299–314. Springer, Cham (2016) 25. Cossetta, A., Palumbo, M.: The co-production of social innovation social innovation: the case of living lab. In: Smart City, pp. 221–235. Springer, Cham (2014) 26. Lombardi, P., Giordano, S., Caragliu, A., Del Bo, C., Deakin, M., Nijkamp, P., Farouh, H.: An advanced triple-helix network model for smart cities performance. In: Regional Development: Concepts, Methodologies, Tools, and Applications, pp. 1548–1562. IGI Global (2012) 27. Simonofski, A., Asensio, E.S., De Smedt, J., Snoeck, M.: Citizen participation in smart cities: evaluation framework proposal. In: 2017 IEEE 19th Conference on Business Informatics (CBI), pp. 227–236. IEEE, July 2017 28. Hashem, I.A.T., Chang, V., Anuar, N.B., Adewole, K., Yaqoob, I., Gani, A., Chiroma, H.: The role of big data in smart city. Int. J. Inf. Manage. 36(5), 748–758 (2016) 29. Li, H., de Jong, M.: Citizen participation in China’s eco-city development. will ‘new-type urbanization’generate a breakthrough in realizing it? J. Cleaner Prod. 162, 1085–1094 (2017) 30. Shelton, T., Zook, M., Wiig, A.: The ‘actually existing smart city’. Camb. J. Reg. Econ. Soc. 8(1), 13–25 (2015) 31. Hollands, R.G.: Will the real smart city please stand up? Intelligent, progressive or entrepreneurial? City 12(3), 303–320 (2008)
Citizen Sentiment Analysis in Social Media Moroccan Dialect as Case Study Monir Dahbi1(&), Rachid Saadane2, and Samir Mbarki1 1
Ibn Tofail University, Kénitra, Morocco [email protected], [email protected] 2 Hassania School of Public Work, Casablanca, Morocco [email protected]
Abstract. Smart cities have millions of sensors and innovative technologies in order to improve the quality of their citizens and to increase the competitiveness of urban infrastructure. Nowadays these citizens like to communicate using social media such as Facebook and Twitter, thus building a smart city is not free from these platforms that have changed citizen’s daily life and becoming a new source of real-time information. These data are named Big Data and are difficult to process with classical methods. To exploit this data, it must be well-processed to cover a wide range of smart city functions, including energy, transportation, environment, security and smart city management. The aim of this paper is to highlight the advantages of social media sentiment analytics to support smart city by detecting various events and concerns of citizens. Towards the end, an illustrative scenario analyses data on citizens’ concerns about traffic in three main cities in Morocco. Keywords: Smart cities
Social media Big Data Sentiment analytics
1 Introduction The Smart City is a connected city, at the forefront of modern technologies with the aim of optimally offering a wide range of services in several areas such as mobility, housing, environment, security and education. The use of Big Data for the Smart City allows efficient data storage and processing to produce information that can improve various services in the Smart City. However, Big Data needs the right tools and methods for effective data analysis. These tools and methods can encourage collaboration and communication between actors and provide services to many sectors of the smart city, while improving the “customer experience” of the citizen. While Smart City currently mainly uses data collected by physical sensors, data from social networks could be another promising source for understanding the city. Thus, local authorities almost systematically use their Facebook, Twitter, Instagram or Snapchat accounts to quickly relay all useful information to connected residents (traffic, works, events, exceptional situations). As for the inhabitants, they have also taken ownership of social networks, particularly to create communities in their cities, which generates a large amount of data that requires serious analysis.
© Springer Nature Switzerland AG 2020 M. Ben Ahmed et al. (Eds.): SCA 2019, LNITI, pp. 16–29, 2020. https://doi.org/10.1007/978-3-030-37629-1_2
Citizen Sentiment Analysis in Social Media
17
Smart city research community has proved that sentiment analysis (or opinion mining) techniques makes it possible to transform data into valuable resources and can contribute to a better understanding of, and timely reactions to, public’s needs and concerns by city governments [1]. Most of the researches that had been done in the field of sentiment analysis were on English texts with little ones on Arabic, which led to the collection of important resources, corpora, the existence of many Arabic dialects, and tools to implement applications such as text classification and topics analysis. In this work, we propose a new methodology for opinion mining using text extraction, text classification and recognition techniques focuses especially of using Moroccan dialect. In this paper, we take these challenges by automatically extracting data related to specific topics, thereafter these data are processed and analyzed in order to generate a situation report about city needs The generated reports can be used as a decision tool by the government, the information contained in these reports can quickly identify opinions of citizens, while allowing managers to better plan their action plans and better coordinate their actions operations. The rest of this paper will be broken into five parts: In section two we will come to a comparison that describes the relevant work. In the Third section, we will elaborate on our proposal Framework. The Fourth will summarize the experiments and analyses the results. Section 5, we present an illustrative scenario of the Framework proposed. Section 6 concludes the paper.
2 Related Work This work is focused on smart cites, sentiment analysis in social network. In the following, some related work will be discussed: 2.1
Smart City
Authors in [2] proposed a sentiment analysis approach which extends positive and negative polarity in higher and wider emotional scales to offer new smart services over mobile devices using microblogging data analysis for chosen topics, locations and time and such an analysis is beneficial for various communities such as policy makers, authorities, the public and in capturing branding success, diffusion in market and emotional states in relevance to different topics. In [3], the authors investigate the spatial and temporal variation of the emotions experienced by individuals whilst using urban green spaces. Through a case study of 60 urban green spaces in Birmingham, United Kingdom, demonstrate the potential for using crowdsourced Twitter data in investigations of emotional responses to urban green space. In [4], the authors report a case study in the real-time acquisition of crime detection from social media feeds, giving to responsible ability to act timely on preventing crime occurrence as detected from tweets posted by real users. In [5], the authors consider that the sentiment approach that used Multinomial Naïve Bayes classifier to build a sentiment classifier gives better results than the famous system in the task of Sentiment Analysis in Twitter in SemEval-2013 in terms of averaged F scores. The novel feature emoji has proved to
18
M. Dahbi et al.
be useful for Sentiment Analysis in Twitter data. This model has applied to real-world tweets and showed how Government agencies can track their citizens’ moods. 2.2
Sentiment Analysis
Authors in [6], Their study introduces an Arabic Jordanian twitter corpus where Tweets are annotated as either positive or negative. Experiments are conducted to evaluate the use of different weight schemes, stemming and N-grams terms techniques and scenarios. results provide the best scenario for each classifier and indicate that SVM classifier using term frequency-inverse document frequency (TF-IDF) weighting scheme with stemming through Bigrams feature outperforms the Naïve Bayesian classifier best scenario performance results. In [7] the authors developed a framework that makes it possible to analyze Twitter comments or “Tweets” as having positive, negative or neutral sentiments. This framework has many new aspects such as handling Arabic dialects, Arabizi and emoticons. Also, crowdsourcing was utilized to collect a large dataset of tweets. Same authors extended their previous work through the use of supervised learning to assign sentiment or polarity labels to tweets written in Arabizi designed A rule-based converter and applied on the tweets to convert them from Arabizi to Arabic the resultant tweets annotated with their respective sentiment labels using crowdsourcing method, Results reveal that SVM accuracies are higher than Naive Bayes accuracies. Secondly, removal of stop words and mapping emoticons to their corresponding words did not greatly improve the accuracies for Arabizi data. Thirdly, eliminating neutral tweets at an early stage in the classification improves Precision for both Naive Bayes and SVM. However, Recall values fluctuated, sometimes they got improved; on other times, they did not improve [8]. The authors present a Saudi dialect Twitter corpus for sentiment analysis [9]. The authors address the challenge of tweeting in dialectical Arabic. They propose a hybrid approach where a lexical-based classifier will label the training data, which is fed to the SVM machine learning classifier [10]. A recent study by [11] presents a novel approach to automatically construct an annotated sentiment corpus for Algerian dialect. In [12] The authors tried to review all the literature that was found on Arabic sentiment analysis, include the proposed corpus and highlighting the results and accuracies reached by each research, most research works have been spent on building and annotating dialectical corpora, however many paper researches addressed the morphology of dialectical Arabic, while most of them ignored the syntactical analysis.
3 Sentiment Analysis Methodology Understanding citizen behavior is a vast and complex task, but with the right mix of research we can better understand citizens and their motivations, thus opinions over social networks platforms can help decision-makers to analyze their needs and to design future plans [13]. This project aims to develop an overarching framework with social media sentiment monitoring to generate reports about city concerns, from extracted tweets and comments. (Figure 1) illustrates the main components of our proposed framework.
Citizen Sentiment Analysis in Social Media
19
As shown in the (Fig. 1) below our methodology is outlined in the fourth section: – – – –
Data collection and annotation Data preprocessing and features selection Opinion classification Results analysis
Fig. 1. Sentiment monitoring methodology.
3.1
Data Collection and Annotation
Moroccan Languages Variants Indeed, from early childhood, Moroccans are confronted with several languages: the mother tongue, which can be dialectal Arabic “darija” (MD) with its different speakers, or Amazigh with its three varieties (tarifit, tamazight, tachelhit) or colonial foreign languages; mainly French and Spanish present, in a minority, in the northern and southern border areas. Then there are the languages of education which, from the first years of primary school in general, are standard Arabic in its two forms, classical and modern (MSA), and recently Amazigh transcribed into “Tifinagh”. But it may also be French or even, in recent years, English, which has established itself in several sectors of professional life, particularly those of training, technology, the economy and business. In recent years have seen the appearance of Arabizi: writing in Arabic with Latin characters and with numbers. Arabizi is often used in informal contexts such as social networks, and alternated with other foreign languages, such as English or French. Corpus More and more in the world of social networks, Twitter and Facebook have become the most popular social media platforms for expressing and sharing opinions on various topics [13]. To begin, we started with collecting data from Twitter and Facebook. For this purpose, we create a Python script that automatically extract streams. Our dataset can be divided into two parts:
20
M. Dahbi et al.
Training Dataset: we collected a set of 4,000 rows that manually labeled each tweet with a binary annotation. Our dataset is generally linked to the city’s interests, such as traffic, health, energy and politics. Test Dataset: To test our system, we have used a set of 2,750 rows related to three main Moroccan cities: Rabat, Marrakech, and Casablanca. 3.2
Data Pre-processing and Features Selection
Our pre-processing and feature extraction approach is summarized as follows: (See Fig. 2).
Fig. 2. Proposed methodology for text filtering.
Pre-processing Step. Data preprocessing is the important step for building a working machine learning model and to keep only understandable data, cleaning is about: – Remove additional white spaces – Removing URL’s and user tagging – Removing punctuations Transformation Step. The goal of this step is about: – Transform every acronym name, negative words, and emoticons into their expression mapped into its corresponding word. – Removing repeated characters, for example: “ ”ﺑﺰﺍﺍﺍﺍﺍﻑis transformed into “”ﺑﺰﺍﻑ – Removing “At-tatwil” of the character “_” for example “ ”ﺷــــــــﻜﺮﺍis transformed into “”ﺷﻜﺮﺍ – Removing “At-tachkil” for example “ ”ﺍﻟﺴﻼﻡis transformed to “”ﺍﻟﺴﻼﻡ
Citizen Sentiment Analysis in Social Media
21
Translation Step (See Fig. 3). The purpose of this part is to translate every word into MSA Arabic. Translation begins with detecting the language of the word and translate each word can be translated into MSA Arabic from every other language, Google API used for this purpose. – Arabish Converter: After translating words into MSA, words in Arabizi are converted into MSA, following a thorough processing, therefore words that have not been translated are subsequently processed into an Arabish converter that converts Arabish words into Arabic letters using a Python transliteration engine in such a way that Arabish characters are replaced by all their possibilities using a correspondence table (Table 1) in order to have a set of options. These options are then searched in a database containing the most commonly Arabic with dialectical words, accompanied by their frequencies, in order to select existing words as candidates. Towards the end these candidates are sorted by their frequencies, to keep only the one that contains the highest frequency. – Dialect Converter: Finally, remaining words are processed in a Dialect converter using a dialect lexicon that links words into their corresponding MSA. This lexicon consists of dialectical words alongside their corresponding MSA words. Table 2 shows example of phrases after being converted to Arabic phrases using our methodology.
Fig. 3. Proposed approach for translation.
Stemming Step. Stemming is the process of reducing words to their word stem, therefore after preprocessing we have only text on MSA language, we apply ISRI Arabic Stemmer from NLTK libraries [14].
22
M. Dahbi et al.
Table 1. Examples of Arabic letters and their corresponding Latin letters used in Arabish.
Characters in Arabic Language ثصس ش طت و ج ىي ي زظذ م ن ذظ ج ف ق ك ل ر
Corresponding Characters in Arabish
s, sa,si,so sh sha,shi sho t, ta,ti,to,tu w, wa,wi,wu,wo j, g y ya yi yo yu z, za,zi,zo,zu m, ma,mo,mi n dh dj, dja,dji, djo f, fa,fi,fo 9a,9o, 9 e,9i k, ka,ki,ko,ke bi r,ra,ri,ro,re
Characters in Arabic Language أإءآئؤ
Corresponding Characters in Arabish
2
ع
3
خ
5
ع
3
ط
6
ح
7
ق
9
ع
3a,3i,3e,3o,3u
ب
b, ba,bi,bo,bu ch,cha,chi d, da,di,do gh, gha, ghi, gho h, ha,hi,ho i , ia,ie,ii,io,iu kh, kha,khi,kho o,oa,oi,ou,oo p, po,pi,pa,pu
ش ضد غ ه ي خ و ب
Table 2. Examples of phrases after being converted to MSA.
Dialect / Arabish
#Salam khoya baraka laho fiik Choukran mais hadchi 3adi 7olm #Moahammed زوﯾﻦ ﺑﺰااااف #Diiiima maghrib !!!!
MSA
اﻟﺴﻼم اﺧﻲ ﺑﺎرك ﷲ ﻓﯿﻚ ﺷﻜﺮا ﻟﻜﻦ ھﺬا ﻋﺎدي ﺣﻠﻢ ﻣﺤﻤﺪ ﺟﻤﯿﻞ ﻛﺜﯿﺮا داﺋﻤﺎ ﻣﻐﺮب
After cleaning and transforming our dataset we Remove stop words. Feature Extraction Step. This part is used to find the most relevant features for the classification task by deleting irrelevant and noisy data. In our experiments, we used unigrams because they would give better results. In the following, we will move to the
Citizen Sentiment Analysis in Social Media
23
creation of the vector, the weight of the word will be given according to the text containing this word. There are several weighting methods such as: – – – –
Binary weighting Term Frequency (TF) weighting Term Frequency-Inverse Document Frequency (TFIDF) Inverse Document Frequency (IDF) weighting
In this paper, we have used binary weighting (presence or not) by determining the weight of every word using a binary model to ensure that a word gets 1 if it is present in the phrase, otherwise the word gets 0. In the next stage, the feature of each labeled phrase is extracted and will be used to construct the feature vector. 3.3
Opinion Classification
Our approach for Sentiment Analysis is part of the supervised learning. In the classification stage we used the following learning algorithms: – – – –
Naive Bayes Classifier (NB) Support Vector Machines (SVM) K-Nearest Neighbors (KNN) Decision Tree (DT)
We input the created feature vectors into the classifier to build the classification model, for this purpose the “Scikit-Learn” Python library is used to build the classification report. (There are several Python libraries which provide solid implementations of a range of machine learning algorithms. One of the best known is Scikit-Learn, a package that provides efficient versions of a large number of common algorithms, it features various classification, regression and clustering algorithms including support vector machines, random forests, gradient boosting, k-means …) [15–17]. A detailed description of the various settings which were used with these four classifiers is found in the next Sect. 4 (Validation and Analysis of Results). SVM (Support Vector Machine) Support Vector Machines [18] are supervised learning models that can be used in prediction also in the classification of the linear and nonlinear data. The principle of the SVM algorithm is to use a non-linear mapping to transform the original learning data into a larger dimension. In this new dimension, it looks for the linear hyperplane of optimal separation. SVM algorithm aims to find a hyperplane with the largest margin named maximum marginal hyperplane (MMH) to be more accurate in the classification of future data i.e. it looks the shortest distance between the MMH and the closest training tuple of each class.
24
M. Dahbi et al.
SVMs are based on the idea of finding a hyperplane that best divides a dataset into two classes, as shown in the image below (See Fig. 4).
Fig. 4. Support vector machine.
Support vectors are the data points nearest to the hyperplane, the points of a data set that, if removed, would alter the position of the dividing hyperplane. Because of this, they can be considered the critical elements of a data set. NB (Naive Bayes Classifier) Naive Bayes is a classification algorithm, which works well on text categorization, based on Bayes Theorem. It makes 2 main assumptions) each feature is independent of the other feature ii) each feature is given same weight/importance. P(A/B) ¼ P(BnA)P(A)/P(B)
ð1Þ
By (1) we mean, the probability of event A given the event B is true. Some popular naive Bayes’ classification algorithms are: – Gaussian Naive Bayes – Multinomial Naive Bayes – Bernoulli Naive Bayes KNN (K-Nearest Neighbors) The k-Nearest Neighbors (KNN) algorithm is one of the simplest of the machine learning algorithms. The general idea of the k-Nearest Neighbors classifier is to classify a given query sample based on the class of its nearest neighbors (samples) in the dataset. The classification process is performed in two phases. In the first phase the nearest neighbors are determined. The neighbors are taken from a set of samples, which the class is known. The optimal number of neighbors (value of k in Fig. 5) can be calculated in different ways. They are described in literature [19]. The neighbors of a query sample are selected based on the measured distances. Therefore, a metric for measuring the distance between the query sample and other samples in the dataset has to be determined. Various metrics can be used: Euclidean, City-block, Chebyshev, etc. The Euclidean distance metric was used in those experiments.
Citizen Sentiment Analysis in Social Media
25
Fig. 5. k-Nearest Neighbors classification for various number.
The aim of the second phase is to determine the class for a query sample based on the outcomes (assigned classes) of the k selected neighbors. The decision about class is obviously straightforward in case when all determined neighbors belong to the same class. Otherwise, in case of neighbors from different classes various ways to select a class are proposed. The most straightforward solution is to assign the majority class among the k neighbors. The other widely used approach applies voting. The neighbors vote on the class, and their votes depend on their distances to the query sample. DT (Decision Tree) Decision trees (See Fig. 6) have a flowchart-like structure. These are those classifiers which work by identifying a splitting node at each step.
Fig. 6. Decision tree.
The split is decided by Information Gain. The attribute with maximum information gain is identified as the splitting node. More is the information gain, less is the entropy. Entropy represents homogeneity in data. A set’s entropy is zero when it contains instances of only one class. The steps to build a decision tree classifier are briefly described below: – Calculate the entropy of the target variable. – The dataset is then split on the different attributes. The entropy for each branch is calculated. Then it is added proportionally, to get total entropy for the split. The resulting entropy is subtracted from the entropy before the split. The result is the Information Gain or decrease in entropy.
26
M. Dahbi et al.
– The attribute with maximum information gain is selected as splitting attribute and the process is repeated on every branch. – A branch with entropy 0 represents the leaf node. Branch with entropy more than 0 requires further splitting. Decision trees are very much prone to overfitting. To fit training data perfectly, splitting is sometimes done to a huge extent. This causes the classifier to lose its generalization capability. And the model performs poorly on test dataset (unseen data).
4 Validation and Analysis of Results After the training phase, we move on to the test phase to evaluate our classifiers. For the performance validation, we use the 60% 40% method to validate our model, such as 60% used for the training phase, and 40% for the test phase. The performance measure used is the Accuracy which is represented by (formula 2) The score of this measure is calculated based on TP, TN, FP, and FN, where TP (true positives) refers to the positive Phrases that are classified correctly as positives, TN (true negatives) refers to the negative Phrases that are classified correctly as negatives, FP (false positive) refers to the negative Phrases that were classified incorrectly as positives, and FN (false negative) refers to the positive Phrases that were classified incorrectly as negatives [10]. Accuracy ¼ TP þ TN=TP þ TN þ FP þ FN
ð2Þ
Figure 7 shows the results obtained with our algorithms, with two adjustments, the first without translation, and the second with translation.
Fig. 7. Performance of classifiers.
As shown in the results (Fig. 7), the SVM and the Naïve Bayes classifiers works better both in terms of accuracy of 94.01%, 92% respectively, when using translation. Our results are coherent with those from other works, drawn from different data sets and with different contexts. For example, in [20], the authors compared between different classifiers namely SVM classifier, Tree classifier, J48 Decision and Naïve Bayes, with the aim of classifying Al Hadith. The authors used a dataset of 795 texts (includes
Citizen Sentiment Analysis in Social Media
27
Hadiths) Their results showed that the SVM achieved the best accuracy followed by the Naïve Bayes and decision tree respectively. In another work [21] a comparative study with various results for four sentiment classifiers, concluded that SVM has high performance than Naïve Bayes, Decision Tree and Neutral Network respectively. Therefore, their results also showed that the SVM classifier achieved the highest accuracy, succeeded by the Naive Bayes classifier.
5 Illustrative Scenario In order to understand the citizens’ opinions and motivations on traffic services we collected data from Facebook and Twitter related to the situation of traffic in #Rabat, #Marrakech and #Casablanca. Afterwards, we applied our approaches to extract sentiments. (Figure 8) shows results obtained by analyzing people’s opinions about traffic in the Moroccan cities: Rabat, Marrakech, and Casablanca over a week period from March 7, 2019, to March 14, 2019.
Fig. 8. Opinions related to the traffic in Morocco main cities.
By analyzing results, a large part of smart cities has a negative sentiment about traffic, with more negative views on the traffic in Rabat. Unlike, Marrakech recorded a lower percentage than other cities, which is not the case for Rabat and Casablanca. as a conclusion, Social networks citizens suffer from traffic conditions therefore those decision-makers must work to improve road traffic infrastructure.
6 Conclusions and Future Work In this article, we have described a system for detecting sentiment analysis in social networks and how those users can be playing as social sensors. In this work, we described how social media analytics could help to analyze urban data streams collected from popular social media sources, such as Twitter and Facebook. Besides we explored the challenge of using texts that include Moroccan dialect and Arabish.
28
M. Dahbi et al.
We presented several sentiment analysis experiments with different adjustments. So we have proven particularly that applying sentiment analysis techniques gave better results on MSA data than on dialectical data directly. We believe that this work will help to give support to research on opinion mining of the Moroccan dialect. In future work, we would like to automate the annotation task and perform a thorough analysis of our results. We analyzed user’s opinions about traffic in three cities in Morocco. The results show the extent of user dissatisfaction with traffic. So their decision-makers need to work hard to resolve this issue.
References 1. Ahmed, K.B., et al. : Sentiment analysis for smart cities: state of the art and opportunities. In: Proceedings on the International Conference on Internet Computing (ICOMP). The Steering Committee of the World Congress in Computer Science, Computer Engineering and Applied Computing (WorldComp) (2016) 2. Vakali, A., Chatzakou, D., Koutsonikola, V., Andreadis, G.: Social data sentiment analysis in smart environments (2013) 3. Roberts, H., Sadler, J., Chapman, L.: The value of Twitter data for determining the emotional responses of people to urban green spaces: a case study and critical evaluation. Urban Stud. 56(4), 818–835 (2019) 4. Souza, A., Figueredo, M., Cacho, N., Araújo, D., Coelho, J., Prolo, C.A.: Social smart city: a platform to analyze social streams in smart city initiatives. In: 2016 IEEE International Smart Cities Conference (ISC2), pp. 1–6. IEEE, September 2016 5. Li, M., et al.: The new eye of smart city: novel citizen sentiment analysis in twitter. In: 2016 International Conference on Audio, Language and Image Processing (ICALIP). IEEE (2016) 6. Alomari, K.M., ElSherif, H.M., Shaalan, K.: Arabic tweets sentimental analysis using machine learning. In: International Conference on Industrial, Engineering and Other Applications of Applied Intelligent Systems. Springer, Cham (2017) 7. Duwairi, R.M., et al.: Sentiment analysis in arabic tweets. In: 2014 5th International Conference on Information and Communication Systems (ICICS). IEEE (2014) 8. Duwairi, R.M., et al.: Sentiment analysis for Arabizi text. In: 2016 7th International Conference on Information and Communication Systems (ICICS). IEEE (2016) 9. Assiri, A., Emam, A., Al-Dossari, H.: Saudi twitter corpus for sentiment analysis. world academy of science, engineering and technology. Int. J. Comput. Electr. Autom. Control Inf. Eng. 10(2), 272–275 (2016) 10. Aldayel, H.K., Azmi, A.M.: Arabic tweets sentiment analysis–a hybrid scheme. J. Inf. Sci. 42(6), 782–797 (2016) 11. Guellil, I., et al.: SentiALG: automated corpus annotation for Algerian sentiment analysis. In: International Conference on Brain Inspired Cognitive Systems. Springer, Cham (2018) 12. Hussien, I.O., Dashtipour, K., Hussain, A.: Comparison of sentiment analysis approaches using modern Arabic and Sudanese dialect. In: International Conference on Brain Inspired Cognitive Systems. Springer, Cham (2018) 13. Alahmary, R.M., Al-Dossari, H.Z., Emam, A.Z.: Sentiment analysis of Saudi dialect using deep learning techniques. In: 2019 International Conference on Electronics, Information, and Communication (ICEIC). IEEE (2019)
Citizen Sentiment Analysis in Social Media 14. 15. 16. 17. 18. 19. 20.
21.
29
ISRI documentation. https://www.nltk.org/_modules/nltk/stem/isri.html Scikit-Learn documentation. https://scikit-learn.org/stable/documentation.html Scikit-Learn. https://en.wikipedia.org/wiki/Scikit-learn Scikit-Learn. https://jakevdp.github.io/PythonDataScienceHandbook/05.02-introducingscikit-learn.html Han, J., Kamber, M.: Data mining: concepts and techniques, 2nd ed., [Nachdr.] Kruczkowski, M., Niewiadomska-Szynkiewicz, E.: Comparative study of supervised learning methods for malware analysis. J. Telecommun. Inf. Technol. (2014) Faidi, K., Ayed, R., Bounhas, I., Elayeb, B.: Comparing Arabic NLP tools for Hadith classification. In: Proceedings of the 2nd International Conference on Islamic Applications in Computer Science and Technologies (IMAN 2014) (2014) Chauhan, P.: Sentiment analysis: a comparative study of supervised machine learning algorithms using rapid miner. Int. J. Res. Appl. Sci. Eng. Technol. 80–89 (2017). https://doi. org/10.22214/ijraset.2017.11011
Social Networks Fake Profiles Detection Using Machine Learning Algorithms Yasyn Elyusufi(&), Zakaria Elyusufi, and M’hamed Ait Kbir LIST Laboratory, Faculty of Sciences and Technologies, Tangier, Morocco [email protected]
Abstract. Fake profiles play an important role in advanced persisted threats and are also involved in other malicious activities. The present paper focuses on identifying fake profiles in social media. The approaches to identifying fake profiles in social media can be classified into the approaches aimed on analysing profiles data and individual accounts. Social networks fake profile creation is considered to cause more harm than any other form of cyber crime. This crime has to be detected even before the user is notified about the fake profile creation. Many algorithms and methods have been proposed for the detection of fake profiles in the literature. This paper sheds light on the role of fake identities in advanced persistent threats and covers the mentioned approaches of detecting fake social media profiles. In order to make a relevant prediction of fake or genuine profiles, we will assess the impact of three supervised machine learning algorithms: Random Forest (RF), Decision Tree (DT-J48), and Naïve Bayes (NB). Keywords: User profiling
Fake profile detection Machine learning
1 Introduction Social media is growing incredibly fast these days. This is very important for marketing companies and celebrities who try to promote themselves by growing their base of followers and fans. The social networks are making our social lives better but there are a lot of issues which need to be addressed. The issues related to social networking like privacy, online bullying, misuse, and trolling etc. are most of the times used by fake profiles on social networking sites [1]. However, fake profiles, created seemingly on behalf of organizations or people, can damage their reputations and decrease their numbers of likes and followers. On the other hand fake profile creation is considered to cause more harm than any other form of cyber crime. This crime has to be detected even before the user is notified about the fake profile creation. In this very context figures this article, which is part of a series of research conducted by our team within the user profiling subject and profiles classification in social networks. Facebook is one of most famous online social networks. With Facebook, users can create user profile, add other users as friends, exchange messages, post status updates, photos, and share videos etc. Facebook website is becoming popular day by day and more and more people are creating user profiles on this site. Fake profiles are the profiles which are not genuine i.e. they are the profiles of persons with false credentials. The fake facebook © Springer Nature Switzerland AG 2020 M. Ben Ahmed et al. (Eds.): SCA 2019, LNITI, pp. 30–40, 2020. https://doi.org/10.1007/978-3-030-37629-1_3
Social Networks Fake Profiles Detection
31
profiles generally are indulged in malicious and undesirable activities, causing problems to the social network users. People create fake profiles for social engineering, online impersonation to defame a person, advertising and campaigning for an individual or a group of individuals [2]. Our research aims at detecting fake profiles at online social media websites using Machine learning algorithms. In order to address this issue, we have chosen to use a Facebook dataset with two thousand eight hundred and sixteen users (instances). The goal of the current research is to detect fake identities among a sample of Facebook users. This paper consists of three sections. In the first section, Machine learning algorithms that we chose to use to address our research issues are presented. Secondly, we present our architecture. In the third section, evaluation model guided by Machine learning algorithms will be advanced in order to identify fake profiles. The conclusion comes in the last section.
2 State of the Art Fake profiles in social media are often used to establish trust and deliver malware or a link to it. Such fake profiles are also used in other types of malicious activities. To solve these problems, a significant body of research to date has focused on fake profiles detection in social media. Generally, following the taxonomy presented by Song et al. [3]. The approaches to identifying fake social media profiles can be classified into the approaches aimed analyzing individual accounts (profile-based techniques as well as graph-based methods), and the approaches capturing the coordinated activities spanning in a large sample of accounts. For instance, Nazir et al. describes in [4] detecting and characterizing phantom profiles in online social gaming applications. The work analyses a Facebook application, the online game “Fighters club”, known to provide incentives and gaming advantage to those users who invite their peers into the game. The authors state that by providing such incentives the game motivates its players to create fake profiles, by introducing those fake profiles into game, the user would increase incentive value for himself. At first, the authors extract thirteen features for each user, and then perform classification using support vector machines (SVMs). This work concludes that these methods do not suggest any obvious discriminants between real and fake users. On the other hand Adikari, S. and Dutta, K., 2014 works on Known fake LinkedIn profiles, posted on special websites. The detection method uses the number of languages spoken, education, skills, recommendations, interests, awards, etc. These features are used to train neural networks, SVMs, and principal component analysis. The accuracy found in this work was 84% True Positive (TP), and 2.44% False Negative (FN). In the same context Stringhini et al. 2010 works on spam accounts registered by honeypots: 173 spam accounts in Facebook and 361 in Twitter [6]. In this research Random forest was constructed based on the following features: ratio of accepted friend requests, URL ratio, message similarity, regularity in the choice of friends, messages sent, and number of friends. The accuracy found in this work was 2% FP, 1% FN for (Facebook); and 2.5% FP, 3.0% FN for (Twitter). Also Yang et al. treat spam Twitter accounts defined as the accounts containing malicious URLs, about 2060 spam accounts are identified [7]. In This latter research authors uses graph based features (local clustering coefficient, betweenness centrality, and bi-directional links
32
Y. Elyusufi et al.
ratio, neighbor-based features, also timing-based features were used to construct different classifiers. The accuracy found was 86% TP, 0,5% FP. The previous approaches assume that the machine learning techniques are too challenging because the attackers create patterns that cannot be trained by machines. But recent works have applied many standard machine learning algorithms, such as ensemble of classifiers, Random Forests, SVM, Decision Trees and Artificial Neural Networks. Several machine learning algorithms are used for the classification of profiles based on their features. The survey on efficient machine learning studies introduces several algorithms and discusses their processing ability with respect to prediction accuracy. Many machine learning algorithms are used in order to identify fake profiles in social networks. The different machine learning techniques used by various works are shown in Fig. 1.
Fig. 1. Machine learning techniques used in recent works.
In several works, we have discussed profiling techniques and their applications [8– 10]. In this work, we identify the minimal set of user profile data that are necessary for identifying fake profiles in facebook, about two thousand eight hundred and sixteen users including fake and genuine profiles are used. In order to identify the most efficient machine learning algorithms, we used three supervised machine learning Algorithms: Random Forest (RF), Decision Tree (J48) and Naïve Bayes (NB). The application section will be made with Jupter tool (Python 3) in order to choose the most efficient algorithms. Finally the comparison of the accuracy and confusion matrix of each model can better explain the interest of our work.
Social Networks Fake Profiles Detection
33
3 Using Naive Bayes Classifiers Naive Bayes classifiers are a family of simple probabilistic classifiers used in machine learning. These classifiers are based on applying Bayes theorem with strong (naive) independence assumptions between the features. Naive Bayes is a simple method for constructing classifiers: models that assign class labels to problem instances, represented as vectors of feature values, where the class labels are drawn from some finite set. It is not a single algorithm for training such classifiers, but a family of algorithms based on a common principle: all naive Bayes classifiers assume that the value of a particular feature is independent of the value of any other feature, given the class variable [11]. Naive Bayes classifiers are a popular statistical technique of email filtering. They emerged in the middle of the 90s and were one of the first attempts to tackle spam filtering problem [11]. Naive Bayes typically use bag of words features to identify spam e-mail, an approach commonly used in text classification. Naïve Bayes classifiers work by correlating the use of tokens (typically words, or sometimes other constructions, syntactic or not), with spam and non-spam e-mails and then using Bayes theorem to calculate a probability that an email is or is not a spam message [11]. In our paper we will assess the impact of using Naïve Bayes classifiers in the prediction of fake or genuine profiles in social networks (Facebook Data set).
4 Using Decision Trees Classifiers A decision tree is a popular classification method that generates tree structure where each node denotes a test on an attribute value and each branch represents an outcome of the test as shown in Fig. 2. The tree leaves represent the classes. This technique is fast unless the training data is very large. It does not make any assumptions about the probability distribution of the attributes value. The process of building the tree is called induction [12].
Fig. 2. Example of fake profiles identification using decision tree.
34
Y. Elyusufi et al.
The decision tree algorithm is a top-down greedy algorithm which aims to build a tree that has leaves as homogenous as possible. The major step in the algorithm is to continue dividing leaves that are not homogeneous into leaves that are as homogeneous as possible until no further division is possible as shown in Fig. 2. In our approach we will assess the impact of using Decision Trees classifiers in the prediction of fake or genuine profiles in social networks (Facebook Data set).
5 Using Random Forest Classifiers Random Forest is one of the most used machine learning algorithms, because its simplicity and the fact that it can be used for both classification and regression tasks. It’s a supervised learning algorithm. As it’s seen from its name, it’s bagged decision tree models that split on a subset of features on each split, it creates a forest and makes it somehow random. The «forest» it builds, is an ensemble of Decision Trees, most of the time trained with the “bagging” method. The general idea of the bagging method is that a combination of learning models increases the overall result. To say it in simple words: Random forest builds multiple decision trees and merges them together to get a more accurate and stable prediction. For better understanding of the RF algorithm is necessary to explain what the main idea behind decision trees is. Depending on the features in each dataset, the decision tree model learns a series of questions to figure out the class labels of the instances. What makes this model successful is that it is nonparametric and it can handle heterogeneous data (ordered or categorical variables, or a mix of both). Furthermore decision trees fundamentally implement feature selection, making them at least to some extent robust to irrelevant or noisy variables and are robust to outliers or errors in labels [13]. To summarize, here is steps that Random Forest algorithm follows: • • • •
Randomly choose n samples from the training set with replacement. Grow a decision tree from the n sample. At each node. Repeat the steps 1 to 2 k-times. Aggregate the prediction by each tree to assign the class label by majority vote.
6 Our Approach In our research work, a novel approach has been presented for the identification of fake profiles on facebook using supervised machine learning algorithms. The proposed model has applied data preprocessing techniques on datasets before analyzing them. A technique has been applied to identify the non significant attributes in datasets and to do attribute reduction. The proposed model was trained using supervised machine algorithms individually for dataset including fake and genuine users. Ensemble classifier has been used to make the prediction more accurate. As shown in Fig. 3, the process of fake profile detection has three levels, in the first level profile features are extracted and then in the second level: Random Forest (RF), Naïve Bayes (NB) and Decision Tree (DT) are used to determine the fake and genuine profiles. The third level, we calculate and compare the accuracy rates across the results of both techniques.
Social Networks Fake Profiles Detection
35
Fig. 3. The process flow in identification of the real or fake profile
6.1
Initial Features
After using the dataset, we proceed for feature selection phase. We noticed that there was much of unneeded features, either have no meaning for our subject or full of NaN values, so to make our models train well, we decided to drop them and to let only those of will affect directly on the results. As shown in Table 1, initially we have 33 profile features which will be used in the Facebook DataSet. 6.2
Features Selection
Feature selection is one of the basic concepts in machine learning which hugely impacts the performance of classification and prediction. In our work, and in order to make our models train well, we decided to use only features which will affect directly the results. The features on the final dataset were: statuses_count, followers_count, friends_count and favorites_count. Below is the meaning of each feature Table 2. After all this steps, and in considering that the algorithm split the dataset into training and testing sets, and to make the row homogenous, we shuffled them. Finally we use the final dataset to train and evaluate the three machine learning algorithms: Random Forest, Decision Tree (J48), and Naïve Bayes.
36
Y. Elyusufi et al. Table 1. Table of initial features. Features Id name screen_name statuses_count followers_count friends_count favourites_count listed_count created_at url lang time_zone location default_profile default_profile_image geo_enabled profile_image_url profile_banner_url profile_use_background_image profile_background_image_url_https profile_text_color profile_image_url_https profile_sidebar_border_color profile_background_tile profile_sidebar_fill_color profile_background_image_url profile_background_color profile_link_color utc_offset Protected verified Description updated
Description Id of user User name Screen Name Statuses Count Followers Count Friends count Favourites Count Listed Count Date of account creation_at Acount Url Language Time zone Geographic Location Default profil Default Profil Image Status Géo localisation Profile Image URL Profile Banner URL Profile Background Image Profile background Image Url Profile Text Color Profile Image Url Https Profile Sidebar Border Color Profile Background Title Profile Side Bar Fill Color Profile Background Image URL Profile Background Color Profile Link Color Offset Status Protection Status Verification Status Description of Account Update Date
Table 2. Table of selected features. Features statuses_count followers_count friends_count favourites_count
Description Statuses Count Followers Count Friends Count Favourites Count
Social Networks Fake Profiles Detection
37
7 Results and Discussion 7.1
Splitting the Dataset
Before the training phase, we have to split the dataset. In this step we started with choosing the input and output values. The input contains the independent variables and the output contains the dependent variable (Fake or Genuine value) that takes the value of 0 and 1, then we split the dataset into the training set and test set. In our work the training test set is defined with 80%, while the test set is defined with 20%. 7.2
Evaluation Metrics
We present the metrics used to evaluate the results in order to select the best supervised machine learning algorithm. We first show the model accuracy and then use the confusion matrix shown in Formula 1. This matrix is used to visualize the performance of the different algorithms using the following metrics. This metric indicates the fraction of returned cases that are valid fake profiles. The accuracy function (Formula 1) Accuracy ¼
TP þ TN TP þ TN þ FN þ FP
ð1Þ
Table 3 represents the confusion matrix used to evaluate the efficiency of proposed methods. Table 3. The confusion matrix Predicted Predicted Examples of this class TP Example not belonging to this class FP
7.3
class No predicted FN TN
Experimental Results
In this research, we have compared the results of three machine learning algorithms (Random Forest, Decision Tree (J48) and Naïve Bayes) to determine the most appropriate approach to differentiate the legitimate profiles from fake profiles in Facebook dataset. Tables 4, 5 and 6 summarize the final confusion matrix values akin to each algorithm by calculating the correctly classified instances and incorrectly classified instances. In addition, it shows the accuracy calculation for each algorithm. Based on the result tests of the tree algorithms, as shown in Fig. 4, it is obvious that Random Forest algorithm is better than Decision Tree (J48) and Naïve Bayes algorithms. It ranked first with an accuracy score of 99,64%. Where Decision Tree (J48) algorithm give only 559 correctly classified instances with an accuracy of 99,28%. Naïve Bayes is the last with an accuracy of 78,33%.
38
Y. Elyusufi et al. Table 4. Classification of profiles on Facebook using (Decision Tree) Decesion Tree (J48) Confusion Matrix 260 1 3 299 Correctly Classified Instances 559 99.2895% Incorrectly Classified Instances 4 0.7105%
Table 5. Classification of profiles on Facebook dataset using (Naïve Bayes) Naïve Bayes Confusion Matrix 260 1 121 181 Correctly Classified Instances 441 78.3304% Incorrectly Classified Instances 122 21.6696% Table 6. Classification of profiles on Facebook dataset using (Random Forest) Random Forest Confusion Matrix 260 1 1 301 Correctly Classified Instances 561 99.6448% Incorrectly Classified Instances 2 0.3552%
Fig. 4. Accuracy comparison between DT, NB and RF algorithms
Social Networks Fake Profiles Detection
39
8 Conclusion and Future Work In this paper, we proposed an approach to identify the fake profile in social network using limited profile data, about 2816 users. As we concluded in our paper, we demonstrate that with limited profile data our approach can identify the fake profile with 99.64% correctly classified instances and only 0.35% incorrectly classified instances, which is comparable to the results obtained by other existing approaches based on the larger data set and more profile information. Our research can be a motivation to work on limited social network information and find solutions to make better decision through authentic data. Additionally, we can attempt similar approaches in other domains to find successful solutions to the problem where the least amount of information is available. In future work we expect to run our model using more sophisticated concepts such as ontology engineering, in order to semantically analyze user posts, and comportments. This later concept can improve the quality of prediction of fake or genuine profiles.
References 1. Boshmaf, Y., Muslukhov, I., Beznosov, K., Ripeanu, M.: The socialbot network: when bots socialize for fame and money. In: Proceedings of the 27th Annual Computer Security Applications Conference, pp. 93–102. ACM (2011) 2. Romanov, A., Semenov, A., Veijalainen, J.: Revealing fake profiles in social networks by longitudinal data analysis. In: 13th International Conference on Web Information Systems and Technologies, January 2017 3. Song, J., Lee, S., Kim, J.: CrowdTarget: target-based detection of crowdturfing in online social networks. In: Proceedings of the 22nd ACM SIGSAC Conference on Computer and Communications Security, CCS 2015, pp. 793–804. ACM, New York (2015) 4. Nazir, A., Raza, S., Chuah, C.-N., Schipper, B.: Ghostbusting Facebook: detecting and characterizing phantom profiles in online social gaming applications. In: Proceedings of the 3rd Conference on Online Social Networks, WOSN 2010. USENIX Association, Berkeley, CA, USA, p. 1 (2010) 5. Adikari, S., Dutta, K.: Identifying fake profiles in Linkedin. Presented at the Pacific Asia Conference on Information Systems PACIS 2014 Proceedings (2014) 6. Stringhini, G., Kruegel, C., Vigna, G.: Detecting spammers on social networks. In: Proceedings of the 26th Annual Computer Security Applications Conference, ACSAC 2010, pp. 1–9 (2010) 7. Yang, C., Harkreader, R.C., Gu, G.: Die free or live hard? Empirical evaluation and new design for fighting evolving Twitter spammers. In: Proceedings of the 14th International Conference on Recent Advances in Intrusion Detection, RAID 2011, pp. 318–337. Springer, Heidelberg (2011) 8. Elyusufi, Y., Seghiouer, H., Alimam, M.A.: Building profiles based on ontology for recommendation custom interfaces. In: International Conference on Multimedia Computing and Systems (ICMCS) Anonymous IEEE, pp. 558–562 (2014) 9. Elyusufi, Y., Alimam, M.A, Seghiouer, H.: Recommendation of personalized RSS feeds based on ontology approach and multi-agent system in web 2.0. J. Theor. Appl. Inf. Technol. 70(2), 324–332 (2014)
40
Y. Elyusufi et al.
10. Elyusufi, Z., Elyusufi, Y., Ait Kbir, M.: Customer profiling using CEP architecture in a Big Data context. In: SCA 2018 Proceedings of the 3rd International Conference on Smart City Applications Article No. 64, Tetouan, Morocco, 10–11 October 2018. ISBN: 978-1-45036562-8 11. Granik, M., Mesyura, V.: Fake news detection using naive Bayes classifier. In: Conference: IEEE First Ukraine Conference on Electrical and Computer Engineering (UKRCON), May 2017 12. Ameena, A., Reeba, R.: Survey on different classification techniques for detection of fake profiles in social networks. Int. J. Sci. Technol. Manage. 04(01), (2015) 13. Beatriche, G.: Detection of fake profiles in Online Social Networks (OSNs), Master’s degree in Applied Telecommunications and Engineering Management (MASTEAM), (2018)
Smart City Mobile Apps Exploratory Data Analysis: Mauritius Case Nawaz Mohamudally(&) and Sandhya Armoogum University of Technology, Mauritius, Pointe-aux-Sables 11134, Mauritius [email protected]
Abstract. Advances in technology are quickly paving the way for smart cities. According to Economic Development Board Mauritius, the Government of Mauritius has set up the Smart City Scheme to provide an enabling framework and a package of attractive fiscal and non-fiscal incentives to investors for the development of smart cities across the island. However, prior to the design and implementation of such technologies, it is important to predict the behavioural intention to use such technology so that smart city technologies effectively empower citizens and improve the quality of life of citizens. In this research work, it is proposed to use the Technology Acceptance Model (TAM) to effectively assess the perception and readiness and the perceived usefulness of certain smart city technologies such as for transportation as well as identifying key smart city applications for Mauritius. The aim of this research project is to evaluate and assess the different factors and condition that can have an impact on the perceived ease of use (PEOU), perceived usefulness (PU), attitudes towards using (ATT), behavioural intention (BI) to use and actual use (AU) of smart city technologies. This chapter is a complementary to the conference paper in SCA2019 which comprised Load Factor Analysis, here the focus is on Cronbach Alpha, Correlation and ANOVA. Keywords: TAM
ANOVA SmartCity App
1 Introduction 1.1
Rationale Behind the Citizen as a Major Stakeholder in Smart Cities
A Smart City encompasses an urban development vision that emphasizes the intelligent management of a city’s resources for the purpose of solving urban challenges. This intelligent management almost always includes the participation of all the stakeholders of the city and not just the municipality itself. A vital prerequisite of this vision is the active participation and engagement of the most important stakeholder of all, the citizen. Courtesy: https://www.alliedtelesis.com/blog/ict-fundamental-enabler-smartcities. Smart cities are mushrooming around the Mauritius Island as depicted in Fig. 1. The Live-Work-Play model has been presented as socio economic development one. However, little has been done to evaluate or understand the citizen evolution in this new environment. This is simply due to the infancy stage of the smart cities on the Island. One aspect which is actually under consideration is about mobile intelligence. © Springer Nature Switzerland AG 2020 M. Ben Ahmed et al. (Eds.): SCA 2019, LNITI, pp. 41–52, 2020. https://doi.org/10.1007/978-3-030-37629-1_4
42
N. Mohamudally and S. Armoogum
In other words when citizens would be roaming around in a smart city, what are the smart services available? Thus enforcing the concept of smart cities. The Smart City is for now the “Blind Men and the Elephant” parable, expectations and perception vary largely. The study presented in this chapter attempts to gauge the user acceptance of ICT based services for smart cities. As a matter of fact, in many cities across the globe for instance in the Taiwanese capital, the majority of the services are accessible via mobile apps. The methodology adopted is therefore a statistical analysis of a survey on smart city mobile apps.
Fig. 1. 16 smart cities locations in Mauritius
1.2
Chapter Description
The rest of the chapter comprises a literature search described in Sect. 2 on the Technology Acceptance Model. Section 3 comprises the methodology including survey information and hypotheses. Exploratory data analysis is covered in Sect. 4 followed by conclusion in Sect. 5.
Smart City Mobile Apps Exploratory Data Analysis
43
2 Technology Acceptance Model Technology Acceptance Model is one of the most popular theory that is used widely to explain Information System usage. Many studies have been conducted which has led to the changes in the originally proposed model. A new model called combined TAMTPB model which integrated the Technology acceptance model and theory of planned behaviour was proposed by Taylor and Todd [1]. Venkatesh and Davis [2] proposed a new version of TAM called TAM2 which added new variables to the existing model namely social influence processes (subjective norm, voluntariness, and image) and cognitive instrumental processes (job relevance, output quality, result demonstrability). Venkatesh et al. [3] further proposed an improved model the Unified Theory of Acceptance and Use of Technology (UTAUT) Model. The various studies conducted by researchers have tried to modify the TAM by adding new variables to it. Agarwal and Prasad [4] modified TAM by adding the construct of compatibility in the Technology Acceptance Model. Moon and Kim [5] has added a new variable playfulness factors to study acceptance of the World Wide Web. Chau [6] in a study reviewed TAM by included two types of perceived usefulness: near-term and long-term. Vander Heijden [7], after analysing the individual acceptance and usage of the website added two new constructs to TAM: perceived entertainment value and perceived presentation attractiveness. Chau and Hu [8] combined the factor of peer Influence with Technology Acceptance Model. Chau and Hu [9] further compared three models, namely the Technology Acceptance Model (TAM), the Theory of Planned Behavior (TPB), and a decomposed TPB model that is potentially adequate in the targeted healthcare professional setting in Hong Kong. The results indicated that TAM was superior to TPB in explaining the physicians’ intention to use telemedicine technology. In [10], the authors examined the consumer acceptance of online banking using an extension of the TAM model where they reported that PU was more influential than PEOU in explaining technology acceptance. Similarly in [11], mobile banking adoption was studied using TAM and TPB. In [12] an extended model based on TRA and TAM approach was used to predict acceptance of e-shopping. It was observed that PEOU and PU significantly determine individual attitudes toward e-shopping. This study also suggests that user acceptance is a better indicator of e-shopping intentions than user satisfaction. In [13], an extension of the TAM to include the four variables (process satisfaction, outcome satisfaction, expectations, and E-commerce use) was used to assess e-commerce where it was reported that the extended TAM explained actual behaviour in E-commerce environments better than the original TAM. Scherer et al. [14] uses TAM to explain teachers’ adoption of digital technology in education. It was concluded that using TAM is relevant but the role of certain key constructs and the importance of external variables contrast some existing beliefs about the TAM. Mobile learning (M-learning), which is based on learning activities using a mobile device like a smart phone, or tablet, is becoming increasingly important in today’s networked environment.
44
N. Mohamudally and S. Armoogum
The TAM model has been adopted to explain and predict the acceptance of Mlearning [15], whereby the PU, PEOU, perceived enjoyment (PE) and the importance of perceived mobility value (PMV) i.e. the ability to learn anytime and anywhere were studied. The predictive power of the added constructs to TAM – PE and PMV – shows that the new variables were imperative. A general extended TAM for e-learning (GETAMEL) is proposed in [17] to include external factors namely Self-Efficacy, Subjective Norm, Enjoyment, Computer Anxiety and Experience. GETAMEL is further adopted to assess students’ behavioral intention to use e-learning in Azerbaijan [21]. TAM has also be used in assessing user acceptance of Mobile Library Applications in Academic Libraries [18], where it was reported that PU, PEOU and interactivity and user satisfaction had significant effects on user attitude and intention to use mobile library applications. An extended TAM has also been used to evaluate user acceptance of YouTube for procedural learning [16]. This extended model took into consideration user satisfaction, content richness, vividness, and YouTube self-efficacy, as well as content richness, apart from PU and PEOU. In a study to understand the factors that influence the adoption of collaborative technologies (namely Google Applications for collaborative learning) to support team work in project-based environments, an enhance TAM is proposed [20]. The ability to share information in the collaborative learning environment was found to influence BI. TAM has also been used to determine acceptance of Iranian agricultural consultants’ of precision agriculture techniques and sustainable farming systems [22]. Similarly in [19], an application of the extended technology acceptance model (TAM2) was used to predict adoption of biological control for various types of pests among Iranian rice farmers. Other than PU, farmers’ intentions of biological control were affected by perceived self-efficacy, facilitating conditions, and compatibility. However, PEOU was not considered as a positive factor towards BI.
3 Methodology The quantitative method was selected for the study through the use of a questionnaire as it is a suitable way to reach a geographically dispersed audience at a relatively low cost. The survey which consisted some 16 pages was administered to a sample of 200 citizens of Mauritius. The convenience sampling method was used where citizens were approached online or through their work place. Respondents were rest-assured about the confidentiality and anonymity of the study. The content validity of the questionnaire was confirmed by a panel of academic lecturers at the University of Technology Mauritius. The reliability of the questionnaire was estimated by conducting a pre-test in which 10 questionnaires were administered to respondents, who were then excluded from the sample. The response was quite spontaneous and the participants showed ease of understanding for the questionnaire. Following the pilot test, the questionnaire was slightly modified based on comments received from respondents. The final questionnaire consisted of a short covering letter, and directions on how to fill the questionnaire. The first section of the questionnaire consists of 9 questions related to the profile of the citizen. The second section of the questionnaire consists of
Smart City Mobile Apps Exploratory Data Analysis
45
21 questions based on the five features of the SmartCity App which was showcased to them through a video of the interaction with the app. The Smart City App is a collection of six different applications. These apps are Nearby Places app, Next Buses app, Weather Info app, Parking app, News app, and Complaint app. The last section of the questionnaire consists of some 30 questions to assess the acceptance as well as to get input about the factors that may affect adoption such as quality of internet connection. Respondents were mainly asked to provide ratings on a 5-point Likert Scale for most questions, which was deemed convenient. 3.1
Sample Size and Hypotheses
The population size (the total number of people being studied) is 1, 263,820 at the end of 2016 [23] out of which 638,267 are women against 625,206 men. A sample of minimum 200 respondents is being targeted. This section outlines the measurement scales used to build research constructs. Respondents were asked to indicate agreement with each statement in a measure using a five-point Likert-type scale (1, strongly disagree; 2, disagree; 3, neutral; 4, agree; 5, strongly agree). The measures related to each construct then were assessed using respondent perceptions: 3.1.1 Perceived Usefulness (PU) TAM posits that PU is a significant factor affecting acceptance of an information system. Davis defined PU as “the degree to which a person believes that using a particular system would enhance his or her job performance” [21]. Hence an application perceived to be more useful than another is more likely to be accepted by users. By applying these into the SmartCity App context we hypothesize: Hypothesis 1. Perceived usefulness (PU) has a positive effect on consumer acceptance of the SmartCity App. 3.1.2 Perceived Ease of Use (PEOU) According to TAM PEOU is a major factor that affects acceptance of information system [7]. PEOU is defined as “the degree to which a person believes that using a particular system would be free of effort”. Therefore, the Smart City App if perceived to be easy to use is more likely to be accepted by users. Therefore, in this case, we hypothesize: Hypothesis 2a. Perceived ease of use (PEOU) has a positive effect on consumer acceptance of the SmartCity App. Hypothesis 2b. Perceived ease of use (PEOU) has a direct effect on perceived usefulness (PU). 3.1.3 User Satisfaction (US) and Perceived Enjoyment (PE) Previous studies have indicated that User Satisfaction (US) affects the effectiveness of IS, system usage, as well as, directly or indirectly, affecting IS performance through IS usage. Although Davis did not include user information satisfaction in his TAM, we revised it based on prior studies and define it as user satisfaction with the
46
N. Mohamudally and S. Armoogum
Internet/WWW. Regarding the use of the SmartCity App, this study is expected to derive the new relations among US, PU and PEOU. Hypothesis 3a.User Satisfaction (US) has a positive effect on user acceptance of the SmartCity App. Hypothesis 3b. PU positively affects User Satisfaction (US) of the Smart City App. Perceived enjoyment refers to the extent to which the activity of using a computer is perceived to be enjoyable in its own right. This is contrasting to the PU, which be seen as an extrinsic motivation whereas perceived enjoyment (PE) as an intrinsic motivation to use information systems. A number of studies on PE have noticed that PE significantly affects intentions to use computers. Igbaria et al. [24] found that PE correlates positively with time of use but not with frequency of use or number of tasks. On this basis, we expect that PE affects the acceptance of the SmartCity App: Hypothesis 4. Perceived enjoyment (PE) has a positive effect on user acceptance of the SmartCity App. 3.1.4 Behavioral Intention to Use (BI) Behavioral Intention refers to the user’s likelihood to engage in the use of the Smart City App. Therefore, in our research, it can be hypothesized: Hypothesis 5a. Perceived ease of use (PEOU) has a direct effect on behavioural intention to use. Hypothesis 5b. Perceived usefulness (PU) has a direct effect on behavioural intention to use. 3.1.5 Quality of Internet (QI) The importance of a decent Internet connection and its quality was raised in our research. Without a proper Internet connection, the use of the Smart City App is not possible. Hence, we posit: Hypothesis 6. The quality of the Internet connection has a positive effect on consumer acceptance of SmartCity App. 3.1.6 Citizen Experience and Readiness We define Citizen Experience and readiness as the level of exposure of citizens to mobile applications. We believe that citizens who are used to consuming ICT services such as online shopping, e-banking, or communication will be more inclined to adopt the SmartCity App. Hypothesis 7. Citizen Experience and Readiness has a positive effect on acceptance of SmartCity App.
4 Exploratory Data Analysis 4.1
Cronbach Alpha
Cronbach’s alpha is a measure of internal consistency, that is, how closely related a set of items are as a group. It is considered to be a measure of scale reliability. Used often in psychometric test.
Smart City Mobile Apps Exploratory Data Analysis
a¼
N c v þ ðN 1Þ c
47
ð1Þ
Where N is the number of items, c is the average covariance between item-pairs, v is the average variance. Cronbach Alpha was also carried out to test for the reliability of the factors obtained from Perceived Usefulness (PU), Perceived Ease of Use (PEOU), User Satisfaction (US), Personal Experience (PE), Behavioral Intention (BI) and Quality of Internet (QI). The results are shown in Table 1 as follows: Table 1. Cronbach alpha results Parameters PU PEOU US PE BI QI
Cronbach alpha 0.962 0.772 0.947 0.754 0.970 0.686
All of the above yielded values greater than 0.5, showing high reliability. Some authors prefer larger values of alpha, beyond 0.9 to characterize an excellent internal consistency. 4.2
Correlation
Table 2 shows the correlation between Perceived Usefulness, User Satisfaction, Behavioural Intention, Personal Experience and Quality of Internet. As expected, a very strong positive correlation (r = +807) was obtained between perceived usefulness and user satisfaction, indicating that the higher the user will perceive the SmartCity App to be useful, the greater he/she will be satisfied. Therefore, perceived usefulness predicts user satisfaction to a very great extent. Likewise, another very strong correlation (r = +.867) was observed between perceived usefulness and behavioural intention which means that, the greater the user will perceive the Smart App, the higher he/she will have the intention to use it again. However, a weak positive correlation (r = +.267) was observed between perceived usefulness and quality of internet indicating that perceived usefulness is less likely to predict the quality of internet. This implies that the extent to which the user will perceive the Smart App as being useful will less likely predict the quality of internet. Furthermore, a positive association between perceived ease of use and user satisfaction (r = +.743) was observed, which implies that the greater the user will perceive the App as being easy to use, the more likely he/she will be satisfied. Finally, a strong positive correlation (r = +.838) between user
48
N. Mohamudally and S. Armoogum
satisfaction and behavioural intention was observed, which means that the greater the user will be satisfied, the more likely the user will have the intention to use it again. Hence, user satisfaction predicts behavioural intention very strongly. The correlation coefficient is given by the formula below: (https://corporate financeinstitute.com/resources/knowledge/finance/correlation/) P ðxi xÞðyi yÞ ffi rxy ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi P P ðxi xÞ2 ðyi yÞ2
ð2Þ
rxy is the correlation coefficient of the linear relationship between the variables x and y xi are the values of the x-variable in a sample x is the mean of the values of the x-variable yi are the values of the y-variable in a sample ȳ is the mean of the values of the y-variable Normally a correlation of 1 indicates a positive correlation of two items in the same direction, −1 in opposite direction and towards zero as a weak correlation. Also to mention that correlation does not necessarily imply causation. In this context a good QI does not imply good usage of adoption of services. Table 2. Correlations between Perceived Usefulness, Perceived Ease of Use, User Satisfaction, Personal Experience, Behavioral Intention and Quality of Internet and their Means and Standard Deviations. Measures PU PEOU US PE BI QI PU 1 PEOU 0.741** 1 US 0.807** 0.743** 1 PE 0.743** 0.592** 0.755** 1 BI 0.867** 0.710** 0.838** 0.773** 1 QI 0.267** 0.237* 0.352** 0.260** 0.372** 1 *Correlation is significant at the 0.05 level (2-tailed) **Correlation is significant at the 0.01 level (2-tailed).
4.3
Anova
The Analysis Of Variance (ANOVA) is an important metric to study the impact of items within a group and impact among groups. Table 3 presents ANOVA results for age groups, whereas Table 4 across genders and Table 5 for educational levels. ANOVA compares the different samples means and we can use it for hypothesis testing too.
Smart City Mobile Apps Exploratory Data Analysis
49
Table 3. Mean Scores (and standard deviations) for measured variables across different age groups Variables 60 4.00 – 4.00 – 4.00 – 4.00 – 4.13 – 2.88 –
ANOVA F(1.172) = .329 F(.464) = .802 F(1.682) = .146 F(1.354) = .249 F(1.616) = .163 F(1.153) = .338
Table shows the F-test values, the F-test is computed as follows: F test ¼ variance of the group means=mean of the within group variances
ð3Þ
It is actually the ratio between the variation between sample means and the variation within the samples. In Table 3 the F-test values are closer to 0.5, for PEOU is around 0.8, which shows that the variance of the group means is higher compared to within the groups. In Table 4, most of the F-test values seem normal except for the QI with an F-test score of 0.027 indicates a discrepancy between male and female citizens regarding the Quality of Internet. There is probably need to investigate into the Quality of Experience (QoE) although the latter is somehow linked to the Quality of Service (QoS) of the Internet connectivity. Table 4. Mean Scores (and standard deviations) for measured variables across gender Variables Male PU 3.56 (.93) PEOU 3.79 (.91) US 3.61 (.98) PE 3.55 (1.06) BI 3.65 (.96) QI 3.15 (1.57)
Female 3.76 (.85) 3.72 (.96) 3.77 (.95) 3.76 (1.18) 3.58 (1.05) 2.55 (.88)
ANOVA F(1.226) = .271 F(.167) = .684 F(.686) = .409 F(.823) = .366 F(.105) = .746 F(5.021) = .027
50
N. Mohamudally and S. Armoogum
Table 5. Mean Scores (and standard deviations) for measured variables across different education levels Variables Primary Secondary PU 2.13 3.44 (1.12) (1.05) PEOU 2.10 3.63 (.14) (1.22) US 2.00 3.48 (1.41) (1.30) PE 2.25 3.67 (1.77) (1.66) BI 2.06 3.29 (1.50) (1.21) QI 2.44 3.11 (1.33) (2.31)
Tertiary 3.74 (.80) 3.84 (.79) 1.00 (.27) 3.67 (.85) 1.06 (.25) 2.84 (.89)
ANOVA F(4.167) = .018 F(3.945) = .023 F(4.164) = .018 F(1.628) = .202 F(4.835) = .010 F(.474) = .624
Table 5 shows the poor F-test values across the broad, it can be deduced from the gap between groups with primary educational level only and that of the tertiary one. Though it seems to be obvious, the table is a showcase of the risk of ICT digital divide in the context of smart cities.
5 Conclusion As expected, a strong positive correlation was observed that the relationship between perceived usefulness and behavioural intention was strong. Therefore, the more users will perceive the Smart City Apps to be useful, the more they will have the intention of using it again. Furthermore, users will be more satisfied with the app if they find it more useful. This statement has been well supported by the results of this study which has shown a very strong positive correlation between perceived usefulness and user satisfaction. Lack of user acceptance is a significant impediment to the success of adoption of Smart City apps. The research has led to a Modified TAM model for Smart City apps. The concept of Smart City is quite subjective. Mauritius Citizens are mature users of Smart City mobile apps. Higher QI may not lead to higher Perceived Usefulness except for Real Time apps. There is need for popularization of smart city underlying technologies. Given the big number and openness of local Smart Cities, there is little risk of Digital Divide. Following a seminar organised by the Mauritius Research Council (MRC) in the context of research findings dissemination, the following points shall be considered for future research. Much discussion was on the perception about smart city. Observations were exposed about improvement of utility services for cities be really smart and lack of motivations for promoters to invest into state of the art technologies. Debate was also held on the legal definition for smart cities in Mauritius. The sample of 200 to be broadened to represent a more realistic sample of the Mauritian population instead of
Smart City Mobile Apps Exploratory Data Analysis
51
looking more at the millennials. More citizens are leaving towns to move to larger open space. One research question that would be interesting to explore is that, are there technological motives for people to move on to cities. Many governmental apps are underutilised. Request has been made to study the reasons behind. Acceptance should not be limited only to mobile apps but other ICT interfaces. Acknowledgment. The authors are thankful to the Mauritius Research Council (MRC) for the funding and dissemination of the research work globally. They are also beholden to the University of Technology, Mauritius research assistant and students support for the Mobile Apps development.
References 1. Taylor, S., Todd, P.A.: Understanding information technology usage: a test of competing models. Inf. Syst. Res. 6, 144–176 (1995). https://doi.org/10.1287/isre.6.2.144 2. Venkatesh, V., Davis, F.D.: A theoretical extension of the technology acceptance model: four longitudinal field studies. Manage. Sci. 46(2), 186–204 (2000). https://doi.org/10.1287/ mnsc.46.2.186.11926 3. Venkatesh, V., Morris, M.G., Davis, G.B., Davis, F.D.: User acceptance of information technology: toward a unified view. MIS Q. Manage. Inf. Syst. 27(3), 425–478 (2003) 4. Agarwal, R., Prasad, J.: A conceptual and operational definition of personal innovativeness in the domain of information technology. Inf. Syst. Res. 9(2), 204–215 (1998) 5. Moon, J.W., Kim, Y.G.: Extending the TAM for a world-wide-web context. Inf. Manage. 38, 217–230 (2001). https://doi.org/10.1016/S0378-7206(00)00061-6 6. Chau, P.Y.K.: An empirical assessment of a modified technology acceptance model. J. Manage. Inf. Syst. 13(2), 185–204 (1996). https://doi.org/10.1080/07421222.1996. 11518128 7. Heijden, H.V.D.: E-Tam: a revision of the Technology Acceptance Model to explain website revisits, Serie Research Memoranda 0029, VU University Amsterdam, Faculty of Economics, Business Administration and Econometrics (2000) 8. Chau, P., Hu, P.: Investigating healthcare professionals’ decisions to accept telemedicine technology: an empirical test of competing theories. Inf. Manage. 39, 297–311 (2002) 9. Chau, P.Y.K., Hu, P.J.H.: Information technology acceptance by individual professionals: a model comparison approach. Decis. Sci. 32(4), 699–719 (2001) 10. Pikkarainen, T., Pikkarainen, K., Karjaluoto, H., Pahnila, S.: Consumer acceptance of online banking: an extension of the technology acceptance model. Internet Res. 14(3), 224–235 (2004). https://doi.org/10.1108/10662240410542652 11. Aboelmaged, M.G., Gebba, T.R.: Mobile banking adoption: an examination of technology acceptance model and theory of planned behavior. Int. J. Bus. Res. Dev. 2(1), 35–50 (2013). ISSN1929‐0977 12. Shih, H.-P.: An empirical study on predicting user acceptance of e-shopping on the web. Inf. Manage. 41(3), 351–368 (2004) 13. Fayad, R., Paper, D.: The technology acceptance model e-commerce extension: a conceptual framework. Procedia Econ. Finance 26, 1000–1006 (2015). https://doi.org/10.1016/S22125671(15)00922-3
52
N. Mohamudally and S. Armoogum
14. Scherer, R., Siddiq, F., Tondeur, J.: The technology acceptance model (TAM): a metaanalytic structural equation modeling approach to explaining teachers’ adoption of digital technology in education. J. Comput. Educ. (2018). https://doi.org/10.1016/j.compedu.2018. 09.009 15. Huang, J.-H., Lin, Y.-R., Chuang, S.-T.: Elucidating user behavior of mobile learning: a perspective of the extended technology acceptance model. Electron. Libr. 25(5), 585–598 (2007). https://doi.org/10.1108/02640470710829569 16. Far, S.T., Rezaei-Moghaddam, K.: Determinants of Iranian agricultural consultants’ intentions toward precision agriculture: integrating innovativeness to the technology acceptance model. J. Saudi Soc. Agric. Sci. 16, 280–286 (2017) 17. Abdullah, F., Ward, R.: Developing a general extended technology acceptance model for elearning (GETAMEL) by analysing commonly used external factors. J. Comput. Hum. Behav. 56, 238–256 (2016) 18. Yoon, H.-Y.: User acceptance of mobile library applications in academic libraries: an application of the technology acceptance model. J. Acad. Librarianship 42, 687–693 (2016) 19. Muk, A., Chung, C.: Applying the technology acceptance model in a two-country study of SMS advertising. J. Bus. Res. 68, 1–6 (2015) 20. Lee, D.Y., Lehto, M.R.: User acceptance of YouTube for procedural learning: an extension of the technology acceptance model. J. Comput. Educ. 61, 193–208 (2013) 21. Cheung, R., Vogel, D.: Predicting user acceptance of collaborative technologies: an extension of the technology acceptance model for e-learning. J. Comput. Educ. 63, 160–175 (2013) 22. Chang, C.-T., Hajiyev, J., Su, C.-R.: Examining the students’ behavioral intention to use elearning in Azerbaijan? The general extended technology acceptance model for e-learning approach. J. Comput. Educ. 111, 128–143 (2017) 23. Statistics Mauritius, Mauritius in Figures. http://statsmauritius.govmu.org/English/Publica tions/Pages/Mauritius-in-Figures.aspx 24. Igbaria, M., Iivari, J., Maragahh, H.: Why do individuals use computer technology? A Finnish case study. Inf. Manage. 29, 227–238 (1995). https://doi.org/10.1016/0378-7206(95) 00031-0
Gamification of Civic Engagement in Smart Cities (New Russian Practices) Olga Sergeyeva1(&) , Elena Bogomiagkova2 Ekaterina Orekh2 , and Natalia Kolesnik1
,
1
2
Sociological Institute of the RAS, a Branch of the FCTAS of the RAS, 7 Krasnoarmeyskaya st. 25, St. Petersburg, Russia [email protected] St. Petersburg University, Smolnogo st. 1/3 Entrance 9, St. Petersburg, Russia
Abstract. The authors review the gamification practices implemented in Russia to enhance civic and political participation. Gamification is the spread of game elements beyond traditional entertainment, and it causes both critical (I. Bogost) and optimistic estimates (J. McGonigal) among researchers. While scientists are discussing gamification effects, managers and political strategists use game apps. The Russian practice of intensifying citizen and political participation through gamification covers three key areas: (1) solving urban problems through the organization of public discussions, voting, environmental performances; (2) the interaction of political leaders and the electorate, (3) the introduction of information agenda. The systematization of gamification cases shows that game designers primarily use the balance of game features, such as interactivity and cooperation, for citizen and political involvement. Keywords: Gamification
Gameful design Citizen participation Russia
1 Introduction A smart city can be defined as an environment of intelligent infrastructures that support the urban economy, transport, the environment, as well as urban governance and politics. Interactivity generated by these technologies mostly determines the nature of digital technologies as the material basis of smart urban infrastructures. That is, between the material objects of urban infrastructure and the citizens, there is an ongoing information exchange: the urban environment “responds”, giving information to the person directly, and the person makes a choice and sets up this environment “for himself”. One of the new trends in citizen interactions in the space of intellectual infrastructures is the gamification or application of computer game elements (points, virtual currencies, levels, progress indicators, etc.) to motivate individuals and keep their attention and interest. Scholars’ views about gamification cover the entire range, from positive to critical. Ideas in support of gamification (like J. McGonigal’s) emphasize the creative potential of the game, which transforms any process into a fascinating action [9]. Criticism of gamification (for example, by I. Bogost) aims at the manipulative potential of game elements introduced into economic and political communications [3]. © Springer Nature Switzerland AG 2020 M. Ben Ahmed et al. (Eds.): SCA 2019, LNITI, pp. 53–62, 2020. https://doi.org/10.1007/978-3-030-37629-1_5
54
O. Sergeyeva et al.
Since gamification in Russia was only taking root in the field of civil-political involvement, in this article, we tried to give an overview of practices that already use game elements. Engagement through gamification is associated with the development of the information infrastructure in a country and citizens’ established skills as technology users. The level of digitalization in Russia does not demonstrate a large gap compared with the average European level. National monitoring data suggests that in May 2019, 84% of the Russian Federation’s population used the Internet (69% used it daily) [11]; in Europe, 86.8% of the total population uses the Internet [6]. The Internet has become a part of everyday life for the majority of Russian people, which means there is a technological basis for the promotion of digital game projects. Collecting Russian cases of civic engagement gamification, we systematized them by two criteria—the level of interactivity and the level of cooperation. These gameplay components are, according to game designers, what brings gamers joy and pleasure [8]. Game apps for politics and urban management are serious games that differ from entertainment games; the serious game is designed to work towards achieving a nongaming result. The discovery of interactivity and cooperation as factors of gaming pleasure, we believe, is essential for understanding the turn to gamification. Also, consideration of the effects of cooperation that can be cultivated by game elements in the field of civic and political participation opens the way for discussion of the solidarity potential that technologies have. By reviewing the practices of gamification, we sought to discover whether people can cooperate through computer-game technologies despite a long-standing social phobia that sees a tendency toward social isolation in screen games. Therefore, we analyzed gamification practices, placing them in the spaces that resulted at the intersection of two axes: the “interactivity axis” and the “cooperation axis”.
2 Literature Review The term “gamification” originated in the environment of IT developers and was taken up by philosophers and social scientists. The first mention of gamification in the scientific databases Scopus and Web of Science relate to 2011. In 2011, the term “gamification” was only referred to in the full texts of two articles in the Russian eLibrary. In 2012, however, Russian publications contained this term in titles, annotations and keywords. It is now clear that for nearly a decade, an interdisciplinary research field that focuses on the manifestation of gamification in various public spheres has been developing. So far, authors have been competing in their attempts to define gamification, and working with this concept fluctuates between two primary meanings. First, gamification relates to the widespread adoption and institutionalization of computer games, which affects everyday life experiences and interactions. Secondly, the use of game elements (points, badges, virtual currencies, levels and indicators of progress) is considered to motivate participants to take any action, increase their activity and retain attention as well as interest. Researcher and game designer S. Deterding stresses that gamification is not the participation in games but is only some entertainment elements
Gamification of Civic Engagement in Smart Cities
55
embedded in other social contexts to make non-game products, services or applications more motivating and/or attractive to use [5]. Also, it is often overlooked that gamification is a metaphorical expression of a tendency to de-rationalize the prevalence of emotional and entertainment components in all social institutions (through infotainment, edutainment, etc.). Much progress in the study of gamification practices has been made in the areas of marketing and organizational communications. The first books on gamification, “Game-Based Marketing” and “Gamification by Design” by G. Zichermann, were published in 2010 [16] and 2011 [15]. At that time, there was already a practical implementation of gamification principles in the form of the Salesforce platform, which had been created in 2007 by Bunchball (a company founded in 2005). In 2010, Badgeville, which also specializes in software that gamifies business processes, was started. Today, these processes associated with the economy—labor, personnel management and consumer behavior—demonstrate the approbation of gamification strategies and the theoretical elaboration of these innovations. The works analyzing the specifics of gamification in political interactions greatly rely on ideas about participatory culture. Digital participation arose where information users and creators’ roles are intertwined; Web 2.0 gives users a platform for civic expression. Joint games also become such areas [12]. Crowdsourcing is another concept explaining the nature of the policy field in connection with gamification. Crowdsourcing means “transferring work to the crowd,” and it is closely related to the idea of a “collective mind” that supports an egalitarian society. Usually, gamers individually seek a goal, while crowdsourcing unites all this human power to solve one particular problem. According to the ideas of well-known media analysts I. Bogost, S. Ferrari, and B. Schweizer, gamification of journalism is the case of political crowdsourcing [4]. In examples from Wired magazine projects, they explain the effects of interactive game models, in which many citizens participate, who can offer solutions to the topical issues of current politics. Along with political crowdsourcing, journalists use the game format to draw attention to current events, which is called newsgames. The term ‘newsgames’ is credited with a renowned game designer and scientist, Gonzalo Frasca. He considered editorial games a combination of political cartoons and computer simulations [7]. In addition to the above, it can be stated that the academic vision of policy-field gamification has two epistemological branches: positive and critical. The first is presented, for example, in the works of game developer and researcher J. McGonigal. In her book “Reality is Broken: Why Games Make Us Better, and How They Can Change the World”, she explores the role of games in empowering social and political participation in societies in which people are tired of aggressive and annoying direct coercion and engagement. Game applications are, in her opinion, the system of “fixing broken reality” [9]. If computer games were first assessed as a form of escapism, then McGonigal explains the participatory game resources with examples in an effort to prove that the connections between people do not disappear; rather, they are regrouping (or, as she writes, “reinventing”). The critical position on the gamification of the policy field denotes the developing trend of libertarian paternalism. Libertarian paternalism implies, for example, that the state gives citizens the freedom to choose but introduces rules that limit the number of
56
O. Sergeyeva et al.
options. People should feel free, but their behavior is regulated. This principle is known to all gamers: the available choices are somewhat limited in most computer games, but within these restrictions, some do good deeds, and some do not. Participants can choose only the options provided to them. Feature options are embedded in a carefully built selection architecture. Critics of gamification see it as a method that controls visible behavior rather than changing a person’s ideas. The computer game researcher I. Bogost expresses polemical judgments: “gamification is a means of exploitation, perversion, and simplification of the game medium, created by marketers and big business for easy gain” [3]. A literature review clarifies the current “growth points” of gamification’s social analytics. Political and civic issues worked much less representatively than economic issues. There is no systematization of gamification practices in contemporary Russia, and this is important while considering its positive and negative effects.
3 Solidarity and Cooperation Through Gamification At the level of local policy, the steadiest trend in gamification is to encourage cooperation. The Russian experience of gamification of civic participation at the municipal level is “Active Citizen,” an online platform that allows Moscow residents to vote on city development. Voting results are implemented either through the adoption of laws by Moscow or through departmental decisions. However, when calling the experience of “Active Citizen” an electronic referendum, we must bear in mind that the voting results have no legal validity, but are an advisory agent in government decisionmaking.
Fig. 1. Screenshot of the “Online Store of Reword”, the Moscow project “Active citizen” https://shop.ag.mos.ru/catalog/
Gamification of Civic Engagement in Smart Cities
57
The actions of “Active Citizen” participants are comparable to the games because awards for activity on this platform were introduced. The “Online Store of Reword” works here, the bright interface of which is typical for any online store (see Fig. 1). A voting Moscow resident can thus get gaming pleasure from both promoting his opinion and gaining points. Gamification’s strength is transforming complicated, tedious, or unpleasant processes into an exciting experience. Garbage collection, like no other activity, embodies such an “unpleasant process.” Therefore, the gamification of this practice is one of the most successful projects implemented by Russian civic activists. “Clean Games” is a social movement born in St. Petersburg in 2014. It is a team competition for cleaning up debris and waste separation. “Clean Games” is supported by the website and mobile application. It is a complete IT platform that coordinates members’ activities. The system allows one to register participants for the game (through social networks and without) and to form teams; to conduct an online rating of competitions (during the game, volunteers enter into the system which team gives how much garbage) using the mobile application, participants can take garbage “check-ins”—photos with GPS tags—and earn extra credits; and after the game, there are detailed statistics of how much garbage different factions have collected. Participants solve environmental puzzles, make photo check-ins, search for artifacts, and collect and share garbage, getting points for it. The best garbage collectors win prizes. The concept of an eco-quest was tested for the first time on the Vuoksa River in the Leningrad region. In 2015, “Clean Games” reached the All-Russian level. In the “Clean Games” case, we see the contribution of the game media components (photo check-ins, online maps, and users’ video clips) that make the project visible, create part of the fun of the game, and monitor the implementation of the rules. Last but not least, the digital gamification of garbage collection appeals to the gamers’ habits, which is relevant for some participants as an attribute of the modern lifestyle. Gamification as a way of maintaining a community of like-minded people is embodied in contemporary Russian electoral interactions. Innovative political strategies have also been used at the federal level, for example, in the presidential election campaign of 2018, when one of the candidates, Ksenia Sobchak, managed to build a working network-oriented political party. There were not only traditional but also, and most significantly, digital methods involved in the conducted election campaign. The interaction process between all the participants in the election was mediated by an automatic system using a whole set of technical tools. Those who supported K. Sobchak’s candidacy registered on the site and was modeled on the info business site on a platform similar to JustClick. On the site, each registered user had a personal account to monitor all the headquarters’ campaign work. K. Sobchak’s website allowed the candidate’s supporters to not only participate in writing the “123 difficult steps” candidate program but also provide financial assistance and receive emails and voice messages from the candidate herself, while the supporters’ activity was rewarded. For example, launching the crowdfunding platform, Bank of Ideas, K. Sobchak addressed the audience: “Advise the presidential candidate. All ideas are considered, published;
58
O. Sergeyeva et al.
you can comment on them and vote for them by liking. Feel free to suggest. Speak out, discuss, vote. We’ll implement the best ideas, and we’ll credit their authors” [14]. Professional membership to K. Sobchak’s election headquarters requires a special mention. The political headquarters included political technologists, both of the old wave (who had already worked in the election headquarters of the first Russian president, B. Yeltsin), and the new one. Among the representatives of the new wave at the headquarters of K. Sobchak was the political strategist V. Shklyarov, who participated in the campaigns of A. Merkel in Germany, B. Obama in the USA, and the team of B. Sanders (D. Trump’s election headquarters); he was the author of the project “Political Uberization” for the Gudkov-Katz team in the Russian municipal elections. V. Shklyarov, who was responsible for technology at the election headquarters of K. Sobchak, determined that its primary goal is to convert millions of K. Sobchak’s Instagram subscribers into volunteers [2]. The researchers of participatory democracy underline that the success of these practices depends on the balance of four factors: “(1) political will, (2) organizational capacity of the social fabric of a territory, (3) autonomy and financial capacity of the promoter political entity and (4) organizational architecture (or “design”) of the participatory process” [1]. Gamification is the design of a participatory process (factor 4) that is expected to be able to “reanimate” the self-organization of citizens tired of traditional involvement.
4 Interactivity and Immersiveness Through Gamification The involvement of different social groups in decision-making processes has been institutionalized as the basis of democracy since the Modern Age. To date, however, such a policy of participative management has acquired the opposite traits. Scientists note the trend of interpassivity, which is triggered by the interactive fatigue and manifested in the delegation of responsibility for decision-making to the state [12, 13]. Public policy actors are constantly looking for ways to attract and keep public attention on important issues as well as ensure political and social participation. Newsgames is a new media technology aimed at drawing the audience’s attention to the public and political agenda. Becoming gamers, previously “passive” users receive the personal experience of information perception turning into the accomplices of news. The newsgames format has several features. First, newsgames are distinguished from usual computer games because they use real social, cultural or political events (current or past) as game spaces (settings). Newsgames also employ the images of political leaders, public figures, news heroes and representatives of different social groups as game characters. So, Ch. Plewe and E. Fürsich consider an example of newsgames in which gamers immerse themselves in the everyday reality of refugees and migrants to feel their experience [10]. Current news becomes a narrative for the game. Second, since potential users of newsgames are large audiences, they need to be quickly and universally accessible for everyone: “Newsgames need to make use of simple or previously existing game mechanics that their gamers are already familiar with” [10, P. 2472]. Also, as a rule, newsgames are posted on the online platforms of media publications. Third, newsgames, which offer specific rules, allow the gamer to
Gamification of Civic Engagement in Smart Cities
59
not only feel like a participant in actual events but also to influence them, “animating” game avatars of real heroes from current news. In Russia, the newsgames format has been implemented by “Lentach”, “Meduza”, “RIA Novosti”, “Esquire Russia Magazine” and others. For several years (since 2014), “Lentach” has used the newsgames format to summarize and evaluate the results of the year. The idea behind these games is that successful actions by the user (who turns out to be a participant in the events of the past year) can change the present, albeit in virtual reality. The game has the slogan, “It is time to change history!” Political and social events that have attracted attention in the past year act as game plots. Game simulations offer the gamer help to prevent or affect what happened, while constantly moving from one situation to another. The game assumes the presence of several levels, and the transition to the next level is possible only after the successful completion of the previous level. (It’s a turn-based game.) The characters of newsgames are well-known political figures, such as the president, deputies, governors, etc. For example, we can consider incorporating D. Trump’s inauguration into the plot of the newsgame, in the victory of which Moscow’s influence is continuously discussed (see Fig. 2). The newsgame plot includes several episodes related to the scandalous image of the president. The gamer must lead Trump’s avatar so that he overcomes all obstacles and comes to the inauguration. This newsgame uses a countdown, the frames of which share the game levels. The music gives a unique atmosphere to the game; in particular, at one of the levels, a gamer acts under the soundtrack of H. Mancini’s “The Pink Panther”.
Fig. 2. Screenshot of the game “Trump’s Inauguration” by the information resource “Lentach” http://lenta.ch/trump/
60
O. Sergeyeva et al.
Lentach is a resource known for using provocative, dramatic political events as plots for newsgames, with caricatures ridiculing them. In this case, the game can be compared with satire. Lentach’s visual content of newsgames is simple and schematic; the process of the game is accompanied by music, which also sets a specific context for immersing and experiencing what is happening on the screen. Newsgame, by Lentach, does not require special skills or technical equipment from the gamer; success in the game depends on the performance of simple functions. Instead, the user needs quick reactions and timely keystrokes. Another example is a game by Meduza that also possesses relatively simple mechanics and contains not only information but also an assessment of current events in Russia or globally. Thus, one of the newsgames by Meduza is dedicated to the recently elected President of Ukraine, Vladimir Zelensky, who was formerly a popular stand-up comic. The game mechanics are based on a test selection of the correct answer. The gamer sees quotes by V. Zelensky on the screen and must quickly decide in what situation this phrase was said: in public speeches by President Zelensky or during comic shows by Zelensky the artist. For each correct answer, the gamer receives points. In Russia, the potential of newsgames is used primarily by opposition information companies. The very mechanics of the games put the users in a position to criticize the phenomena that underlie the games’ plots. By offering some rather than other situations as elements of the games, the companies designate them as necessary and significant and highlight them from the whole space of problems that might become visible to users, thereby contributing to agenda-setting. The creators of these games determine the specific perspectives on the topics. Newsgames contain an explicit or hidden assessment of political and social events, and gamers can only act in conditions that are set by the logic of the game.
5 Conclusion The infrastructures of modern cities are called “smart” because they accumulate data on a variety of physical and social processes. This data is designed for active users, that is, those who can regulate the information flow and apply the received information to solve their civic and professional tasks. However, the 21st century has become an era of disappointment in politics and civil structures, expressed in absenteeism, diminished interest in cooperation and collaboration with neighbors and colleagues, and the avoidance of political news. All of these are global processes. In this context, the tools for increasing the motivation with which the political elites (both governing and opposition) and civic activists of any movement try to gain interest and mobilize citizens for their support have become especially in demand. Managers and politicians consider the use of gaming principles and technical resources of digital media as an instrument for social inclusion and achievement of organizational goals. Gamification is the reality of today, and it is logical to imagine that this process is the result of social changes in contemporary society. In this sense, we are not just talking about trendy technology used by someone for some purpose but also about the ubiquitous trend of our time brought about by certain circumstances. The need to use
Gamification of Civic Engagement in Smart Cities
61
unique mechanisms of involvement in any process is caused by weakening attention, a decreasing interest, a reduction or lack of desire to participate among consumers. It is hard to disagree with G. Zichermann and D. Linder that “the classical ways of engaging in the modern world no longer work, given the huge number of distractions that are rapidly increasing” [16, P. 23]. The Russian practice of intensifying citizen and political participation through gamification covers three key areas: (1) the solution of city problems, (2) the interaction of political leaders and the electorate, (3) the introduction of information agenda. The design of game elements implemented in the communications of citizens, the state, political parties and social movements primarily uses the balance of such properties of the game as interactivity and cooperation. Interactivity is characteristic of newsgames, embodying a turn towards immersive journalism. The effects of cooperation due to game interactions are produced in citizens’ projects to solve urban problems and in election campaigns. Gamification in civic and political communications is becoming more common. At the same time, elements of serious computer games change the status of a game in a culture that used to be connected with something unproductive, but today we can see how games work.
References 1. Allegretti, G., Antunes, S.: The Lisbon participatory budget: results and perspectives on an experience in slow but continuous transformation. Field Actions Science Reports 11 (2014). http://journals.openedition.org/factsreports/3363 2. Azar, I.: Polittekhnolog shtaba Sobchak Vitalij Shkljarov: «Interesno poshhupat’ molodezh’, do kotoroj ne dotjanulsja Naval’nyj» [Vitaly Shklyarov, political strategist at the headquarters of Sobchak: “It’s interesting to feel the youth that Navalny did not reach”]. Novaja gazeta [Novaya Gazeta] 120(2697), 12–13, 27 October 2017. [in Russian] 3. Bogost, I.: Why gamification is bullshit. In: Walz, S.P., Deterding, S. (eds.) The Gameful World: Approaches, Issues, Applications, pp. 65–80. The MIT Press, Cambridge (2015) 4. Bogost, I., Ferrari, S., Schweizer, B.: Newsgames: Journalism at Play. The MIT Press, Cambridge (2010) 5. Deterding, S., Sicart, M., Nacke, L., O’Hara, K., Dixon, D.: Gamification: using gamedesign elements in non-gaming contexts. In: Proceedings of ACM CHI 2011 Conference on Human Factors in Computing Systems, pp. 2425–2428. Vancouver, BC, Canada (2011) 6. Internet Word Stats, 31 March 2019. https://www.internetworldstats.com/stats.htm. Accessed 14 Sept 2019 7. James, J.: Newsgames – journalism innovation through game design. Am. J. 34(3), 379–383 (2017) 8. Kazakova, N.Ju., Nazarov, Ju.V.: Tselevaja auditorija gejm-dizajna i igrovoj protsess [Game Design Target Audience and Game Process]. Dekorativnoe iskusstvo i predmetnoprostranstvennaja sreda. Vestnik MGKhPA [Decorative Art and Subject-spatial Environment. Bull. Moscow State Acad. Arts Ind. 1, 393–414 (2015). [in Russian] 9. McGonigal, J.: Reality is Broken: Why Games Make Us Better and How They Can Change the World. The Penguin Press, New York (2011) 10. Plewe, C., Fürsich, E.: Are newsgames better journalism? Empathy, information, and representation in games on refugees and migrants. J. Stud. 19(16), 2470–2487 (2018)
62
O. Sergeyeva et al.
11. Russian Public Opinion Research Center (VTsIOM): polling data № 3948, 06 May 2019. https://wciom.ru/index.php?id=236&uid=9681. Accessed 14 Sept 2019. [in Russian] 12. Van Deth, J.W.: Democracy and involvement: the benevolent aspects of social participation. In: Torcal, M., Montero, J.R. (eds.) Political Disaffection in Contemporary Democracies: Social Capital, Institutions and Politics, pp. 101–129. Routledge, New York (2006) 13. Van Oenen, G.: Three cultural turns: how multiculturalism, interactivity, and interpassivity affect citizenship. Citizsh. Stud. 14(3), 293–306 (2010). https://doi.org/10.1080/ 13621021003731856 14. Urazalieva, G.K.: Cifrovaja partija v politicheskom prostranstve: opyt prezidentskoj kampanii v RF 2018 goda [Digital Party in the Political Space: The Experience of the Presidential Campaign in Russia in 2018]. In: Kuleshova, A.V. (ed.) Sociolog 2.0: transformacija professii: Materialy VIII Mezhdunarodnoj konferencii [Sociologist 2.0: Profession Transformation: Proceedings of the VIII International Conference], pp. 357–361. VTsIOM [Russian Public Opinion Research Center], Moscow (2018). [in Russian] 15. Zichermann, G., Cunningham, C.: Gamification by Design: Implementing Game Mechanics in Web and Mobile Apps. O’Reilly Media, Sebastopol (2011) 16. Zichermann, G., Linder, J.: Game-Based Marketing: Inspire Customer Loyalty Through Rewards, Challenges, and Contests. Wiley, Hoboken (2010)
Machine Learning for Sentiment Analysis: A Survey Zineb Nassr(&), Nawal Sael, and Faouzia Benabbou Laboraty of Modeling and Information Technology, Faculty of Sciences Ben M’SIK, University Hassan II of Casablanca, Casablanca, Morocco [email protected]
Abstract. The scope of this research fits in sentiment analysis. This latter is becoming more and more an active field of research where to extract people’s opinion concerning political, economic and social issues. The objective of sentiment classification is to classify opinions of users as positive, negative or neutral from textual information alone. For that, purpose researchers used data mining classification techniques such as naïve Bayes classifier and the Neural Networks. The sentiment analysis area is confronted to several problems that distinguish it from traditional thematic research, since the sentiment is expressed in a very varied and very subtle ways. In the last few years, several researches focused on sentiment analysis in order to study attitudes, opinions, and emotions. In this paper we present an analytical and comparative study of different researches conducted on sentiment analysis in social networks using machine Learning. This study analyzes in more detail the preprocessing steps which are very important in sentiment analysis process success and are the most difficult especially in the case where the comments are written in not structured language. Keywords: Sentiment analysis
Machine learning Preprocessing
1 Introduction In the past ten years, many social network sites (Facebook, Twitter, Instagram etc.) have increased the presence on the web. These sites have an enormous number of users who produce a massive amount of data which include texts, images, videos, etc. According to [1], the estimated amount of data on the web will be about 40 thousand Exabytes, or 40 trillion gigabytes, in 2020. Analysis of such date could be valuable. There are many different techniques for data analytics on data collected from the web, with sentiment analysis a prominent one. Sentiment analysis (see for example [2]) is the study of people’s attitudes, emotions and options, and involves a combination of text mining and natural language processing. Sentiment analysis focuses on analyzing text messages that hold people opinions. Examples of topics for analysis include opinions on products, services, food, educations, etc. [3]. Twitter is a popular social media platform where a huge number of tweets are shared and many tweets contain valuable data. As [4] reported: in March 2014, active © Springer Nature Switzerland AG 2020 M. Ben Ahmed et al. (Eds.): SCA 2019, LNITI, pp. 63–72, 2020. https://doi.org/10.1007/978-3-030-37629-1_6
64
Z. Nassr et al.
Arabic users wrote over 17 million tweets per day. There are huge numbers of tweets generated every minute and many of them in the Arabic language. Topics about health services appear frequently on Twitter trends. The aims of this paper is to introduce a new Arabic data set on health services for opinion mining purposes. Also, to explain the process of collecting data from Twitter, preprocessing Arabic text and Annotating the data set. After collecting and annotating the dataset, some data processing tasks are applied, such as feature selections, machine learning algorithms and deep neural networks. The efficiency of these methods are assessed and compared. The objective of this work is to develop a comparative and statistical study of researches on sentiment analysis in these last few years. The paper is organized as follow: Sect. 2 develops the sentiment analysis background. Section 3 presents the related works. Section 4 discusses our comparative study and some statistical analysis. Finally, the conclusion and the research future works are detailed in Sect. 5.
2 Background In general, there are two types of approaches for opinion detection and sentiment analysis, those that are based on the lexicon and those that are based on machine learning. Lexicon-based approaches use a dictionary of subjective words. This dictionary can be general as for example Vader [4], SentiWordNet [4], Opinion Finder [4], or it is inferred from the corpus studied. Unsupervised approaches to sentiment classification can solve the problem of domain dependency and reduce the need for annotated data for training. Approaches based on machine learning consist of assigning data so-called “features” to a classification algorithm for learning. The latter generates a model that is used for the test part of the learning. This type of approach includes two aspects: extracting features and learning the classifier. The main features used are: Single word (Unigram), Bigrams, Trigrams, part of speech (POS) and polarity. The main classifiers always used in different researches are SVM (Vector Machine Support), naive Bayes, Maximum entropy and logistic regression. Most researchers consider that only adjectives carry subjectivity, while others believe that certain adverbs, nouns and verbs may also contain subjectivity. 2.1
Main Concepts
Nowadays, sentiment analysis has become one of the most active areas of research in automatic language processing. Its objective is to define automatic tools capable of extracting subjective information from text in natural languages such as opinions and feelings, in order to create a structured knowledge that can be used by decision support systems. In Sentiment Analysis [7], an opinion is defined by referring to a quintuple (o; a; so; h; t) that consists of:
Machine Learning for Sentiment Analysis: A Survey
65
– Object ‘o’, which is the opinion target. It can be a product, service, topic, issue, person, organization, or an event. – Aspect ‘a’, which is the targeted attribute of the object ‘o’. – Sentiment orientation ‘so’ that indicates whether an opinion is positive, negative or neutral. – Opinion holder ‘h’, which is the person or organization that expresses an opinion. – Time ‘t’: the moment in which this opinion is expressed. Sentiment Analysis can be investigated at three levels of granularity, namely document level, sentence level, and aspect level [8]. Its application is a highly challenging research area that involves different complex tasks. The most studied tasks are subjectivity classification, sentiment classification, lexicon creation, aspects extraction, aspect sentiment classification and opinion spam detection [9]. Under this innovative communication scenario, sentiment analysis solutions are faced to many problems related to traditional sentiment analysis, automatic natural language processing and also new and complex challenges. This complexity is mainly due to variants messages characteristics such as: there are usually very short, but rich in semantic [10], are not structured, writhed in variant languages and contain noise. Among the characteristics mentioned above, natural language texts analysis is a key element on which work the sentiment analysis community. 2.2
The Preprocessing Task in Sentiment Analysis
Data in sentiment analysis is highly susceptible to inconsistency, redundancy and noisy. The pre-processing phase attempts to improve this data quality and reduce its complexity [24] to allow knowledge discovery and extraction. It is a very important as it decides on the effectiveness of the other steps that follow [11]. Its steps cover syntactic correction, data cleaning, data transformation, stemming…. Preprocessing phase is faced to several problems which are related to the sentiment analysis context. Indeed, Words belonging to different parts of speech must be treated according to their linguistic role (adjective, nouns, verbs, etc.). The Word style (bold, italic and underline) is not always available in online social media platforms and is often replaced by some language conventions. The lengthening of words like “it’sseeeeeerious” (commonly known as expressive elongation or Word stretching) is an example of new language conventions that are today very popular in online platforms [12]. Other problems are related to additional terms such as the abbreviation expressions that are additional paralinguistic elements used in non-verbal communication in online social networks [13]. The Hashtags witch are widely used in online social networks [14, 15] to express one or more specific feelings. The distinction between sentiment hashtags and subject hashtags is a challenge that must be properly addressed for polarity classification. And the emoticons which are introduced as nonverbal expressive components in the written language [16] to reflect the role played by facial expressions in speech.
66
Z. Nassr et al.
Other very important preprocessing task is to detect and analyze the Uppercase letters since the positive and negative expressions are commonly reported by the uppercase of certain specific words (for example, ‘#StarWars was UNBELIEVABLE!’) to express the intensity of the user’s feelings [17].
3 Related Works Review Several studies and variety of approaches have been developed on sentiment analysis. This section highlights the objective of these studies and their outcome. Alowaidi et al. [18] analyzed the behavior of Tunisians during the Revolution through publications on Facebook pages, 260 posts were used as sets of training, evaluation and testing, classifiers Support Vector Machines (SVM) and naive Bayes (NB) have been used in the experiment. The result showed that the SVM algorithm had a better result than NB. Alaybaand et al. [19] collected about 126,959 Arabic Tweets on opinions about health services. The data was classified into positive and negative classes using three supervised machine learning algorithms: Naïve Bayes, Vector Support Machine and Logistic Regression. The SVM obtained better result, about 91%. EL abdouli et al. [20] presented a new way to the sentiment analysis on Twitter data written in English, French and Moroccan dialect, during the preprocessing phase they manually created a Python file to transform words written in Moroccan dialect or in a Tamazight Berber dialect into standard Arabic. The data were classified in positive and negative classes using the symbols of emotion and for the classification they applied the naive classifier Bayes that gave 69% of precision. Abinash et al. [20] mentioned that social networks reviews and blogs are very important resources for analysis and decision making. They used classification and clustering to analyze reviews and adopted four different machine learning algorithms that are: Naïve Bayes, SVM, Maximum Entropy, and Stochastic Gradient Descent for classifying the human sentiments. Umar et al. [21] proposed a negation handling approach based on linguistic features which determine the effect of different types of negation. The proposed approach improved the accuracy of negation identification and overall sentiment analysis. AkshayAmolik et al. [22] proposed a highly accurate model for sentiment analysis of tweets about reviews of upcoming Bollywood or Hollywood movies. They used Naïve Bayes and SVMs to detect opinions polarity. The average precision obtained was 75% and the average recall was 65%. Rouby et al. [23] analyzed the impact of some preprocessing steps such as noise rejection, stemming, and normalizations in the user’s comments analysis in facebook. The preprocessing operations involve tokenization, stop words elimination, normalization, stemming, negation words, emotions, and intensifiers words. The results showed that the performance was also improved when using negations and emotions. Nabil et al. [24] classify 10,006 ArabicTweets manually annotated using Amazon Mechanical Turk (AMT) service. They applied a wide range of machine learning algorithms (SVM, MBN, BNB, KNN, stochastic gradient descent) to classify the data in four classes.
Machine Learning for Sentiment Analysis: A Survey
67
Manar et al. [25] proposed a novel framework for events and incidents’ management in smart cities, the framework uses stemming techniques to extract the intelligence needed from Arabic social media feeds. Five supervised machine learning algorithms were applied, i.e., Support Vector Machines (SVM), Polynomial Networks (PNs), Rachio, K-NN, and Naïve Bayes classifiers. The SVM and Polynomial Networks gives the better results, around 96.49%. Ali Hasan and Sana Moin [26] compared three sentiment analyzers to determine the approach with the highest rate of accuracy on tweets in Urdu, the tweets were translated into English during the pre-processing phase. The result showed that the TextBlob and WSD parsers are better than the SentiWordNet approach.
4 A Comparative Study 4.1
Criteria
In our comparative study, we regroup and synthesize the researches done in the last few years. The comparison criteria adopted are; – Paper objective: It means the objective and the context of the paper – Dataset: Size and Source of dataset. – Preprocessing steps: In this study, we focused on the most important steps in preprocessing tasks. • Data cleaning that deal with the noisy data; • Normalization: which allow to generate consistent word forms (Stemming, Lemmatization, Normalizing repeated letters and Replaces languages • Lexicon development, which focuses on the informal language of online social networks such as Emoticons, acronyms, Stop words dialect, dialect…. – Language: It shows if the language translation is used or not • Before Preprocessing: refer to the original comments language • After Preprocessing: the translation language. – Algorithm: It mean which algorithms is used: NB, SVM, KNN, NN… – Accuracy: this parameter gives the precision obtained. 4.2
Analysis and Discussion
Table 1 shows that most studies have worked on political data because the opinions made are very important and their analysis can make the difference in this area, others have focused on health and education and with less importance the Smart City field. From dataset point of view, we can observe that most studies have treated small databases for better accuracy improvement. The most used data source is twitter with 72% as it is the easiest platform to analyze and its APIs facilitate the recovery of data unlike Facebook APIs, which no longer allow recovering publication and comments since the past year. The preprocessing tasks are very important in sentiment analysis and are key element to its success.
68
Z. Nassr et al. Table 1. Summary of sentiment classification works.
Ref
Objectif
Dataset
Preprocessing Data cleaning
[21] Political views analysis (election)
Twitter
[22] Introduce a framework to help health officials [23] Present a framework for classifying tweets using a hybrid classification scheme [25] Sentiment regarding revolution in Tunisia [26] Moroccan tweets analysis
Twitter
[27] The impact of the mobile phone usage in education [28] Opinions on health services
Twitter
[29] Suicide sentiment prediction
Twitter
Algorithm Accuracy Normalization
Lexicon development
Slang words, URL, Stemming special characters and Symbols Emojis, URL, Hashtag, Stemming Username, Stop words
SVM NB
76.00% 79.00%
LSTM
Lemmatization Slang dictionary
SVM
96% A 89% R 86% F1 85%
Facebook Stop words
Stemming
NB SVM
74.05% 75.31%
Twitter
Stopwords of standard Arabic, French and English Number- punctuationstop words
Elongated words
NB
69
_
Emoticons – acronyms – interjection Moroccan dialect to Arabic _
Non-health-related opinions- UsernameURLs- hashtagspunctuation Punctuation
_
_
SVM NB HMM NB SVM
−91% −86% −89% 85% −91%.
_
Vocabulary related to suicide Stop words
SVM
82.9%
NB SVM
94% 95%
_
MSP
91.2%
_
_
SVM
Stemming
Stop words emoticon
L2R2 RNN SVM
Twitter
Twitter
Tokenization, Stop Words, hashtags, RT, punctuation, URL, white space
[30] Semantic Arabic SA Twitter
URLs, (RT), usernames, numbers, single Arabic letters, non-letter characters, punctuation Twitter Stop-words, URL, RT, Usernames Facebook Numbers, punctuation, Twitter stopwords
[22] SA using the model MSP [31] Analyze Arab educational sentiment Twitter [25] A hybrid Incremental learning approach for Arabic tweets
URLs, non-Arabic words and numbers, diacritical marks
Lengthening
72.47 75.19 74.94
The analysis of all this researches allows as confirm that almost all studies have applied: “data cleaning” to eliminate noisy data, the commonly eliminated data are: URLs, Hashtags and specials characters; “data Normalization”: in this task when the language is a structured, most researches apply directly the Natural Language data Processing techniques to standardize and normalize words; otherwise, they attempt to adapt this process to the language context and realize some steps semi automatically or manually (stemming) The development of Lexicon is used only in cases where we deal
Machine Learning for Sentiment Analysis: A Survey
69
with dialects and the most common dictionary developed is the stop word one since there are widely used in text and can affect heavily the sentiment analysis and classification. This study clearly shows that Naive Bayes and SVM algorithms are more efficient than other algorithms. SVM is best suited to large scales and data; however, NB remains better than SVM when the technical selection functions are used. From the Table 1 we can see that most of the work focused on structured language and especially on English texts and very few studies were interested in written opinions in unstructured languages because of it preprocessing difficulties. We can notice that for this type of comments, there is no standards process and most of works offer a case study that reflect the research context and language constrains. 4.3
Statistical Review
In this section, we present our finding in form of graphs, to give a statistical review of the most important parameters. The graph in Fig. 1 presents the most analyzed social media platform.
Fig. 1.
As we can see, for the source or DataSet we found that the most analyzed social media platform is Twitter with 71%. In the graph below, we present the type of language written by users in social networks (Fig. 2).
70
Z. Nassr et al.
Fig. 2.
We can observe that, the most of the work deals with English language, and very few studies were interested in written opinions in unstructured languages such as the Moroccan dialect. It is noted that the most used classification technique is the naïve algorithm bye and SVM support vector machine (Fig. 3).
Fig. 3.
Machine Learning for Sentiment Analysis: A Survey
71
5 Conclusion and Future Work Sentiment analysis plays an essential role in decision-making in different fields such as politics, digital marketing (product and service evaluation), and for studying social phenomena. Because of its high value for practical applications, there has been an explosive growth in research in academia and applications in the industry. However, there is a remarkable lack of treatment on unstructured languages such as the Arabic dialect “Darija as an example”, yet These dialects represent a rich source of information especially that they are the most used by the population on nonprofessional social networks. This lack may be due to the difficulty of processing these languages, especially in terms of pretreatment, something that pushes us to take up the challenge and try to fill this gap in order to exploit a little used wealth.
References 1. Zheng, L., Wang, H., Gao, S.: Sentimental feature selection for sentiment analysis of Chinese online reviews. Int. J. Mach. Learn. Cybern. 9(1), 75–84 (2018) 2. Kušen, E., Strembeck, M.: Politics, sentiments, and misinformation: an analysis of the Twitter discussion on the 2016 Austrian presidential elections. Online Soc. Netw. Media 5, 37–50 (2018) 3. Aswani, R., Kar, A.K., Ilavarasan, P.V., Dwivedi, Y.K.: Search engine marketing is not all gold: Insights from Twitter and SEOClerks. Int. J. Inf. Manage. 38(1), 107–116 (2018) 4. Collins, M., Karami, A.: Social media analysis for organizations: Us northeastern public and state libraries case study. arXiv preprint arXiv:1803.09133 (2018) 5. Morente-Molinera, J.A., Kou, G., Peng, Y., Torres-Albero, C., Herrera-Viedma, E.: Analysing discussions in social networks using group decision making methods and sentiment analysis. Inf. Sci. 447, 157–168 (2018) 6. Jianqiang, Z., Xiaolin, G., Xuejun, Z.: Deep convolution neural networks for twitter sentiment analysis. IEEE Access 6, 23253–23260 (2018) 7. Liu, B.: Sentiment analysis and opinion mining. Synth. Lect. Hum. Lang. Technol. 5(1), 1– 167 (2012) 8. Ma, Y., Peng, H., Cambria, E.: Targeted aspect-based sentiment analysis via embedding commonsense knowledge into an attentive LSTM. In: Thirty-Second AAAI Conference on Artificial Intelligence, April 2018 9. Arif, M.H., Li, J., Iqbal, M., Liu, K.: Sentiment analysis and spam detection in short informal text using learning classifier systems. Soft. Comput. 22(21), 7281–7291 (2018) 10. Al-Smadi, M., Al-Ayyoub, M., Jararweh, Y., Qawasmeh, O.: Enhancing aspect-based sentiment analysis of Arabic hotels’ reviews using morphological, syntactic and semantic features. Inf. Process. Manage. 56(2), 308–319 (2019) 11. Cirqueira, D., Pinheiro, M.F., Jacob, A., Lobato, F., Santana, Á.: A literature review in preprocessing for sentiment analysis for Brazilian Portuguese social media. In: 2018 IEEE/WIC/ACM International Conference on Web Intelligence (WI), pp. 746–749. IEEE, December 2018 12. Sohrabi, M.K., Hemmatian, F.: An efficient preprocessing method for supervised sentiment analysis by converting sentences to numerical vectors: a twitter case study. Multimedia Tools Appl. 1–20 (2019)
72
Z. Nassr et al.
13. Soong, H.C., Jalil, N.B.A., Ayyasamy, R.K., Akbar, R.: The essential of sentiment analysis and opinion mining in social media: introduction and survey of the recent approaches and techniques. In: 2019 IEEE 9th Symposium on Computer Applications & Industrial Electronics (ISCAIE), pp. 272–277. IEEE, April 2019 14. Haddi, E., Liu, X., Shi, Y.: The role of text pre-processing in sentiment analysis. Procedia Comput. Sci. 17, 26–32 (2013) 15. Chiong, R., Fan, Z., Hu, Z., Adam, M.T., Lutz, B., Neumann, D.: A sentiment analysisbased machine learning approach for financial market prediction via news disclosures. In: Proceedings of the Genetic and Evolutionary Computation Conference Companion, pp. 278– 279. ACM, July 2018 16. Krouska, A., Troussas, C., Virvou, M.: The effect of preprocessing techniques on Twitter sentiment analysis. In: 2016 7th International Conference on Information, Intelligence, Systems & Applications (IISA), pp. 1–5. IEEE, July 2016 17. Hasan, A., Moin, S., Karim, A., Shamshirband, S.: Machine learning-based sentiment analysis for twitter accounts. Math. Comput. Appl. 23(1), 11 (2018) 18. Hamouda, S.B., Akaichi, J.: Social networks’ text mining for sentiment classification: the case of Facebook’statuses updates in the ‘Arabic Spring’era. Int. J. Appl. Innov. Eng. Manage. 2(5), 470–478 (2013) 19. El Abdouli, A., Hassouni, L., Anoun, H.: Sentiment analysis of moroccan tweets using naive bayes algorithm. Int. J. Comput. Sci. Inf. Secur. (IJCSIS) 15(12) (2017) 20. Khan, F.H., Bashir, S., Qamar, U.: TOM: Twitter opinion mining framework using hybrid classification scheme. Decis. Support Syst. 57, 245–257 (2014) 21. Tartir, S., Abdul-Nabi, I.: Semantic sentiment analysis in Arabic social media. J. King Saud Univ.-Comput. Inf. Sci. 29(2), 229–233 (2017) 22. Safeek, I., Kalideen, M.R.: Preprocessing on Facebook data for sentiment analysis (2017) 23. Wang, X., Zhang, C., Ji, Y., Sun, L., Wu, L., Bao, Z.: A depression detection model based on sentiment analysis in micro-blog social network. In: Pacific-Asia Conference on Knowledge Discovery and Data Mining, pp. 201–213. Springer, Heidelberg (2013) 24. Martinez, L.S., Hughes, S., Walsh-Buhi, E.R., Tsou, M.H.: “Okay, We Get It. You Vape”: an analysis of geocoded content, context, and sentiment regarding e-cigarettes on Twitter. J. Health Commun. 23(6), 550–562 (2018) 25. Corredera Arbide, A., Romero, M., Moya Fernández, J.M.: Affective computing for smart operations: a survey and comparative analysis of the available tools, libraries and web services. Int. J. Innov. Appl. Res. 5(9), 12–35 (2017) 26. Rodríguez-Martínez, M., Garzón-Alfonso, C.C.: Twitter health surveillance (THS) system. In: Proceedings of IEEE International Conference on Big Data, vol. 2018, p. 1647. NIH Public Access, December 2018 27. Asghar, M.Z., Kundi, F.M., Ahmad, S., Khan, A., Khan, F.: T-SAF: Twitter sentiment analysis framework using a hybrid classification scheme. Expert Syst. 35(1), e12233 (2018) 28. Amolik, A., Jivane, N., Bhandari, M., Venkatesan, M.: Twitter sentiment analysis of movie reviews using machine learning techniques. Int. J. Eng. Technol. 7(6), 1–7 (2016) 29. Wang, X., Zhang, C., Ji, Y., Sun, L., Wu, L., Bao, Z.: A depression detection model based on sentiment analysis in micro-blog social network. In: Pacific-Asia Conference on Knowledge Discovery and Data Mining, pp. 201–213. Springer, Heidelberg, April 2013
A Survey on Hand Modalities and Hand Multibiometric Systems Farah Bahmed1(&) and Madani Ould Mammar2 1
Department of Computer Science, Ahmed Zabana University of Relizane, Relizane, Algeria [email protected], [email protected] 2 Department of Electrical Engineering, University Abdelhamid Ibn Badis, Mostaganem, Algeria [email protected]
Abstract. Nowadays, hand biometric systems are very used to perform personal recognition as well as for airports as for companies. These systems can be used also for payment to avoid paying using cash or bank card. This chapter presents a survey of different multibometric hand systems developed in the literature. Various biometric modalities offered by the hand and that can be used as an alternative to the fingerprint are presented. Concepts and techniques of multibiometric hand systems which are used to improve recognition accuracy are also detailed. Examples of some commercialized systems are finally exposed. Keywords: Multibiometry control
Hand biometry Finger recognition Access
1 Introduction Hand biometric systems are commonly used in various domains as access control, attendance monitoring, passengers control in airports and even payment. In fact, biometric systems are very practical since a user doesn’t need to remember a password, and doesn’t need to be careful about the steal of his password too. These systems allow also time gain since personal recognition is performed automatically without any human intervention and in only few seconds. Recognition using hand is in reality a very old technique, since prehistoric people were signing their drawings with their handprints, and Babylonian and ancient Chinese people were later also signing their commercial contracts with their handprints. In USA, a hand geometry system was used in the 1996 Atlanta Olympic Games for the access control of the Olympic Village, while Walt Disney World Park in Florida has also used in the 2000’s a finger geometry system to entrance control and to avoid ticket fraud. Nowadays, laptops use fingerprint or even palm vein to secure access. Historically, hand systems used to utilize only one modality to authenticate or identify a person. Now, the new researches direction is to combine many modalities to overcome some limitations of classical unimodal or monomodal systems. © Springer Nature Switzerland AG 2020 M. Ben Ahmed et al. (Eds.): SCA 2019, LNITI, pp. 73–88, 2020. https://doi.org/10.1007/978-3-030-37629-1_7
74
F. Bahmed and M. Ould Mammar
The further sections will present basic concepts of biometrics then different concepts of hand biometry. Thus, different hand modalities will be described, and principles of multibiometric systems will be explained. Finally, some practical hand systems available in the market will be exposed.
2 Hand Biometric Systems 2.1
General Biometrics Concepts
As seen in previous section, a biometric system consists to use one or more biometric modalities to perform personal recognition. But what does it mean exactly “recognition”? And what does we mean exactly by “biometric modality”?. Biometrics is composed of two words, “bio” which means life in Greek, and “metrics” which means “measure”. So, in field of security, biometrics means to collect measures of human’s characteristics to perform person recognition. In France, Alphonse Bertillon (1853–1914), the developer of the judiciary anthropometry, used to collect palmprints, footprints and finally fingerprints to resolve criminal cases. Personal recognition in any biometric system can be of two kinds: • Identification: Consists to answer the question: “Who is this person?”. So, this type of recognition implies 1-vs-N comparisons. Based on comparison results, system returns the decision “Accepted” or “Denied”. • Verification or Authentication: Consists to answer the question: “Is this person really who he/she claims to be?”. So, this type of recognition implies 1-vs-1 comparison. Based on comparison result, system returns the decision “Accepted” as genuine user, or “Rejected” as imposter. A biometric modality can be a physiological characteristic (face, iris, fingerprint, etc.), or a behavioral characteristic (gait, handwriting, etc.), or a biological characteristic (DNA, saliva, etc.). To build a biometric system two phases are required: 1. Enrollment or Registration: For every genuine or authorized user, acquisition and preprocessing are performed, then features are extracted and stored in a database as templates. 2. Recognition: From a querying user X, acquisition and preprocessing are performed, then extracted features are compared to templates to take a decision. Each phase is composed of a series of steps or modules: 1. Acquisition or Capture: Image is acquired using an acquisition device containing one or more sensors. The sensor is usually a classical 2D visible light sensor (as a digital camera or a webcam), while some studies use an infrared sensor or combine both sensors, and more rarely, a 3D sensor. 2. Preprocessing: Modality is extracted from the image, and various operations are performed to facilitate further features extraction. 3. Features extraction: A features extraction method is applied on extracted modality. In enrollment phase, these features are stored as templates.
A Survey on Hand Modalities and Hand Multibiometric Systems
75
4. Comparison or Matching: In recognition phase, features extracted are compared to templates to perform biometric recognition. Thus, to evaluate performance of any system, standard metrics are defined [1]: • FAR (False Acceptance Rate): Is the rate of imposters accepted by the system. FAR ¼
Number of accepted imposter claims total number of imposter claims
ð1Þ
• FRR (False Rejection Rate): Is the rate of genuine users rejected by the system. FRR =
Number of rejected genuine claims total number of genuine claims
ð2Þ
• EER (Equal Error Rate): Is the rate at which both FAR and FRR are equal. It can be obtained with threshold graph of FRR and FAR values. • CIR or CRR or TSR (Correct Identification Rate or Correct Recognition Ratio or True Success Rate): Is the rate of users correctly identified by the system. CIR =
Number of correct matches total number of identification attempts
ð3Þ
• DI (Discriminability Index): Is the measurement of separability between genuine and imposter scores. It is defined as: lg li ffi DI ¼ qffiffiffiffiffiffiffiffiffiffi 2 2 rg þ ri 2
ð4Þ
Where lg and li are the means and rg and ri are the standard deviations of the genuine and imposter scores, respectively. A biometric system is said efficient when EER, FAR and FRR are minimum and when CIR and DI are maximum. Another commonly method used to illustrate authentication system performance is to plot the Receiver Operating Characteristic (ROC) curve, which illustrates relation between FAR and FRR for different decision threshold values. Figure 1 Depicts examples of FAR and FRR curves and ROC curve. 2.2
Hand Recognition Systems
Additionally to general concepts about biometrics, hand’s systems present some other specifities. Firstly, generally hand systems proposed in the literature are developed for verification purpose, not for identification purpose. Secondly, in preprocessing stage, hand is detected and extracted from the background. This operation can be performed using region-based or contour-based segmentation techniques. Generally, other operations are performed in this step, as orientation correction of the hand and detection of
76
F. Bahmed and M. Ould Mammar
fingers’ tips and valleys which are considered as keypoints. Thus, hand systems can be categorized into two main categories: • Contact-based: Acquisition is performed using guiding parts as pegs or using a flat surface as a scanner to force the user to put his hand in a certain manner. • Contact-free or contactless or peg-free: Acquisition is performed without any pegs or surface, the user has only to put his hand in front of a sensor. Note: Some studies use hand acquisition systems known as “contact-free”, while in fact they force the user to put his palmer/dorsal hand-side on a certain surface to acquire the dorsal/palmer side image of the hand. Figure 2 illustrates samples of some popular public hand databases used in different multibiometric studies.
(a)
(b)
Fig. 1. Curves used to represent performance of biometric systems: (a) EER finding (b) ROC curve
Bosphorus H DB [2] HKPU Contactless FK DB (V1.0) [3] (a)
HKPU Contactless H Dorsal DB [4] (b)
HKPU Contact free 3D/2D Hand DB (V2.0) [7]
Technocampus H DB [5,6]
IIT Delhi Touchless PP DB [8]
(c)
Fig. 2. Some public databases used for evaluation by multibiometric systems: (a) contact-based, (b): contact-free but using a surface and (c) contact-free.
A Survey on Hand Modalities and Hand Multibiometric Systems
77
3 Hand Modalities What makes the hand so useful to biometric recognition is its richness with parts that can be used as biometric traits. Figure 3 illustrates these parts. Hand modalities can be classified into four great classes: Geometry, Contour, Surface and Vein. In next sections, a review of every class of modalities is performed, and then a review of some state-of-art multibiometric hand systems is proposed.
(a) Fingerprint
(b) Palmprint
(f) Fingervein
(c) Hand Geometry
(g) Finger Back Knuckle print
(d) Hand Countour
(h ) Finger surface
(e) Palm Vein
(i) Finger Inner-Knuckle print
Fig. 3. Different modalities provided by the hand
3.1
Geometry
This modality has been used commonly since 1970, and it is still now used especially in companies for access control or attendance monitoring of employees. Features describing hand geometry are generally fingers’ widths, fingers’ length, palm widths, fingers’ area, fingers’ perimeter etc. Ratios can also be used to derive new features. Some studies propose a selection or reduction approach to decrease amount of features. The most challenging task in such system is to combine the best set of features with the best classifier to obtain the best accuracy rate possible. Table 1. Comparison of some state-of-art methods (Geometry) Ref. Nb of subjects
Contact- Surface free? used?
[9]
No
[10] 100 [11] 100 (CASIA) 144 (GPDS) 137 (IITD) [12] 60 [13] 250
50
Nb of features
Classif. method
Results
–
16
Yes No No Yes No
No – – No –
34 50
62
Yes
Yes
3
Weighted Euclidean – Distance SVM CIR = 96.2% LDA CIR = 97%, EER = 4.64% CIR = 100% CIR = 98%, EER = 4.51% New proposed CIR = 100%, distance EER = 0.59% Euclidean distance FAR = 2%, FRR = 1%
78
F. Bahmed and M. Ould Mammar
A famous contact-based system has been proposed in [9] using pegs and extracting 16 features. [10] have proposed a set of 34 hand geometry features for contact-free recognition. [11] have studied efficiency of geometry features using Genetic algorithm and Local Discriminant Analysis (LDA) to reduce number of features from 400 to 50 on average for both contact-based and contact-free recognition. More recently, [12] have designed a contact-based system which makes use of 62 features, and introduced new features like crookedness of fingers, while [13] have studied the case of nonseparated fingers, and used 3 triangles features. This modality presents the advantage of good acceptance by users, especially when the acquisition is contact-free, but the inconvenient of not being unique, since members of the same family can have very similar hand geometry characteristics. Aging and some diseases as arthritis can modify also this kind of features. Table 1 summarizes a comparison between above studies. 3.2
Contour (or Shape)
Consists to use hand’s contour features to perform biometric recognition. As for geometry, a selection or reduction method can also be used to decrease features’ set size. [2] have proposed a contact-based system which compares the use of Hausdorff distance and Independant Components Analysis features. [14] have used high order Zernike moments to capture hand shape features. [15] have introduced a contact-based system using circular graph, for the case of non separated fingers, while for contact-free applications, [16] have proposed to use a new shape recognition method called Coherant Distance Shape Contexts, CDSC. Table 2 presents more details. This class of modalities presents the same advantages and inconvenients as geometry features. It can be noted also that contour features are not very used in literature in comparison to other modalities. Hand shape is more used for hand gesture and hand pose recognition purpose than for biometric recognition purpose. Table 2. Comparison of some state-of-art methods (Contour) Ref. Nb of subjects
Contact- Surface free? used?
[2] 458 [14] 40 [15] 52 [16] 200
No Yes No Yes
3.3
– Yes – Yes
Nb of features
Classif. method
Results
200 30 / 60 angle bins, 16 distance bins
ICA Euclidean distance Superposition + Matching Method of dynamic programming
CIR = 97.31% EER = 2% EER = 0.13% CIR = 99.65, EER = 0.9%
Surface (or Texture)
One of the main advantages of the hand is its large dermal surface. The aim there is to find a larger zone than the classical fingerprint and which is also wealth with lines, ridges and wrinkles to perform biometric recognition using cheaper acquisition devices. Yet in 1858, Sir William Herschel used handprints of Indian workers as a mean of recognition to manage the day of payment.
A Survey on Hand Modalities and Hand Multibiometric Systems
79
Distal Phalanx Distal Interphalangeal joint Middle Phalanx Proximal Interphalangeal joint Proximal Phalanx
Fig. 4. Finger anatomy from an Inner knuckle print and a Back knuckle print surface
Table 3. Comparison of some state-of-art methods (Surface) Ref. Nb of subjects
Contactfree?
Surface used?
Features
Classif. method
[17] 100 (PolyUpp) [19] 400
Yes No
Yes –
DWT decomposition 3D features
[18] 193 100 [22] 177 148 100 100 [20] 177 148 100 [21] 177 100 [23] 120 [24] 165
Yes
Yes
Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
No No No No No No No No No No No
Multi-orientation and multiscale features No features extraction method
No
–
EER = 1.37% FRR = 5% FAR = 5.88% Hamming distance EER = 0.23% EER = 0.30% Deep NN VA = 100% VA = 98.65% VA = 100% VA = 98% PNN EER = 0.34% EER = 1.35% EER = 3% FCFNN EER = 0.23% EER = 2% Angular distance EER = 1.09% Angular & Magnitude EER = 0% distances Normalized crossEER = 2.4% correlation
[25] 100
(PolyUpp) (UST) (PolyU2D3D) (IIT Delhi) (CASIA-BLU) (CASIA-WHT) (PolyU2D3D) (IIT Delhi) (CASIA-MS) (PolyU2D3D) (CASIA-MS) (PolyU FKP)
ELLBP
MSALBP Orientation features Orientation and magnitude features Line features
Results
RSVM
Palmprint region was largely investigated by researches, as in [17] where wavelet decomposition was used, or in [18] where multi-orientation and multi-scale features were extracted using Gabor filter. [19] have also proposed the use of 3D acquisition device, since 3D palmprints are harder to counterfeit than 2D palmprints. Also, since few years ago, the entire finger surface has became to be studied as biometric trait, as in [20] where an enhanced features extraction method, based on Local Binary Pattern (LBP) was proposed especially for finger texture recognition. Another method based on Sobel direction operators and LBP too was proposed by the same authors in [21]. Deep learning has been used for a finger texture recognition in [22]. It can be noted that the three studies used methods based on neural networks. To deal with smaller ROIs, some studies introduced a new modality: The knuckle print. Both back or dorsal side and palmer or inner side were investigated. Figure 4 Details finger anatomy. Major knuckle (middle knuckle) corresponds to the proximal knuckle, and minor knuckle (upper knuckle) corresponds to the distal knuckle. Finger
80
F. Bahmed and M. Ould Mammar
Back Knuckle Print, FBKP, was first used in [23, 24], with Gabor filter as a features extraction tool, while finger Inner-Knuckle Print, FIKP, was studied as in [25] using line features. Generally, studies work only on Major knuckle. So, hand surface is constituted of so many parts that can be used for identity recognition, and it is the most used approach in the literature. The major disadvantage is the sensitivity to injuries (scars, cuts and burns) which modify epidermal appearance. Presence of dirt or moisture can also alter recognition efficiency. 3.4
Vein
As seen, previous modalities have several disadvantages. Hand’s geometry and contour are not unique and can be modified by some diseases, and hand’s surface can be altered by physical problems. Hand vascular network is unique to everybody even among identical twins as fingerprint, but additionally cannot be affected by diseases, aging, cuts and scars. Veins offer also more security, because they are internal to human body and cannot be falsified.
Table 4. Comparison of some state-of-art methods (Vein) Ref. Nb of subjects
Contact- Surface free? used?
Features
[26] 207
Yes
No
(2D)2LDA
[27] 100
Yes
No
[28] 105 (HKF) 85 (Tw. U.) [29] 312 (HKF) 636 (SDU) [30] 50
Yes No Yes Yes No
Yes – Yes Yes –
Classif. method
Results
Minimum EER = 0.34% distance classifier Relative distance and Absolute distance FRR = 1.5%, angles FAR = 2% Coeffs. of Radon transform Proposed elastic EER = 2.86% matching EER = 0.69% Orientation map guided Proposed method EER = 0.38% curvature method EER = 1.39% RCNN-3 RCNN-3 CIR = 89.43%
[26] have designed a contact-less palm vein recognition with a modified version of the LDA((2D)2LDA). [27] Have proposed to use relative distances and angles between a reference point and features points of the hand palmer vein. Moreover, finger vein was also investigated as in [28] where vein anatomy was extracted using curvatures in Radon space, and in [29] where a new anatomy analysis method was proposed. [30] have proposed the use of an improved convolutional neural network to perform vein recognition. Veins are considered to be the most powerful class of modalities; it allows reaching an EER very close to 0% in many systems, using only palm vein or 1 or 2 fingers, when for other modalities, it requires using 4 or 5 fingers to reach such accuracy. Contact-free payment systems use generally hand vein [31]. The inconvenient of such systems is that they require special infrared sensors, which are more expensive than visible light sensors.
A Survey on Hand Modalities and Hand Multibiometric Systems
81
4 Multibiometrics Old biometric systems were using only one modality, one capturing sensor, one features extraction method, etc. These classical systems present in practice some limitations, as relative low recognition rate or inability to collect sometimes the modality from the user, as fingerprints from manual workers. The solution to overcome these limitations is generally multibiometrics. Multibiometrics combines use of several biometric systems and each signal source is considered as a subsystem of the whole multibiometric system [1]. There are six main kinds of multibiometric systems: • Multisensor systems: One modality is captured using several sensors. E.g., fingerprint can be captured using both optical and compactive sensors. • Multi-instance systems: The same biometric modality is captured several times. E.g., the capture of the index and the middle of the same hand. • Multisample systems (or multi-impression): Is a variant of multi-instance systems, where one modality has to be captured several times. E.g., the capture of the face can be captured from several angles of view. • Multialgorithm systems: One modality is processed but using several recognition algorithms. E.g., comparison of both minutiae and texture of fingerprint. • Multimodal systems: Several modalities are combined to perform identity recognition. E.g., recognition using both hand geometry and palmprint. • Hybrid systems: Constituted of several of above-mentioned systems. Data provided by different subsystems can be fused at several levels of the architecture of the multibiometric system: • Sensor level: Consists in fusion of captures to form a new capture. This kind of fusion requires a multisensor system. • Feature (Or representation) level: Consists in fusion of subsystems’ features’ sets. Sometimes, a simple concatenation of features’ vectors can be applied. • Score level: Consists in fusion of scores provided by each subsystem after comparison. Generally, scores are normalized, then a fusion is carried out using well known function rules or using or more elaborate classifier. • Decision level: Each subsystem generates a decision according to its own score, and these decisions will be used to take a final decision. Generally, the lower the fusion level is achieved, better are the performance. For score fusion, normalization is a necessary step, since scores’ distributions of subsystems are rarely compatible. E.g.: the distribution of A subsystem can be [0,1] while the distribution of B subsystem is [1, 10]. Most used normalization methods are: • Minimax normalization: It is used when distribution scores bounds of the subsystem are known, i.e. S 2 [min, max]. The new scores distribution will be [0, 1]. From a score S, the new normalized score is obtained as follows: S0 ¼
S min max min
ð5Þ
82
F. Bahmed and M. Ould Mammar
The bounds min and max can be estimated using a training set. • Z-score normalization: It is calculated using arithmetic mean l and standard deviation r of the data. The normalized score is obtained using: S0 ¼
Sl r
ð6Þ
l and r parameters can be estimated using a training set also. • MAD normalization: the two previous methods are subject to noise because of the use of min, max, l and r. To overcome this problem, the median absolute deviation, MAD, which use the median of the scores can be used: S0 ¼
S median MAD
ð7Þ
Where: MAD = median (|S – median|). • Tanh normalization: it s a robust and efficient normalization method, which is based on Hampel’s estimators: S0 ¼
S lgenuine 1 tanh 0:001 þ1 2 rgenuine
ð8Þ
Where lgenuine and rgenuine are the mean and standard deviation of the genuine scores’ distribution. After score normalization, fusion of normalized scores of subsystems is achieved to obtain an overall score which will be used to take the final decision. If N denotes the number of subsystems, Si obtained scores are combined using a set of rules: • Sum rule: the fused score is the sum of scores’ subsystems: S¼
XN i¼1
Si
ð9Þ
• Product rule: the fused score is the product of scores’ subsystems: S¼
YN i¼1
Si
ð10Þ
• Maximum rule: the fused score is the maximum of scores’ subsystems: S ¼ maxðSi Þ; i ¼ 1; . . .; N
ð11Þ
A Survey on Hand Modalities and Hand Multibiometric Systems
83
• Minimum rule: the fused score is the minimum of scores’ subsystems: S ¼ minðSi Þ; i ¼ 1; . . .; N
ð12Þ
• Weighted sum rule: A weight is assigned to each sub-system according to its performance, and the fused score is the weighted sum of scores’ subsystems: S¼
XN i¼1
w i Si
ð13Þ
Where wi denotes weight assigned to subsystem number i. Weights shall also be computed using a training set. An alternative to fusion using rules is fusion using a classifier. This classifier could be for instance a Support Vector Machine, or a K-Nearest Neighbors or an Artificial Neural Networks classifier, which input’s takes scores of each sub-system as a vector S = [S1, S2, …, SN] and provides as output a decision “Accepted” or “Rejected”. The classifier is designed using a prior learning stage with training scores, while fusion using rules doesn’t need any learning stage (except for W.S. rule). For fusion at decision level only decisions, the most used methods are: • Majority vote: the decision is “Accepted” if most of subsystems return the decision “Accepted”. • AND logical operator: the decision is “Accepted” if all subsystems return the decision “Accepted”. This method leads to a low overall FAR and a high FRR. • OR logical operator: the decision is “Accepted” if at least one subsystem returns the decision “Accepted”. This method leads to a low FRR and a high FAR. 4.1
Comparison Study of Multibiometric Systems
Last years, various multibiometric hand systems have been proposed in the literature. Table 5 summarizes a comparison of some state-of-art researches. All presented systems are multimodal and some are hybrids. For fusion level, some studies have used several fusion methods. In this case, only the method showing the best accuracy is cited. As it can be seen, studies are very different each one from another and a huge number of methodologies were investigated. [32] Have proposed a system which combines 24 finger geometry features and major FBKP subspace features. The first use of minor FBKP was introduced then by [33] in combination with major FBKP using 1D Log-Gabor filter, and later, [34] have designed a system which makes use of minor FBKP and major FBKP. Combination use of FIKP, Finger geometry and palmprint was proposed in [35] with decision level fusion, while [36] have designed a special acquisition device which allows the capture and combination use of fingerprint, fingervein and FIKP. A bimodal system using geometry and shape has been developed in [37], and another bimodal system that uses fusion of palmprint and hand shape at feature level was proposed in [38], which is a rarely type of fusion level used in multimodal systems.
84
F. Bahmed and M. Ould Mammar
An EER equal to 0% was reached by [39], who designed a special acquisition device which allows the use of 5 biometric modalities: Hand geometry, palmprint, palmvein, FIKP and finger vein. More recently, [40] have proposed the use of minor FIKP for contact-less recognition. The combination major FIKP, minor FIKP and finger geometry has shown the best accuracy using a kind of decision level fusion. As Table 5 shows, a large scale of multibiometric systems have been proposed, each one making use of certain modalities, with certain features extraction methods, with a certain fusion level. Some studies developed its own acquisition device, while others tested their proposed methodology on standard databases. Score fusion level is the most used, since it is easy to implement and most of systems’ architecture and features extraction methods don’t allow a fusion at lower level. Multibiometric systems show error rate equals or almost equals 0%, even among complete unconstraint systems, in contrary of classical unimodal systems which don’t reach generally this performance (See Tables 1, 2, 3 and 4). According to Table 5 too, it can be noted that using cheap visible light devices instead of infrared devices and keeping in the same time very low error rate is possible, if two or more modalities are combined. Table 5. Comparison of some state-of-art multibiometric systems Ref. Biometric modalities
Nb of subjects
Contactfree?
Surface used?
Fusion level Results
[32] FG + Major FBKP
105
Yes
Yes
[35] FIKP + FG + PP
190
Yes
Yes
[33] Major FBKP + Minor FBKP
202
Yes
Yes
[39] HG + PP + PV + FIKP + FV 136
Yes
No
[36] FP + Major FIKP + FV
378
No
–
[37] HG + HS
50 (JUET) No 240 (IITD) Yes
– No
[34] Major FBKP + Minor FBKP + FG [38] HS + PP [40] Major FIKP + Minor FIKP + FG
150
Yes
Score level (Combined rule) Decision level (AND rule) Score level (W.S. Rule) Score level (Sum rule) Score level (W.S. Rule) Score level (Combined rule) Score level (product rule) Feature level Decision level (Hierarchical)
Yes
139 Yes 100 (IITD) Yes
No No
EER = 1.39%
FRR = 0.898% FAR = 2.52 e−6 EER = 0.16% EER = 0.00% EER = 0.109% EER = 0.31% EER = 0.52% EER = 0.44% EER = 7% FRR = 0% FAR = 0.12%
PP = Palmprint, PV = Palm Vein, FIKP = Finger Inner Knuckle Print, FG = Finger Geometry, FBKP = Finger Back Knuckle Print, HG = Hand Geometry, FV = Finger Vein, HS = Hand Shape
Different multibiometric hand systems are already commercialized. IDEMIA, the world leader in biometric security systems proposes the MorphoWaveTM Compact, a
A Survey on Hand Modalities and Hand Multibiometric Systems
(a)
85
(b)
Fig. 5. Example of multibiometric systems: (a) MorphoWaveTM Compact, (b) M2-FuseID™
multi-instance 3D contact-less access control, which makes use of four fingerprints of either the right or the left hand to perform authentication or identification. It is a wallmounted version of the MorphoWaveTM Tower, developed especially for companies. M2SYSTM proposes a multibiometric system too, M2-FuseID™ finger reader, which is a bimodal system combining fingerprint and finger vein, to perform also authentication or identification. Using the capture of finger vein allows protection against fake and spoofed fingerprints since it is a liveness detection method, and allows in the same time to ensure high level of security for some business transactions. Figure 5 illustrates MorphoWaveTM Compact and M2-FuseID™ systems. Some multimodal softwares are also available in the market. First, DERMALOG ABIS by DERMALOG, which uses fingerprints, face, iris, palmprints and even DNA. Secondly, MBIS SUITE by IDEMIA, which makes use of face, fingerprints, palmprints, Iris and even tattoos. The aim of these systems is to collect and store as many as possible biometric traits of criminals in order to identify them in further crimes. It can be noted that most of hand’s modalities investigated in scientific researches are not utilized widely in practice in commercial products, unlike the classical fingerprint and finger vein/palm vein, even for authentication task. Palmprint is used as a part of DERMALOG ABIS and MBIS SUITE systems, only because in some crime scenes police officers cannot find clear fingerprints, but find exploitable palmprints. According to [41], in the period 2007–2012 hand geometry has got only 5% world biometry market parts and multibiometry has got only 3% of the market, while fingerprint remains the most used modality with almost 60% of the market. So, it appears to be very interesting for commercial companies to integrate new modalities in their hand systems, and to carry out new combination modalities. Designing friendly contact-less hand systems should also be considered to satisfy hygienic requests.
5 Conclusion In this chapter, an overview of different hand biometry techniques was presented. Different hand biometric modalities were detailed, and unimodal systems using these modalities were exposed too. Principles of multibiometrics were introduced, and a comparison of well-known multibiometric systems of the literature was also performed. Experimental results have proved that these systems outperform classical unimodal
86
F. Bahmed and M. Ould Mammar
systems. Some commercial systems were also presented, and it was seen that for criminal applications many biometric traits are already used, among them palmprint. So, we can say that hand recognition is very useful for a large variety of security applications, and huge progress has been accomplished in this field of research these last years, and it should take more parts in world biometry market in the future, due to its high acceptability by users, and the number of exploitable modalities it contains.
References 1. Naït-Ali, A., Fournier, R.: Signal and Image Processing for Biometrics. ISTE Ltd and Wiley, New Jersey (2012) 2. Yoruk, E., Konukoglu, E., Sankur, B., Darbon, J.: Shape-based hand recognition. IEEE Trans. Image Process. 15(7), 1803–1815 (2006) 3. Kumar, A.: Importance of being unique from finger dorsal patterns: exploring minor finger knuckle patterns in verifying human identities. IEEE Trans. Info. Forensics Secur. 9(8), 1288–1298 (2014) 4. Kumar, A., Kwong, Z.X.: Personal identification using minor knuckle patterns from palm dorsal surface. IEEE Trans. Info. Forensics Secur. 11(10), 2338–2348 (2016) 5. Faundez-Zanuy, M., Mekyska, J., Font-Aragonès, X.: A new hand image database simultaneously acquired in visible, near-infrared and thermal spectrums. Cognit. Comput. 6 (2), 230–240 (2014) 6. Font-Aragonès, X., Faundez-Zanuy, M., Mekyska, J.: Thermal hand image segmentation for biometric recognition. IEEE Aerospace Electron. Syst. Mag. 28(6), 4–14 (2013) 7. Kanhangad, V., Kumar, A., Zhang, D.: Contactless and pose invariant biometric identification using hand surface. IEEE Trans. Image Process. 6(3), 1014–1027 (2011) 8. Kumar, A.: Incorporating cohort information for reliable palmprint authentication. In: 6th Indian Conference Computer Vision Graphics and Image Processing, pp. 583–590. India (2008) 9. Ross, A., Jain, A.: A prototype hand geometry based verification system. In: Proceedings of 2nd Conference on Audio and Video Based Biometric Person Authentication, pp. 166–171. Washington (1999) 10. Guo, J.M., Hsia, C.H., Liu, Y.F., Yu, J.C., Chu, M.H., Le, T.N.: Contact-free hand geometry-based identification system. Expert Syst. Appl. 39(2012), 11728–11736 (2012) 11. Luque-Baena, R.M., Elizondo, D., López-Rubio, E., Palomo, E.J., Watson, T.: Assessment of geometric features for individual identification and verification in biometric hand systems. Expert Syst. Appl. 40(2013), 3580–3594 (2013) 12. Klonowski, M., Plata, M., Syga, P.: User authorization based on hand geometry without special equipment. Pattern Recogn. 73(2018), 189–201 (2018) 13. Khaliluzzaman, Md., Mahiuddin, Md., Monirul Islam, Md.: Hand geometry based person verification system. In: 2018 International Conference on Innovations in Science, Engineering and Technology (ICISET), Chittagong, Bangladesh. IEEE (2018) 14. Amayeh, G., Bebis, G., Erol, A., Nicolescu, M.: Peg-free hand shape verification using high order zernike moments. In: Proceedings of the 2006 Conference on Computer Vision Pattern Recognition Workshop (CVPRW 2006), New York, USA. IEEE (2006) 15. Bakina, I., Mestetskiy, L.: Hand shape recognition from natural hand position. In: 2011 International Conference on Hand-Based Biometrics, Hong Kong, China. IEEE (2011) 16. Hu, R.-X., Jia, W., Zhang, D., Gui, J., Song, L.-T.: Hand shape recognition based on coherent distance shape contexts. Pattern Recogn. 45(2012), 3348–3359 (2012)
A Survey on Hand Modalities and Hand Multibiometric Systems
87
17. Prasad, S.M., Govindan, V.K., Sathidevi, P.S.: Palmprint authentication using fusion of wavelet based representation. In: 2009 World Congress on Nature & Biologically Inspired Computing (NaBIC), Coimbatore, India. IEEE (2009) 18. Ma, F., Zhu, X., Wang, C., Liu, H., Jing, X.-Y.: Multi-orientation and multi-scale features discriminant learning for palmprint recognition. Neurocomputing 348, 169–178 (2019) 19. Zhang, B., Li, W., Qing, P., Zhang, D.: Palm-print classification by global features. IEEE Trans. Syst. Man Cybern. Syst. 43(2), 370–378 (2013) 20. Al-Nima, R.R.O., Dlay, S.S., Al-Sumaidaee, S.A.M., Woo, W.L., Chambers, J.A.: Robust feature extraction and salvage schemes for finger texture based biometrics. IET Biometrics 6 (2), 43–52 (2016) 21. Al-Nima, R.R.O., Abdullah, M.A.M., Al-Kaltakchi, M.T.S., Dlay, S.S., Woo, W.L., Chambers, J.A.: Finger texture biometric verification exploiting multi-scale sobel angles local binary pattern features and score-based fusion. Digital Sig. Process. 70(2017), 178–189 (2017) 22. Omar, R.R., Han, T., Al-Sumaidaee, S.A.M., Chen, T.: Deep finger texture learning for verifying. IET Biometrics 8(1), 40–48 (2019) 23. Zhang, L., Zhang, L., Zhang, D.: Finger-knuckle-print: a new biometric identifier. In: 16th IEEE International Conference on Image Processing (ICIP), Cairo, Egypt. IEEE (2009) 24. Zhang, L., Zhang, L., Zhang, D., Zhu, H.: Online finger-knuckle-print verification for personal authentication. Pattern Recogn. 43(7), 2560–2571 (2010) 25. Liu, M., Tian, Y., Li, L.: A new approach for inner-knuckle-print recognition. J. Vis. Lang. Comput. 25(2014), 33–42 (2014) 26. Lee, Y.-P.: Palm vein recognition based on a modified (2D)2 LDA. Sig. Image Video Process. 9(1), 229–242 (2013) 27. Hu, Y.-P., Wang, Z.-Y., Yang, X.-P., Xu, Y.-M.: Hand vein recognition based on the connection lines of reference point and feature point. Infrared Phys. Technol. 62, 110–114 (2014) 28. Qin, H., He, X., Yao, X., Li, H.: Finger-vein verification based on the curvature in Radon space. Expert Syst. Appl. 82(2017), 151–161 (2017) 29. Yang, L., Yang, G., Yin, Y., Xi, X.: Finger vein recognition with anatomy structure analysis. IEEE Trans. Circ. Syst. Video Technol. 28(8), 1892–1905 (2018) 30. Wang, J., Wang, G.: Hand-dorsa vein recognition with structure growing guided CNN. Optik 149(2017), 469–477 (2017) 31. Keyo - Secure Biometric Network for Access Control, Ticketing and Payments. https://keyo. co/payments. Accessed 12 September 2019 32. Kumar, A., Ravikanth, Ch.: Personal authentication using finger knuckle surface. IEEE Trans. Info. Forensics Secur. 4(1), 98–110 (2009) 33. Kumar, A.: Can we use minor finger knuckle images to identify humans? In: 2014 IEEE Fifth International Conference on Biometrics, Columbus, OH, USA. IEEE (2012) 34. Usha, K., Ezhilarasan, M.: Fusion of geometric and texture features for finger knuckle surface recognition. Alexandria Eng. J. 55(2016), 683–697 (2016) 35. Zhu, L.-Q., Zhang, S.-Y.: Multimodal biometric identification system based on finger geometry, knuckle print and palm print. Pattern Recogn. Lett. 31, 1641–1649 (2010) 36. Kang, W., Chen, X., Wu, Q.: The biometric recognition on contactless multi-spectrum finger images. Infrared Phys. Technol. 68(2015), 19–27 (2015) 37. Sharma, S., Dubey, S.R., Singh, S.K., Saxena, R., Singh, R.K.: Identity verification using shape and geometry of human hands. Expert Syst. Appl. 42(2015), 821–832 (2015) 38. Chen, W.-S., Wang, W.-C.: Fusion of hand-shape and palm-print traits using morphology for bi-modal biometric authentication. Int. J. Biometrics 10(4), 368–390 (2018)
88
F. Bahmed and M. Ould Mammar
39. Michael, G.K.O., Connie, T., Teoh, A.B.J.: A contactless biometric system using multiple hand features. J. Vis. Commun. Image Represent. 23(7), 1068–1084 (2012) 40. Bahmed, F., Ould Mammar, M., Ouamri, A.: A multimodal hand recognition system based on finger inner-knuckle print and finger geometry. J. Appl. Sec. Res. 14(01), 48–73 (2019) 41. Singla, S.K., Kumar, S.: A review of data acquisition and difficulties in sensor module of biometric systems. Songklanakarin J. Sci. Technol. 35(5), 589–597 (2013)
A Social Media Ontology-Based Sentiment Analysis and Community Detection Framework: Brexit Case Study Moudhich Ihab(&), Loukili Soumaya(&), Bahra Mohamed(&), Hmami Haytam(&), and Fennan Abdelhadi(&) Faculty of Sciences and Technologies of Tangier, LIST Laboratory, Abdelmalek Essaâdi University, Tétouan, Morocco [email protected], [email protected], [email protected], [email protected], [email protected]
Abstract. Nowadays social media content is the fuel of almost all kinds of domains, due to its rich and ever-increasing quantity of data. Digging this content can lead to extracting valuable information that can enhance business products and services, politic decisions, socio-economic systems and more. To this end, sentiment analysis and community detection represent two of the main methods used to analyze and comprehend human interactions within social media. Also, to get meaningful results, filtering social content is needed, here where domain ontology can be a great assistant in collecting specific data, as it describes the domain’s features and their existing relationships. This current work depicts our social media analysis Framework, where we apply lexiconbased and machine learning approaches to extract expressed sentiments of social media users toward a subject, and also used a community detection algorithm to highlight the formed groups within the network. Besides, the resulting Framework not only focuses on analyzing textual data (by taking into account the negation and sentence POS tags), but also visual content shared by users, such as images. For the testing purpose of our Framework, we chose to analyze the British exit (“Brexit”) case by collecting ontology-based data from Twitter and Reddit, and it had some promising results. Keywords: Sentiment analysis Opinion mining Community detection Classification Lexicon Machine Learning Social media analysis Ontology
1 Introduction Currently, we live in an information-sharing era where people can express their opinion more freely, and this result in the exponential growth of data available on the internet, which touch a variety of domains, making it a priceless asset to rely on for developing and enhancing all facet of our eco-system. In the current time, one minute is sufficient to have more than 456,000 tweets sent, 527,760 photos shared on Snapchat, more than 4,146,000 videos viewed on YouTube, just to name a few. People tend to freely share their views online, to offer their © Springer Nature Switzerland AG 2020 M. Ben Ahmed et al. (Eds.): SCA 2019, LNITI, pp. 89–103, 2020. https://doi.org/10.1007/978-3-030-37629-1_8
90
M. Ihab et al.
recommendations about a service, or just generally to express their opinions with others. In parallel, we all acquired the reflex to look up what is said about a subject in particular before forming our own opinions. That is exactly why analyzing users’ sentiments on social media is efficient and provides valuable assets to understand and develop our life experiences. It is also quite interesting that the community detection matter shifted recently because the geographical limits between individuals simply vanished through social media. This allows studying new types of relationships and communities formed through users’ interactions. But due to these massive amounts of data available online, applying sentiments analysis or community detection techniques on a domain of interest means filtering online data accordingly, that is where ontologies come in handy; they allow us to represent the subject and its entities, and therefore gather only relevant information. In our work, we grouped lexicon-based and machine learning approaches along with community detection algorithm in one Framework, so we can understand in-depth the impact of a subject on social media users, and we chose the British Exit as a field of study because it is one of the subjects of matter in the time of developing the Framework.
2 Related Work Kaushik et al. [1] have proposed before, sentiment analysis of social media data. They worked on mining general web content and analyzing the sentiment that it holds. The different approaches of this field have been studied thoroughly, by Medhat et al. [2]. Their work centered mainly around comparing all the different sub-methodologies used in the field; whether it’s under the machine learning approach or the Lexicon-based one. This latter has been discussed and implemented by Chopra et al. [3] in their paper. They worked on creating an algorithm that analyzes the sentiment contained in sentences, using the Lexicon approach. Sentiment Analysis using Machine Learning has also been proposed several times. The work of Gaikwad et al. [4] is an illustration of it. They focused on developing a Naïve Bayes classifier, that divides data into positive or negative. They also heavily worked with ontologies. They relied on the domain ontology to identify the feature they were meaning to analyze if it was present. There’s also the paper submitted by Wang et al. [5]. They contributed to the subject by developing a Recurrent Neural Network for the classification of diseases based on symptoms contained in tweets. For community detection over Social Media, Lim et al. [6] discussed it in their paper that describes their work aiming to find communities with common interests using following links of celebrities. Musto et al. [7] proposed a solution to extract sentiments from tweets through a Framework. They developed a multi-level Framework that collects data, contextualizes them and analyzes them. They published two study cases: one of violence in Italy, and the other about the L’Aquila Earthquake. Also, another Framework with the same idea is by Bouktaib et al. [8]. They intended to predict the results of French elections in 2017 strictly based on the citizens’ sentiments about the candidates. Their results were accurately correct.
A Social Media Ontology-Based Sentiment Analysis
91
The concept of smart cities is very popular at the moment and has been for the last few years. That is because it allows better management of the city and entails that every aspect of it is improved thanks to the system. Using sentiment analysis for that goal has been explored by Ahmed et al. [9], as they discussed the state-of-the-art of this particular topic.
3 Main Concepts In this chapter, we’ll discuss several concepts that we explored and incorporated into our Framework. 3.1
Sentiment Analysis
Xiang et al. [10] define sentiment analysis as “a special type of text mining with the focus on identification of subjective statements and contained opinions and sentiments, particularly in consumer-generated content on the Internet”. Sentiment analysis is a subfield of natural language processing that provides a methodology for detecting the sentiment in non-structured information on computer. A given text can be analyzed as an entire document, a sentence or only a part of a sentence [11]. Typically, we perform a preliminary treatment on the text with tools such as stemming, tokenization, speech tagging, entity extraction and relationship extraction. Then there are other steps like the subjectivity/opinion detection, the detection of polarity, the detection of intensity, the detection of a specific emotion, or sentiment sensing about a specific aspect. There are three main approaches to implement sentiment analysis: either a lexiconbased one [12], or by applying machine learning [13], or by combining these two in what we call the hybrid approach [14]. Lexicon Based Sentiment Analysis. In sentiment analysis, a lexicon is a dictionary of words and their correspondent polarity scores. It can either be domain-specific and have a theme, or be general and contain as many terms as possible. The main idea behind the lexicon-based approach is the following: build a lexicon (or use other already built solutions), create a bag-of-words from the text we want to analyze, use preprocessing techniques to clean the input text, and finally calculate the sentiment score, which equals the average of each word’s score from the lexicon. Machine Learning Based Sentiment Analysis. Machine Learning is an application of artificial intelligence that allows systems to learn and improve automatically from the experience without being explicitly programmed. For the learning process, the method can either be supervised (i.e., creating an inferred function from labeled data), or unsupervised (i.e., learning from its mistakes). For sentiment analysis, we create a model by training the classifier with a dataset of labeled examples. Hybrid Sentiment Analysis Approach. It is a combination of the two previous approaches to obtain the best accuracy in the results. The idea is to have the best of both approaches. That is because, in case we don’t have a dataset for the machine
92
M. Ihab et al.
learning approach, or it is inaccurate, it allows to train the model using data scored with the Lexicon-based method. As a result, we have a robust classifier model. 3.2
Community Detection
The existence of communities in a network corresponds to the presence of groups of vertices (nodes) more strongly connected to each other than with the others vertices of the graph; these are related classes, which have a higher density than the graph in its totality. In social networks, analyzing these communities can highlight valuable information, as it can show a subject of interest, common activities of a group, etc. while in the Web, a group of connected sites is often the case of relatively similar discussed themes, using this information can help to improve search engines. Community detection algorithms can be divided into two main categories. The first one focuses on the extraction of information about the actors of the network, while the second helps to understand the structure and characteristics of the social network. Community Detection Algorithms – Actor’s Information. These algorithms highlight the most important actors and the strategic positions of the network. Their main objective is studying the network centrality. Here is a brief description of the implementation of these algorithms: • Degree Centrality: The actor with the largest number of connections is the most central node of the graph. • Betweenness Centrality: The nodes that are most often on the shortest paths between the other nodes are considered to be the central nodes. • Closeness Centrality: The central nodes are those with the smallest average length of paths linking one node to another. Community Detection Algorithms – Network Structure. In addition to centrality, several measures provide the information needed to understand the distribution of individuals and activities within the social network and to measure its flexibility in communicating and distributing messages. The algorithms that exploit these measures fall into two categories: • Hierarchical algorithms: the main purpose of these algorithms is the construction of a hierarchical community tree called Dendrogram, which is a tree of denser communities from top to bottom. • Heuristic algorithms: are based on heuristics related to the community structure of the network and the characteristics of the community. Hereafter, a table (Table 1) in which we listed a comparison of the most used algorithms in the community detection field.
A Social Media Ontology-Based Sentiment Analysis
93
Table 1. Algorithm
Compatible with oriented graph Edge-Betweenness True Leading-Eigenvector False Fast-Greedy False
Compatible With weighted graph True False True
Manipulate multiple components True true True
Walktrap Label-Propagation InfoMAP Spinglass
True True True True
False False True False
3.3
False False True True
Complexity
#ðE2 N Þ #ðN ðE þ N ÞÞou #ðN 2 Þ #ðN log2 ðN ÞÞ #ðN 2 logðN Þ #ðEÞ #ðEÞ #ðN 3:2 Þ
Ontologies
In computer science, the most quoted definition of ontology is: “An ontology is an explicit specification of a conceptualization” [15] - Gruber. The conceptualization part of the definition represents an abstract simplified view of the world. Which means that it’s based on concepts, objects, and other entities that are alleged to exist in the area of interest, as well as between them, while the specification part, means a formal and declarative representation. In the data structure representing the ontology, the type of concepts used and the constraints of their use are explicitly declared, making of it a taxonomy that provides the machines with the capability of making sense of data.
4 Proposed Framework
Fig. 1. Global architecture of proposed Framework
Our proposed Framework is made of several modules, each with a specific task to accomplish. In a typical pipeline, the designed Framework is capable of massive data extraction using the domain’s ontology for accurate filtering. Moreover, with a set of
94
M. Ihab et al.
technologies we can preprocess the filtered data (i.e., tokenization, stemming, negation handling, etc.), identify the sentiment that it holds with the Lexicon-based and Machine Learning approaches, and also extract the formed communities, and finally displaying the results on an analytics console, in multiple formats. The above figure (Fig. 1) is a simplified representation of the architecture of the proposed Framework, along with the technologies and methods used in each module. The following section is a description of each module of the proposed Framework. 4.1
Social Media Extractor
This module is the entry point of the Framework. It is in this component that the extraction of data from social networks takes place. It establishes the connection with Twitter and Reddit, captures any content that meets the defined criteria, and stores it into the database. Before starting the data extraction, it is key to define the domain of interest’s ontology, for the Framework to gather the right data, that is related to the subject. The process of extracting data from Social Media goes through the official Twitter and Reddit APIs. – Selected data sources: the social media platforms we chose to extract data from are Twitter and Reddit. The main reason we picked Twitter is: its primary purpose is the expression of different points of view related to a subject, in a short textual format and it allows us to extract only relevant content with its Hashtags system. Also, note that the endpoints available through its official API meet our needs and do not have as many restrictions as other APIs. The second one is Reddit, and we chose to work with it because its content is UGC (User Generated Content) and it has different communities called subreddits, that contains relevant and passionate opinions, which is fitting for our study. 4.2
Text Preprocessing
Applying a lexicon-based approach in Sentiment Analysis means to work with preestablished dictionaries, so to have the text sentiment score we need to provide clean text to the algorithm, and here where the preprocessing module takes place. We start by removing any existing URLs, extra whitespaces and tab spaces, and any screen names, or numbers. The next step is to substitute emojis by their text format, along with any abbreviations (e.g. OMG becomes oh my god), and contractions (e.g. haven’t become have not). Afterward, we normalize the entire text into lower case. After the first phase, we proceed to determine which words are affected by negation in a sentence [16]. We did that so we can invert their sentiment scores in the Lexicon method. To determine the negation scope, we define a number of principles to follow: We detect the start of the negation scope through negation cues, such as: no, not, never, than we rely on part-of-speech tagging and coordinating conjunctions (e.g. and, or) to determine if the term is affected by negation, and we only invert the score of verbs, adjectives, and nouns, because they’re the meaning holders, and also because the rest likely will not exist in the dictionary.
A Social Media Ontology-Based Sentiment Analysis
95
For example, for the sentence ‘i do not like or love you.’, the words ‘like’ and ‘love’ are the ones that are affected by the negation. We got that because first, there is the negation cue ‘not’. Right after it, we have a verb, followed by a coordinating conjunction ‘or’, and a verbal phrase. The first verb, ‘like’, is going to be inverted due to the cue ‘not’. The second one, ‘love’, because it is the first verb of the verbal phrase, that is preceded by a negated verb (i.e. same nature) and conjunction. Lastly, we remove any remaining stop words (e.g. the, on, in), and punctuation, and we lemmatize the text. Lemmatization is a lexical treatment that consists of returning the canonical neutral form of a term, the one that exists in the dictionary. 4.3
Sentiment Analysis
In this module, we analyze the sentiment held by every text, following the two approaches (i.e. Lexicon and Machine Learning). Lexicon Based Approach. We used the lexicon-based approach for two types of data: textual and visual data. Text Sentiment Analysis. To calculate the sentiment score of a text, we proceed as follows: the algorithm first tokenizes the text. It retrieves the sentiment score of each term from the SenticNet dictionary [17]. For the terms that are affected by the negation, it inverts their polarity. The average of all scores obtained is calculated. It assigns the total score to the text, (see Fig. 2).
Fig. 2. Text sentiment analysis model
Images Sentiment Analysis. For each image, we extract all the bigrams ‘noun, adjective’ that it contains, using the SentiBank [18] visual sentiment ontology Framework. These obtained terms are then processed like the previous textual data, in order to get their average score using SenticNet.
96
M. Ihab et al.
Algorithm: calculateImageSentimentScore Input: image Variables: features, feature, score, counter, scores_total Output: image’s sentiment’s score (image_sentiment) 1.
features ← extractFeaturesSentibank(image);
2.
scores_total ← 0;
3.
counter ← 0;
4.
foreach feature
features do
5.
score ← calculateTextSentimentScore(feature);
6.
scores_total ← scores_total + score;
7.
counter ← counter +1;
8.
image_sentiment ← scores_total / counter;
9.
return image_sentiment;
Machine Learning Approach. For the machine learning approach, we created a recurrent neural network (RNN) that was trained with over one million labeled tweets. The label is either ‘positive’ or ‘negative’. The RNN architecture combines the following layers: – Embedding: transforms data into same-size dense vectors. – LSTM: Long Short-Term Memory networks that learn long-term dependencies, which make them perfect for usage in natural language processes [19]. – Dense: since the output of an LSTM layer is not a SoftMax [20], it is important to add the dense layer. Each text’s sentiment is then predicted by the generated model. The quality of a model is assessed based on its accuracy, which formula is: Accuracy ¼
correct amount of guesses total amount of guesses
ð1Þ
In our case, the accuracy is equal to: 90%. 4.4
Community Detection
At this module, we analyze data to identify the relations that exist between users, depending on their interactions on social media. First, we create a network in which their nodes are users and edges are retweets existing between them. We run the Edge Betweenness [21] algorithm by Girvan and Newman, which returns a list with the formed communities. The way this algorithm works is by calculating the Betweenness Centrality score between all edges and then removing the ones with the highest score. We reiterate this operation until no edge can be removed.
A Social Media Ontology-Based Sentiment Analysis
gð v Þ ¼
X s6¼v6¼t
rst ðvÞ rst
97
ð2Þ
Where rst is the total number of shortest paths from node s to node t and rst ðvÞ is the number of those paths that pass-through v. 4.5
Analytics Console
The final layer is a mean for the user to take advantage of and visualize the results of the previous modules. It displays the gathered data and the results of both sentiment analysis and community detection. The user can consult the generated reports, charts and graph from his own account. Reports and maps are used to interpret the results. This means that there is a percentage of negative, positive and neutral data, a comparison between different methods, an evolution over time, and the feeling according to the regions of the world. For the network, it is a representation of the formed communities, as well as the details of each one if you click on it. The details of each community consist of the number of the community, the number of nodes and edges it has, the influencer nodes, and the more recurrent tweets.
5 Brexit Analysis Case Study The Brexit is a term that was born in 2016 and refers to the United Kingdom’s exit from the EU (e.i. European Union). It was the former Prime Minister David Cameron who took this initiative, by organizing a referendum back in 2016, in which 51.9% of citizens chose to leave the EU. The whole process is still ongoing, and there are a lot of people who spoke up against the Brexit itself. The fact that there are individuals who want the Brexit to take place, and others who would rather remain in the EU, makes this topic highly fitting to test the Framework. That is because not only it provoked a lot of mixed feelings within the population, but it also separated the country into communities who share not only the same point of view but also the approach they see fit to solve the problem. The first step was to define the keywords which we were going to use to collect Brexit related data on Twitter and Reddit. The selected keywords are highly important and are game-changers for the accuracy of the Framework because it is crucial to gather all the relevant data and to not omit any important ones. We started by studying the different political parties in the UK and their position on the matter to build our domain ontology. From there, we gathered politicians’ usernames, the parties’ official usernames, and their used hashtags. Also, it is important to collect data using hashtags that are proper to each side of the movement, such as “Remainer”, “LeaveEU”, “RevokeArt50”, etc.
98
M. Ihab et al.
Fig. 3. Brexit domain’s ontology
The previous figure (Fig. 3) represents the Brexit domain’s ontology, which we used further on to collect data. We started the data collection on March 1st, 2019 until the current day, both through Twitter API and Reddit API. We managed to collect over 5 million tweets, retweets, quotes, replies, posts and comments, which were all stored in a Mongo database. Going forward, we submitted the data to the proposed Framework. The results have different forms, so we can understand in depth the subject. 5.1
Sentiment Analysis – Overall Pie Charts
The first charts represent the negative and positive quantity of data that is related to the Brexit, both on Twitter and Reddit, and with both approaches – Lexicon based and Machine Learning. The color green is mapped to positive, blue to negative, and yellow to neutral sentiment.
Fig. 4. Sentiment analysis approaches comparison – overall Brexit sentiment
A Social Media Ontology-Based Sentiment Analysis
99
As can be seen in Fig. 4, the majority of data is negative in this topic, which clearly indicates that people strongly dislike the Brexit idea and prefer to stay within the EU. In order for us to see which approach is more accurate and closer to real results, we looked up recent polls that were conducted by NatCen, which is the largest independent social institution in the UK, which full title is “National Centre for Social Research”. The following table synthesizes the comparison between our Framework’s results with both approaches, and NetCen’s polls answer to the question: “If there was another referendum on Britain’s membership of the EU, how would you vote?”. All results are from the same time frame [22] (Table 2). Table 2. Proposed Framework’s results comparison with NatCen’s results Leave the EU Remain in the EU NatCen’s results 44.45% 55.55% Proposed Framework – Lexicon Based Approach Results 35.01% 64.99% Proposed Framework – Machine Learning Results 45.12% 54.88%
The results show that the Machine Learning approach is more accurate, as its percentages are very close to the ones we retrieved from actual polls. 5.2
Sentiment Analysis – Results on Map
To further our understanding of the Brexit subject, and to understand which region are most likely to vote “Remain” in case of a second referendum, we generated the following map on the proposed Framework.
Fig. 5. Proposed Framework’s analysis results by region on map
sentiment
Fig. 6. Survation’s poll results by region on map
As seen in Fig. 5, most regions’ sentiments about the matter are negative, except for two regions: “East Midlands” and “Yorkshire and The Humber”. The results generated by the proposed Framework are similar to the ones established by Survation, which is a polling and market research agency with British origins [23]. Except for one
100
M. Ihab et al.
region, “East”, that has a 50% leave vote. These results are represented in the figure (Fig. 6). 5.3
Sentiment Analysis – Overtime Stacked Bars
The following figure (Fig. 7) represents the overtime variation of sentiments for the Brexit. The first bar is the proposed Framework results, and the second is the same results published by Survation. We used the results of Machine Learning approach, as it is more accurate.
Fig. 7. Sentiment analysis over time – Framework’s results comparison with Survation’s poll.
Even though there is a variation in weeks’ results, they all indicate that people strongly dislike the Brexit, and thus want to remain in the EU.
Fig. 8. Sentiment analysis over time - BMG polls on Brexit
The above figure (Fig. 8) is a representation of polls’ results [23], conducted from November 2016 to the start of April 2019. These polls had been done by BMG research, a Birmingham based research center, very trustworthy in the UK, their results go in line with the results of our proposed Framework, as they show a slight increase in the Remainers camp, rising to 51%.
A Social Media Ontology-Based Sentiment Analysis
5.4
101
Community Detection Results
Fig. 9. Simple visualization of founded communities
Figure 9 represents the formed communities within social media users who are invested in the Brexit subject. Interestingly, we have two separate subgraphs; each composed of several communities formed. This is explained by the fact that all the communities of the first subgraph (on the right), for example, are against Brexit. The difference between them is that each has a different argument, or the intensity of their remarks is of a different degree. For the second subgraph, all users want the Brexit to take place, and are disappointed to see that it is not executed. Some communities that feel betrayed and present this argument, others that are more practical and who organize petitions to sign, etc. N.B: If the “Leaver” community is much bigger, it is because we have added conditions to the Framework, such as limiting the number of tweets and retweets to be analyzed, thus a relatively small amount of random data is used, for the community detection.
6 Conclusion In our work we present a sentiment analysis Framework that integrates domain ontology along with machine learning and lexicon-based techniques, to extract sentiments expressed by social media users, also we introduced community detection analysis to understand the structure of the groups formed according to the studied case. The pipeline we used consist of filtering data using domain ontology, then a preprocessing phase take place to clean text, following, we apply lexicon-based and machine learning algorithm to identify and calculate the sentiment score, and finally, display the results in various format in our analytics console.
102
M. Ihab et al.
As a perspective for this Framework, there are many ideas we can execute to enhance the performance of our Framework and have even better results than the one we currently have. Regarding the Lexicon-based approach in the sentiment analysis section of the Framework, it could be improved by combining multiple Lexicons, so we can ensure the presence of nearly every term we want to analyze. Therefore, the accuracy will automatically get better, because the analysis would take into consideration the majority of words. Also, in case a word isn’t found in any Lexicon, we could look up the score of its synonym. Another interesting perspective is to combine the two approaches in Sentiment Analysis. Using a hybrid, one would increase the accuracy, as it takes the best of each. Training data that have been labeled using a Lexicon to train a model would be ideal. For community detection, it would be quite interesting to explore other algorithms, to see which ones of them are the quickest and most accurate. Having used the Edge Betweenness algorithm, our next step will consist of implementing InfoMap algorithm [24], as they both treat weighted and directed edges. Using machine learning for this part of the Framework is also something we are looking forward to achieving.
References 1. Kaushik, A., Kaushik, A., Naithani, S.: A study on sentiment analysis: methods and tools. Int. J. Sci. Res. (IJSR) (2015). ISSN (Online) 2319-7064 2. Medhat, W., Hassan, A., Korashy, H.: Sentiment analysis algorithms and applications: a survey. Ain Shams Eng. J. 5, 1093–1113 (2014) 3. Chopra, F.K., Bhatia, R.: Sentiment analyzing by dictionary based approach. Int. J. Comput. Appl. (0975 – 8887) 152(5), 32–34 (2016) 4. Gaikwad, A.S., Mokhade, A.S.: Twitter sentiment analysis using machine learning and ontology. Int. J. Innovative Res. Sci. Eng. Technol. 6, Special Issue 1 (2017). ISSN (Online) 2319 – 8753 ISSN (Print) 2347 – 6710 5. Wang, C.-K., Singh, O., Tang, Z.-L., Dai, H.-J.: Using a recurrent neural network model for classification of tweets conveyed influenza-related information. In: International Workshop on Digital Disease Detection using Social Media 2017 (DDDSM - 2017), pp. 33–38 (2017) 6. Lim, K.H., Datta, A.: Finding Twitter communities with common interests using following links of celebrities. In: MSM 2012 - Proceedings of 3rd International Workshop on Modeling Social Media (2012). https://doi.org/10.1145/2310057.2310064 7. Musto, C., Semeraro, G., de Gemmis, M., Lops, P.: Developing smart cities services through semantic analysis of social streams. Copyright is held by the International World Wide Web Conference Committee (IW3C2) (2014) 8. Bahra, M., Bouktib, A., Hanafi, H., El Hamdouni, M., Fennan, A.: Sentiment analysis in social media with a semantic web-based approach: application to the French presidential elections 2017. In: Innovations in Smart Cities and Applications. Proceedings of the 2nd Mediterranean Symposium on Smart City Applications (2017) 9. Ahmed, K.B., Radenski, A., Bouhorma, M., Ahmed, M.B.: Sentiment Analysis for Smart Cities: State of the Art and Opportunities. CSREA Press (2016). ISBN 1-60132-439-1 10. Ko, C.-H.: Exploring big data applied in the hotel guest experience. Open Access Libr. J. 5, e4877 (2018)
A Social Media Ontology-Based Sentiment Analysis
103
11. Kolkur, S., Dantal, G., Mahe, R.: Study of different levels for sentiment analysis. Int. J. Curr. Eng. Technol. (2015). E-ISSN 2277 – 4106, P-ISSN 2347 – 5161 12. Kundi, F.M., Khan, A., Ahmad, S., Asghar, M.Z.: Lexicon-based sentiment analysis in the social web. J. Basic Appl. Sci. Res. (2014). ISSN 2090-4304 13. Hasan, A., Moin, S., Karim, A., Shamshirband, S.: Machine learning-based sentiment analysis for Twitter accounts. Math. Comput. Appl. 23, 11 (2018). https://doi.org/10.3390/ mca23010011 14. Kolchyna, O., Souza, T., Treleaven, P., Aste, T.: Twitter Sentiment Analysis: Lexicon Method, Machine Learning Method and Their Combination (2015) 15. Gruber, T.R.: Toward principles for the design of ontologies used for knowledge sharing. Int. J. Hum. Comput. Stud. 43, 907–928 (1993) 16. Dadvar, M., Hauff, C., de Jong, F.: Scope of negation detection in sentiment analysis (2011) 17. Cambria, E., Poria, S., Bajpai, R., Schuller, B.: SenticNet 4: a semantic resource for sentiment analysis based on conceptual primitives (2016) 18. Yu, Y., Lin, H., Meng, J., Zhao, Z.: Visual and textual sentiment analysis of a microblog using deep convolutional neural networks. Algorithms 9, 41 (2016). https://doi.org/10.3390/ a9020041 19. Yin, W., Kann, K., Yu, M., Schutze, H.: Comparative Study of CNN and RNN for Natural Language Processing (2017). arXiv:1702.01923v1 [cs.CL], 7 February 2017 20. Zhang, L., Wang, S., Liu, B.: Deep learning for sentiment analysis: a survey. Wiley Interdisciplinary Rev. Data Min. Knowl. Discov. 8, e1253 (2018) 21. Polanco, X., Juan, E.S.: Text data network analysis using graph approach. In: I International Conference on Multidisciplinary Information Sciences and Technology, Mérida, Spain, October 2006, pp. 586–592 (2006). ffhal-00165964f 22. NatCen, Whatukthings, Second Referendum Vote Results, May 2019. https://whatukthinks. org/eu/questions/if-a-second-eu-referendum-were-held-today-how-would-you-vote/ 23. BMG Research, Remain or Leave The UK? Poll, April 2019. https://www.bmgresearch.co. uk/latest-eu-voting-intention-figures-show-remain-continuing-to-record-leads-over-leave/ 24. Deitrick, W., Hu, W.: Machine learning-based sentiment analysis for Twitter accounts. J. Data Anal. Inf. Process. 1, 19–29 (2013)
Fuzzy Questions for Relational Systems Rachid Mama(&) and Mustapha Machkour Faculty of Sciences, Information Systems and Vision Laboratory, Ibn Zohr University, Agadir, Morocco [email protected], [email protected]
Abstract. Databases are becoming more and more inevitable for websites and applications that handle large amounts of data, such as bank accounts (games, social networks, videos, etc.). However, they are unable to respond to humantype questions that are generally imprecise, uncertain and vague, and therefore need to be appropriately addressed. Therefore, fuzzy interrogation systems have become indispensable to represent and manage these data, and particularly facilitate the interrogation to a non-expert user. Fuzzy logic provides a powerful tool to take into account various aspects of fuzzy information. In this paper, we present a short study of fuzzy querying relational databases. We start by the design and operation of a fuzzy system and present the different types of architectures of fuzzy question database systems. Lastly, we introduce a comparison of the most pertinent features in fuzzy query systems of databases. Keywords: Relational databases Fuzzy query Flexible queries Fuzzy logic
1 Introduction Databases are created to manage huge quantities of data. Obtaining this data requires a database language such as Structure Query Language (SQL). But this language is not easy for everyone to extract data [1, 2]. An immediate consequence is that since natural language is the means of communication used by human beings innately in everyday life, human interrogation with database systems must be primarily in natural language. The purpose of a database query system is to allow users to compose human language questions and receive the answer in natural language. however, the natural question is intrinsically fuzzy [3] (this is called a fuzzy query). Example: “Find all young person whose salary is very high.”
where “young” and “high” are examples of fuzzy predicates and “very” is a modified predicate. This request cannot be processed and manipulated directly by a conventional SQL command, but it can be handled and manipulated easily by means of a fuzzy query in a flexible and humane way, based on the use of the fuzzy logic [19].
© Springer Nature Switzerland AG 2020 M. Ben Ahmed et al. (Eds.): SCA 2019, LNITI, pp. 104–114, 2020. https://doi.org/10.1007/978-3-030-37629-1_9
Fuzzy Questions for Relational Systems
105
Traditional interrogation systems have excellent interrogation capabilities but are unable to deal with vague and imprecise language terms [21], they do not have the flexibility to deal with fuzzy question. Several works have been proposed in the literature to introduce flexibility in querying databases [20]. Most of these works have used the fuzzy sets and fuzzy logic formalism [22, 23] to model linguistic terms such as (“young”, “high”) and to evaluate predicates with such terms (see Fig. 1). The main idea in this work is to extend the SQL language and add an additional layer of a classical DBMS to evaluate fuzzy predicates [4, 5, 18].
Fig. 1. Fuzzy sets representing human age and salary
In this paper, we give a short examination of fuzzy query systems for relational systems. In Sect. 2, we give a presentation a bit more detail on the design and operation of a fuzzy system. Section 3 shows architectures of fuzzy question of relational systems. We present in Sect. 4 a comparison of most suitable features in fuzzy query systems of the database. Finally, we give conclusions.
2 Fuzzy Systems 2.1
The Operation of a Fuzzy System
The principle of a fuzzy system is to be able to compute output parameters by providing the system with a set of rules formulated in natural language. A fuzzy logic system consists of three parts (see Fig. 2). The first part that will translate a sensor’s digital data into a linguistic variable is called fuzzification. Depending on the membership function created by the fuzzy system designer, we will be able to transform a quantitative sensor data into a qualitative linguistic variable) for example, data from a sensor could be age = 32 years. After fuzzification, we would have Age = 75% young, 25% middle, 0% aged).
106
R. Mama and M. Machkour
Fig. 2. Block diagram of a fuzzy system
The second part is the inference engine that will apply each of the inference rules. These rules of inference represent the knowledge that one has of the system due to the human expertise. Each rule will generate an exit command. Finally, the third step is defuzzification. This is the step to merge the different commands generated by the inference engine to give it only one output command and to transform this output linguistic variable into numeric data. Fuzzification The purpose of fuzzification is to transform a digital datum into a linguistic variable. For this, the designer of the fuzzy system must create membership functions. A membership function is a function that makes it possible to define the degree of membership of a numerical datum to a linguistic variable. To highlight the fundamental principle, consider the following table (Table 1): Table 1. Exemplary data from tbl_emps ID_EMP 1 2 3 4
NAM kamal tazi ahmed safi siham rami soufian aka
AGE 26 33 39 50
SALARY 8000 3500 4700 6000
we want to transform this numeric data into a linguistic variable, we can find several linguistic variables qualifying the Age, for example: Young, Average, Old. Now, just create a membership function of the Age to each of these variables. Since these membership functions qualify the same type of data, they can be represented on the same graph (see Fig. 3).
Fuzzy Questions for Relational Systems
107
Fig. 3. Fuzzification
Inference Engine Now that we have linguistic variables, we will be able to pass them in the inference engine. Each rule of the inference engine is written by the designer of the fuzzy system based on the knowledge available to him. A rule must be in the form if condition, then action. For example, IF the speed is high AND the distance to the fire is short THEN braking hard is a valid inference rule. In the conditions we may find several variables that are linked together by AND or OR. Indeed, conventional logic operators (AND, OR) are no longer valid in fuzzy logic. We must redefine them ourselves. The AND Operator There are several definitions of the AND operator in fuzzy logic. Among the most used, we have: The minimality operator: a AND b ¼ minða; bÞ
ð1Þ
The product operator: a AND b ¼ a:b
ð2Þ
AND fuzzy: a AND b ¼ c:minða; bÞ þ ð1 cÞ:
aþb 2
ð3Þ
The fuzzy AND has a parameter c which is between 0 and 1 and which must be set by the designer of the fuzzy system. The OR Operator As for logical AND, there are several logical OR definitions in the context of fuzzy logic: The minimality operator: a OR b ¼ maxða; bÞ
ð4Þ
The product operator: a OR b ¼ 1 ð1 aÞ:ð1 bÞ
ð5Þ
OR fuzzy: a OR b ¼ c:maxða; bÞ þ ð1 cÞ:
aþb 2
ð6Þ
108
R. Mama and M. Machkour
The NOT Operator The NOT operator in fuzzy logic corresponds to the complementary set and is defined simply: NOT a ¼ 1 a
ð7Þ
Once a list of rules of inference has been drawn up, it is enough to apply to the linguistic variables calculated in the fuzzification stage. The results of these rules can be directly related to the final stage of defuzzification. Defuzzification In the second step, we generated a set of commands in the form of linguistic variables (one command per rule). The purpose of defuzzification is to merge these commands and transform the resulting parameters into exact numeric data. There are several methods to fuzzify. Among the most used are the average maxima method and the center of gravity method. • The average maxima method (MM): The average maxima method is equivalent to taking the abscissa corresponding to the average of the abscissa having the maximum value of the membership functions as ordered. Formally, it is expressed as: R x:dx value ¼ RS with S ¼ fx; lðxÞ ¼ supðlðxÞÞg S dx
ð8Þ
• The center of gravity method (COG): It consists of taking the abscissa of the center of gravity of the surface of the result curve: R lðxÞ:x:dx value ¼ RS with S, the domain of the membership function S lðxÞ:dx
ð9Þ
Defuzzification is a tricky part to implement in a fuzzy system. Indeed, it usually consumes a lot of computing resources to be able to transform linguistic variables into digital data, because in this part, we manipulate functions. This can be a critical point in an embedded system and the choice of the defuzzification method is therefore crucial.
3 Fuzzy Querying System Architectures 3.1
Fuzzy Query to Database
A fuzzy query is a query that allows a user to describe their requests with vague and imprecise language terms (quasi-natural expressions) in the data quantization criteria and to express preferences between these criteria.
Fuzzy Questions for Relational Systems
109
These preferences can be set using conditions without exact (fuzzy) criteria. These fuzzy conditions include several possible forms, such as: fuzzy predicates (e.g., young and old), similar linguistic terms (e.g., old aged), fuzzy preferences (e.g., far more than, much less than), fuzzy comparators (e.g., a little more, approximately), fuzzy quantifiers (e.g., approximately the half, most), modified predicates (e.g., relatively, very, too much), Compound predicates(conjunction and disjunction of predicates), and so forth. Consider the table tbl_emps (Table 1), if the user asks the question: which employees earn more than 5,000, the output will show two results. This is a simple query generating a simple output. But what if we want to know if the employee in the above query is young or not (fuzzy query). The solution of the above problems can be found using the fuzzy value sets: • Age: very young, young, a little old, old… • Salary: small, medium, high, very high… 3.2
Fuzzy Interrogation System Architectures
Most fuzzy database query approaches are based on fuzzy-set theory [6, 7], which benefits from the great expressivity of the latter in modeling different types of gradual criteria and in their combination. Several works have been proposed in the literature to introduce flexibility in querying databases. Most of these works have used the fuzzy sets and fuzzy logic formalism to model linguistic terms such as (“moderate”, “medium”) and to evaluate predicates with such terms. The main idea of these systems is to extend the SQL language and add a supplementary layer of a classic database management system (DBMS) to evaluate fuzzy predicates (see Fig. 4).
Fig. 4. Fuzzy interrogation system
110
R. Mama and M. Machkour
The trapezoidal membership functions are the most used in the literature to express the vagueness of linguistic terms. However, it is difficult to use the same membership function to model a given linguistic term in all circumstances. Thus, there are no linguistic concepts with universal distributions. For example, the term “average” that characterizes the age linguistic variable can be described in Fig. 5.
Fig. 5. Two models of “average age”
The response to a fuzzy query phrase is usually a list of records, sorted by degree of match. The implementation of a fuzzy interrogation system raises various constraints [8], among which the main ones are: • Commercial DBMS do not provide tools to model vague and inaccurate linguistic terms and evaluate predicates with such terms. • When evaluating a fuzzy selection (or join) query, it not possible to directly use existing indexes. • The result of the fuzzy query is a set of rows organized according to the degree of satisfaction and result calibration, which leads to an additional cost during the evaluation. Consequently, the development of a fuzzy interrogation system can be based on to three types of architecture [8, 17]: Low Incorporation A software layer has been developed above the DBMS to retrieve all tuples from the database and to calculate the satisfaction degree for each of them against the fuzzy predicates already specified by the user. An additional step dedicated to the calibration of the result to return only those tuples whose degree of satisfaction greater than or equal to a qualitative thresholding. This naive version of Fuzzy Query basically translates the fuzzy query into a Boolean query and sends it to a DBMS to retrieve the result. For example, the fuzzy query introduced in Sect. 4. A would thus be derived in: SELECT FROM tbl emps WHRERE salary [ ¼ 5000 AND age BETWEEN 16 AND 30
Fuzzy Questions for Relational Systems
111
The advantage of this type of architecture is that it works independently of the database language because any DBMS can be used. But unfortunately, the performance is poor because all the tasks are supported by the software layer. Medium Incorporation This architecture relies on the use of DBMS mechanisms such as stored procedures to define the membership functions of fuzzy predicates and the operations necessary for computation to form the fuzzy result relationship. Its main strength lies in its better performance because the data is managed directly by the DBMS kernel. Strong Incorporation New features such as fuzzy relational operators are embedded in the very core of the DBMS. This architecture, which is obviously the most efficient in terms of query evaluation, is based on rewriting the DBMS evaluation engine which requires a very consistent implementation effort.
4 Comparison of Most Relevant Characteristic in Fuzzy Query Systems The first models are mainly theoretical models of fuzzy relational database, like that of Buckles and Petry [9, 24, 25] which paves the way for future development, and Medina [12] which introduced the model GEFRED (generalized model for fuzzy relational database) in 1994. It is an eclectic synthesis of the different models published to deal with the problem of representing and processing fuzzy information using RDBs. One of the main advantages of this model is that it consists of a general abstraction that makes it possible to deal with different approaches, even those that may seem very disparate. It is based on the generalized fuzzy domain and the generalized fuzzy relation, which include domains and classical relations respectively (Table 2). Table 2. Comparison of most relevant characteristic in fuzzy query systems Model Manage scalar data Manage nonscalar data Similarity relationship Possibility distributions Degree in attributes level Degree in tuple level
Medina Bosc Zemankova [12] [14] [11]
Prade Martinez Umano Kacprzky Buckles [10] [16] [13] [15] [9]
(continued)
112
R. Mama and M. Machkour Table 2. (continued)
Model Fuzzy modifiers Fuzzy quantifiers Fuzzy comparison operators Fuzzy group by Fuzzy joins Nesting Store fuzzy data Fuzzy queries Extension SQL language
Medina Bosc Zemankova [12] [14] [11]
Prade Martinez Umano Kacprzky Buckles [10] [16] [13] [15] [9]
Other models have been presented as the model of Umano and Fukami [13, 27, 28] who presented FOOBD in SQL, it is based on the theory of possibilities. The difference with other models is how to describe fuzzy information. And in 1985, Prade and Testmale [10] presented a model based on the theory of the possibilities of representing inaccurate or incomplete data. It allows to incorporate the “incomplete” or “imprecise” data of the theory of possibility. The data structure is like that used in the UmanoFukami model. It uses measures of possibility and necessity for the satisfaction of the conditions established in the consultation. In 1995 Bosc [14] introduced the first version of a flexible query language called SQLF, then with Jose Galindo who introduced the FSQL language like that of Bosc but presenting new approaches such as comparators, attributes and fuzzy constants. The most complete implementations were provided by Bosc and Pivert called SQLF and by Kacprzky and Zadrozny [15, 26] called FQuery in Microsoft Access. In 2016 Martinez-Cruz [16] suggested an implementation based on fuzzy logic to handle fuzzy queries and ontology for manipulate semantic queries, this is an implementation extracted from the Medina proposal. The GIFRED model, introduced by Medina, theoretically defined some features that have not yet been included in the implemented version as fuzzy quantifiers and fuzzy links. Thus, these implementations showed their powers in the flexible consultations (fuzzy request of type SELECT). Unfortunately, these tools do not allow the user to describe the schema of the fuzzy database (FDB) nor to manipulate its FDB.
Fuzzy Questions for Relational Systems
113
5 Conclusion This paper presents an examination of the fuzzy question systems. these systems enable users to extract data from the database using practically human language. Many works have been introduced to make flexible database query. Most of these works use the fuzzy set formalism to model linguistic terms and evaluate predicates with such terms. The main idea in these works is to extend the standard SQL and add a layer over relational systems to estimate these predicates. Although different variations exist, we can distinguish three types of architectures: low, medium and strong incorporation. Several models have appeared in the literature such as the GEFRED model of medina [12]. The FSQL language was developed based on the GEFRED model by Bosc and Pivert [14], and represents an extension of SQL to support flexible queries. Despite the fact that the previously described models have shown desirable properties for BDFs, none of them has completely satisfied the characteristics of a FDB model, however, two problems have not yet found definitive solutions that are appalling [8] : (i) the efficient implementation of a fuzzy query system (which raises the issue of fuzzy query optimization), (ii) how a non-expert user can be helped to define his fuzzy quantifiers, set the weights to the predicate of a complex condition, or express nested fuzzy queries. These problems can be solved if using, inter alia, machine learning techniques.
References 1. Siasar djahantighi, F., Norouzifard, M., Davarpanah, S.H., Shenassa, M.H.: Using natural language processing in order to create SQL queries. In: IEEE International Conference on Computer and Communication Engineering (ICCCE), Kuala Lumpur, Malaysia, 13–15 May 2008, pp. 600–604. IEEE (2008) 2. Simpson, H.: Dumb Robots, 3rd edn, pp. 6–9. UOS Press, Springfield (2004) 3. Tahani, V.: A conceptual framework for fuzzy query processing—a step toward very intelligent database system. Inf. Process. Manage. 13, 289–303 (1977) 4. Zemankova-Leech, M., Kandel, A.: Fuzzy Relational Databases—A Key to Expert Systems. Verlag TUV Rheinland GmbH, Köl (1984) 5. Simpson, B., et al.: Title of paper goes here if known (unpublished) 6. Bosc, P., Pivert, O.: SQLf: a relational database language for fuzzy querying. IEEE Trans. Fuzzy Syst. 3, 1–17 (1995) 7. Dubois, D., Prade, H.: Using fuzzy sets in flexible querying: why and how?. In: Proceedings of the 1996 Workshop on Flexible Query-Answering Systems, pp. 89–103 (1996) 8. Smits, G., Pivert, O., Girault, T.: Fuzzy queries and RDBMSs: towards a tighter coupling. In: LFA 2013, Reims, France (2013) 9. Buckles, B.P., Petry, F.E.: A fuzzy representation of data for relational databases. Fuzzy Sets Syst. 7(3), 213–226 (1982) 10. Prade, H., Testemale, C.: Generalizing database relational algebra for the treatment of incomplete or uncertain information and vague queries. Inf. Sci. 34(2), 115–143 (1984) 11. Zemankova, M., Kandel, A.: Implementing imprecision in information systems. Inf. Sci. 37 (1–3), 107–141 (1985)
114
R. Mama and M. Machkour
12. Medina, J.M., Pons, O., Vila, M.A.: GEFRED: a generalized model of fuzzy relational databases. Inf. Sci. 76(1–2), 87–109 (1994) 13. Umano, M., Hatono, I., Tamura, H.: Fuzzy database systems. In: Proceedings of the FUZZIEEE/IFES 1995 Workshop on Fuzzy Database Systems and Information Retrieval, Yokohama, Japan, 53–36 (1995) 14. Bosc, P., Pivert, O.: SQLf: a relational database language for fuzzy querying. IEEE Trans. Fuzzy Syst. 3(1), 1–17 (1995) 15. Kacprzyk, J., Zadrożny, S.: SQLf and FQUERY for access. In: Proceedings of the IFSA World Congress and 20th NAFIPS International Conference, vol. 4, pp. 2464–2469 (2001) 16. Martinez-Cruz, C., Noguera, J.M., Vila, M.A.: Flexible queries on relational databases using fuzzy logic and ontologies. Inf. Sci. 366(20), 150–164 (2016) 17. Urrutia, A., Tineo, L., Gonzalez, C.: FSQL and SQLf: towards a standard in fuzzy databases. In: Information Science Reference, Hershey, PA, USA, pp. 270–298 (2008) 18. Galindo, J., Urrutia, A., Piattini, M.: Fuzzy Databases: Modeling. Design and Implementation. Idea Group Publishing, Hershey (2006) 19. Andreasen, et al.: Flexible Query Answering Systems. Springer, Roskilde (1997) 20. Ma, Z.: Fuzzy Database Modeling of Imprecise and Uncertain Engineering Information. Springer, Heidelberg (2006) 21. Abadi, D., et al.: The Beckman report on database research. Irvine, CA, USA (2013). http:// beckman.cs.wisc.edu/beckman-report2013.pdf 22. Zadeh, L.A.: The role of fuzzy logic in the management of uncertainty in expert systems. Fuzzy Sets Syst. 11, 199–227 (1983) 23. Zedeh, L., Kacprzyk, J. (eds.): Fuzzy Logic for the Management of Uncertainty. Wiley, New York (1992) 24. Buckles, B.P.: F.E. Petry.: Query languages for fuzzy databases. In: Kacprzyk, J., Yager, R. (eds.) Management Decision Support Systems using Fuzzy Sets and Possibility Theory, pp. 241–251. Verlag TÜV Rheinland, Cologne (1985) 25. Buckles, B.P., Petry, F.E., Sachar, H.S.: A domain calculus for fuzzy relational databases. Fuzzy Sets Syst. 29, 327–340 (1989) 26. Kacprzyk, J., Zadrozny, S., Ziółkowski, A.: FQUERY III+: a “Human-consistent” database querying system based on fuzzy logic with linguistic quantifiers. Inf. Syst. 14(6), 443–453 (1989) 27. Umano, M.: FREEDOM-0: a fuzzy database system. In: Gupta, M.M., Sanchez, E. (eds.) Fuzzy Information and Decision Processes, pp. 339–347. North Holland Publishing Company, Amsterdam (1982) 28. Umano, M., Fukami, S.: Fuzzy relational algebra for possibility-distribution-fuzzy-relational model of fuzzy data. J. Intell. Inf. Syst. 3, 7–27 (1994)
Smart Education
Using Machine Learning Algorithms to Predict the E-orientation Systems Acceptancy Rachida Ihya1(&), Mohammed Aitdaoud1, Abdelwahed Namir1, Fatima Zahra Guerss2, and Hajar Haddani3 1
Laboratory of Information Technologies and Modeling, Department of Mathematics and Computer Science, Faculty of Sciences Ben M’Sik, University Hassan II of Casablanca, Casablanca, Morocco [email protected], [email protected], [email protected] 2 Computer Laboratory of Mohammedia, Computer Sciences Department, Faculty of Sciences and Technicals Mohammedia, University Hassan II of Casablanca, Casablanca, Morocco [email protected] 3 Laboratory of Search Optimization, Emerging System Networks and Imaging, Computer Sciences Department, Faculty of Science, University Chouaib Doukkali of El Jadida, El Jadida, Morocco [email protected]
Abstract. The orientations are the trends and behaviors that express an individual’s desire to pursue or apply oneself to a specific occupation and, together, these orientations affect the individual’s decision-making process with respect to occupational choice. This study aimed at generating an acceptance model of the e-orientation Moroccan platform “orientation-chabab.com” that can be used during the conceptual design of the future E-orientation platforms. Firstly We established a qualitative questionnaire based in the Technology Acceptance Model (TAM) as a theoretical model. Our experiment was conducted with the WEKA machine learning software by using four algorithms namely: NaïveBayes, J48, NLMT and SimpleLogistic. The results indicated that the highest classification accuracy performance is for the J48 and classifier gives us the best performance outcomes. Keywords: E-orientation
TAM Machine learning
1 Introduction The orientations are important insofar as they address an individual’s specific abilities. However, recognizing or identifying an individual’s interests and strengths is considered one of the more difficult and complicated tasks that teachers and organizational leaders face as they strive to guide students and employees to pursue appropriate careers (Ari, Vatansever, & Uzun) [1]. The orientation should begin in the school environment, as it increases student awareness of the varied natures of different school and helps to identify the specialty that are most suitable for the individual student. Such orientation is considered even more important with respect to the social environments © Springer Nature Switzerland AG 2020 M. Ben Ahmed et al. (Eds.): SCA 2019, LNITI, pp. 117–130, 2020. https://doi.org/10.1007/978-3-030-37629-1_10
118
R. Ihya et al.
associated with various careers. Institutions of higher education are responsible for developing students’ specific skills, such as being aligned with the chosen career workplace requirements. This continuity creates a connection between the actual framework of the study and the expectations of the external labor market (Milloshi) [2]. Special programs help in developing academic and professional orientations in educational environments by providing students with the appropriate support and by helping them to recognize their physical, cognitive and personal traits. The appearance of TICE in the field of education has revealed another means of orientation is through the existence of orientation platforms that considerate like a special programs. There are guidance platforms that can help a student in their choice of orientation by leading them to reflect on their tastes and abilities but also on their knowledge of the socioeconomic environment and training. Subsequently, they experienced a clean development related to the possibilities of the computer tool. Today there is a lack of current research on the acceptance of Moroccan electronic guidance systems. Thus, most of the examples focus on “Meta-model of e-orientation platforms” [3] and “Modernization of a domain E-orientation Meta-model” [4]. However, research that would focus on the acceptance prediction’s model for e-orientation system has not been previously conducted except for our previous work [34]. The TAM comprises several variables explaining behavioral intentions and the use of technology directly or indirectly (i.e., perceived usefulness, perceived ease of use, attitudes toward technology) [5]. We established a qualitative questionnaire based in TAM theoretical model, sharing with social networks, SMS sending, individual interviews in collaborations with the experts in the field of orientation. After we collected the feedback from our study sample, we use machine learning algorithms to generate our model. This study aims to predict the acceptance of the E-orientation systems by users that will help us to achieve an acceptance model of E-orientation platform “Orientationchabab.com” [6]. Since our data is tremendously increasing, it becomes difficult for us to establish a relationship between multiple features, which makes it difficult for us to manually analyzing the data for strategic decision making. Machine learning is a method of data analysis that automates analytical model building. Our database includes 256 samples. Using various Machine learning classifier algorithms, the best results were obtained by a J48 with accuracy rates of “80, 46%”. In this paper a review of literature regarding Technology Acceptance Model (TAM) and E-orientation is carried out before the methods used and the results of this study are presented. Finally, we discuss the findings of the analysis and the conclusions.
2 School Guidance and TIC TICE is a great addition to the guidance counselor in that it will enable them to use referral systems. Through TIC will enable the guidance counselor to highlight the three axes of education Orientation namely:
Using Machine Learning Algorithms
119
• Self-awareness • Knowledge of the socio-economic environment • Knowledge of the formations Orientation education, for example, provides an education of choice, that is, it aims to provide learners with methods and skills to help them be autonomous in their orientation [7]. Several orientation platforms in sight the day among its platforms we have: inscription-bachelor.com, almowajih.com, insription-ecole.com, orientation-pourtous. com, and orientation-chabab.com. The Inscription-bachelor.com [8] platform provides users accounts and guide to the sub-branches inside schools (commerce, specialized, of engineers). This orientation (guidance) is based only on the information about included on users’ diplomas. Almowajih.com [9] platform proceeds in the same manner asinscriptionbachelor.com platformexcept that it integrates also a professional orientation. Inscription-ecole.com [10] platform and Inscriptionbachelor.com platform are similar except that Inscriptionecole.com focuses more on other information: level of training (bac, bachelor,..), its option (distance, local, both). Orientation-pour-tous [11] platform and Almowajih.complat form are similar apart from Orientation-pour-tous platform guides to foreign establishments. Orientation-chabab.com platform and inscriptionbachelor.com platform provide the same functions apart from Orientation-chabab.com platform doesn’t demand a user account and it precised a field linked to training or sub-branch (engineering, agrifood, …). A comparison of his 5 platforms was made as shown in the Table 1 and we observe that the platform “orientation-chabab.com” contains a number of important features cited and it will be note platforms study:
Table 1. The E-orientation system evaluation. Indicator Inscriptionbachelor.com Almouwajih.com Inscriptionecole.com Orientationpourtous.com Orientationchabab.com
User account +
Higher education −
Static search engine −
Learning −
Search engine −
− +
+ −
− −
− −
− −
+
−
+
−
−
+
+
−
+
−
120
R. Ihya et al.
3 Theoretical Background In this section we present the TAM and using it in the e-orientation systems. 3.1
Technology Acceptance Model
TAM, proposed by Davis in 1985 [12], explains and predicts the usage of information technologies based on the Theory of Reasoned Action (TRA) [13]. The TAM includes perceived usefulness and perceived ease of use as the main influencing variables of an individual’s acceptance of information technologies [14]. Figure 1 illustrates the TAM model [15].
Fig. 1. Technology Acceptance Model (TAM).
Perceived usefulness (PU) is “the degree to which a person believes that a particular technology would enhance his or her performance”. Perceived ease of use (PEOU) is “the degree to which a person believes that using a particular technology would be effortless”. Behavioral intention (BI) refers to possible actions of individuals in the future, which can be based on forecasting people behavior [16]. The using of external variables depends on the type of research and reflects the flexibility of TAM [17]. According to “Li, Yuanquan, Jiayin Qi, and Huaying Shu” [22], attitude toward using technology is the connection between belief variables (PEOU, PU) and BI. BI is the trend of the user’s cognition about likes or dislikes to use the information systems (IS). Usage Behavior (UB) is the final IS use behavior. Thus, the TAM has been verified for several information technologies by researchers and practitioners [18–20]. 3.2
Using TAM in E-orientation Systems
Among the most used platforms in Morocco we named “orientation-chabab.com”, which is a guide for high school Moroccan graduates for access to private and public universities and colleges. However, the establishment of this e-orientation platform has never been exposed to a study that shows its acceptance by users. To develop a successful orientation system, the designer must familiarize him or herself with the specifics of that environment, as well as the typical and learned behavioral patterns that occur within it. Orientation systems need to be accessible and understandable for as many people as possible.
Using Machine Learning Algorithms
121
The acceptance and the usage of the platform “orientation-chabab.com” have been examined using TAM. To understand the behavior of the individual towards the orientation systems, it is essential to research for the factors which explain the users’ acceptance of E-orientation systems. We divided our extending TAM into three categories: The external variables, the mediator variables and the variable to predict (see Fig. 2).
Fig. 2. Theoretical model of acceptance of E-orientation systems.
External Variables: Presents the external variable that affects the decision making of an e-orientation: Individual and Social factors. Individual Factors: It consists on individual variables characteristics (gender, age, level education, formation educational, experience and resources. Social Factors: The concept of social influence is based on the subjective norm proposed in the TAM, and describes the influence of people who are important to the subject making decisions. In our study context, we are talking about the social factors that affect the acceptance and use of an e-orientation (influence of: professional categories and study’s level of parents, career professional of relatives, support of relatives, effect of relative’s and networks financial dependence) [21, 22].
122
R. Ihya et al.
As Mediator Variables, we positioned four factors: Perceived Usefulness (PU). The influence of user perception on the usability of the eorientation Platform (Behavior Intention) [23]. Perceived Ease of Use (PEOU). The influence of perceived ease of using Eorientation platform on users’ intentions to use the e-orientation platform (Behavior Intention) [23]. As a Variable to predict we have the user’s prediction of decision to accept or not using the e-orientation platform. The variables cited in our theoretical model (Fig. 2) are supported by previous literature review by experts as seen in Table 2: Table 2. The E-orientation system evaluation. Factors determine e-orientation platform intention to use Individual factors Social factors Perceived usefulness Perceived ease of use
Supported literature
References
Hung et al. 2006 Hung et al. 2006, Van Dijk et al. (2008) Hung et al. 2006, Davis, Fred D (1989) Hung et al. 2006; Davis, Fred D (1989)
[24] [24, 25] [24, 26] [24, 26]
4 Methodology The approach ATM is used to construct a questionnaire developed using a qualitative approach including the explanatory and mediating factors, as major items for questionnaire development. This approach makes it possible to adapt the theoretical constructs (the factors confirmed by the literature) to a specific content (questionnaire) according to the population studied. The study population consists of women and men aged from 18 until the age of 60, who are mainly (students, employers, and traders, farmers, unemployed and inactive). Questionnaire distribution was carried out via questionnaire sharing with social networks, sending SMS messages, individual interviews. These persons are informed about our project and they were invited to test the existing platforms during a month delay and then to answer the questionnaire (Table 3). The distribution of the questionnaire was administered online by mail or by SMS and distributed among groups, forums and social networks, and paper form by realizing direct interviews. Their returns are recorded on an Excel file in Google drive. The data collection was then carried out in April 2018 which lasted 6 months. We received 256 Returns. In this study 140 respondents were male (54.69%) as shown in Fig. 3. Those who sent more returns are those who live in big cities (72%) as seen in Fig. 4. About 237 of
Using Machine Learning Algorithms
123
Table 3. Heading level Gender Age socio-professional categories Education level Marital Status Lodgment Situation Commune Size
Font size and style - Male; Female - 18–20; 21–24; 25–45; 46–60; > 60 - Student; Farmer; Merchant; artisan; Entrepreneur; Senior, Professor, Intellectual, Supervisor; Intermediate Occupation; Employee; Worker; Unemployed; Inactive; Other - College; High school; Baccalaureate; Baccalaureate + 2; University degree, Mastery(Bac + 3or4); Master, DEA, DSS; PhD; No diploma - Single; married; divorced; widowed - At my parents; I live alone; I live with other students; I live in a couple - Big City; Small Town; Campaign; Town
Fig. 3. The returns of women and men.
Fig. 4. The size of the municipality of responders to the survey.
124
R. Ihya et al.
the respondents are between the ages of 18–45, and only (7, 42%) of the respondents are 60 years or older. Most respondents had higher education (96, 48%). Once we have collected our data, we start to examine it and work out what we can do with it. The objective we have is one of prediction: given the data we have, predict what the next person will make the decision form the E-orientation platform. 4.1
Waikato Environment for Knowledge Analysis (Weka)
Waikato Environment for Knowledge Analysis (Weka) is a suite of machine learning software written in Java, developed at the University of Waikato, New Zealand. It is free software licensed under the GNU General Public License [27]. Weka contains a collection of visualization tools and algorithms for data analysis and predictive modeling, together with graphical user interfaces for easy access to these functions. Weka supports several standard data mining tasks, more specifically, data preprocessing, clustering, classification, regression, visualization, and feature selection. All of Weka’s techniques are predicated on the assumption that the data is available as one flat file or relation, where each data point is described by a fixed number of attributes (normally, numeric or nominal attributes, but some other attribute types are also supported). In our data analysis, we have downloaded and installed the software “WEKA” version 3.9 which is available from WEKA University of Waikato website . We have exported in CSV format our data file in the “WEKA” tool which will in turn show us the 27 attributes after application of the feature selection that will allow us to implement a model to predict the acceptation of the E-orientation systems. 4.2
Feature Selection
Feature Selection (FS) is a crucial part of preprocessing step in the process of Knowledge Data Discovery (KDD). Attribute Selection, Instance Selection, Data Selection, Feature Construction, Variable Selection and Feature Extraction are some of the different names assigned to Feature Selection Algorithms (FSA). They are predominantly used for data reduction by removing irrelevant and redundant data. As mentioned in [2], feature selection improves the quality of the data and increases the accuracy of data mining algorithms by reducing the space and time complexity. Feature selection focuses on eliminating redundant and irrelevant data [28]. Information Gain Attribute Ranking is one of the simplest (and fastest) attribute ranking methods and is often used in text categorization applications, where the sheer dimensionality of the data precludes more sophisticated attribute selection techniques if A is an Attribute and C is the class, Eqs. 1 and 2 give the entropy of the class before and after observing the attribute. The Information Gain Attribute evaluates [29] the value of an attribute in the correlation between it and the OUTPUT class: “make the decision to use an electronic guidance system” and we got 26 attributes (Table 4).
Using Machine Learning Algorithms
125
Table 4. Attribute Evaluator (supervised, Class(nominal): 20 Decision_orientation): Information Gain Ranking Filter. Ranked 0.4158 0.4045 0.3693 0.3198 0.2857 0.2696 0.1923 0.189 0.1848 0.1698 0.1478 0.1427 0.1265 0.1222 0.1216 0.1152 0.1064 0.1045 0.0968 0.0706 0.0638 0.0536 0.0499 0.0432 0.03 0.0297
4.3
Attributes Perceived_ease_use Quality_information Contents Perception_risk Perceived_usefluness Form_presentation Pro_socio_category Influence_level_formation Socio_pro_categories_parents Influence_speciality_formation Influence_career_pro_relatives Level_studies Influence_networks_relatives Formation_E-orientation Level_studies_parents Age Ressource_rate Influence_formation_relatives Support_relatives Financial_dependence Lodgement_situation Type_last_school_university Experience_information_technology Size_municipality Gendre Marital_status
Classification Algorithms
Machine learning (ML) is the scientific study of algorithms and statistical models that computer systems use to effectively perform a specific task without using explicit instructions, relying on patterns and inference instead. It is seen as a subset of artificial intelligence. Machine learning algorithms build a mathematical model based on sample data, known as “training data”, in order to make predictions or decisions without being explicitly programmed to perform the task [30]. Classification is the process of assigning labels to test patterns, based on previously labeled training patterns. This process is commonly divided into a learning phase, where the classification algorithm is trained, and a classification phase, where the algorithm labels new [30].
126
R. Ihya et al.
Our research study uses different well-known classifiers, such as NaiveBayes, J48, LMT and SimpleLogistic for validating the output of “decision making for using the platform “orientation-chabab.com”. Naive Bayes Classifier: The Naive Bayes algorithm is a simple probabilistic classifier that calculates a set of probabilities by counting the frequency and combinations of values in a given data set. The algorithm uses Bayes theorem and assumes all attributes to be independent given the value of the class variable. This conditional independence assumption rarely holds true in real world applications, hence the characterization as Naive yet the algorithm tends to perform well and learn rapidly in various supervised classification problems . Naïve Bayesian classifier is based on Bayes’ theorem and the theorem of total probability [31]. Decision Tree Algorithm J48: J48 classifier is a simple C4.5 decision tree for classification. It creates a binary tree. The decision tree approach is most useful in classification problem. With this technique, a tree is constructed to model the classification process. Once the tree is built, it is applied to each tuple in the database and results in classification for that tuple. The basic idea is to divide the data into range based on the attribute values for that item that are found in the training sample. J48 allows classification via either decision trees or rules generated from them [31]. Simple Logistic Algorithm is also known as Logit Regression or Logit Model, is a mathematical model used in statistics to estimate (guess) the probability of an event occurring having been given some previous data. Logistic Regression works with binary data, where either the event happens (1) or the event does not happen (0). So given some feature x it tries to find out whether some event y happens or not. So y can either be 0 or 1. In the case where the event happens, y is given the value 1. If the event does not happen, then y is given the value of 0. For example, if y represents whether a sports team wins a match, then y will be 1 if they win the match or y will be 0 if they do not. This is known as Binomial Logistic Regression. There is also another form of Logistic Regression which uses multiple values for the variable y. This form of Logistic Regression is known as Multinomial Logistic Regression [32]. Logistic Model Tree Algorithm (LMT) is a classification model with an associated supervised training algorithm that combines logistic regression (LR) and decision tree learning. Logistic model trees are based on the earlier idea of a model tree: a decision tree that has linear regression models at its leaves to provide a piecewise linear regression model (where ordinary decision trees with constants at their leaves would produce a piecewise constant model). In the logistic variant, the LogitBoost algorithm is used to produce an LR model at every node in the tree; the node is then split using the C4.5 criterion. Each LogitBoost invocation is warm-started [vague] from its results in the parent node. Finally, the tree is pruned the text should be in two 8.45 cm (3.33”) columns with a .83 cm (.33”) gutter [33]. Once the preliminary testing is judged to be satisfactory, the classifier is available for routine use. The classifier’s evaluation is most often based on prediction accuracy (the percentage of correct prediction divided by the total number of predictions)
Using Machine Learning Algorithms
4.4
127
Classifier Accuracy Measures
There are some parameters on the basis of which we can evaluate the performance of the classifiers such as TP rate, FP rate, Precision and Recall F- Measure areas which are explained below. The Accuracy of a classifier on a given test set is the percentage of test set tuples that are correctly classified by theclassifier. True Positive (TP): If the outcome from a prediction is p and the actual value is also p, then it is called a true positive. True positive rate ¼ diagonal element=sum of relevant row False Positive (FP): If the actual value is n then it is said to be a false positive. False positive rate ¼ non diagonal element=sum of relevant row: Precision and Recall: Precision is the fraction of retrieved instances that are relevant, while recall is the fraction of relevant instances that are retrieved. Both precision and recall are therefore based on an understanding and measure of relevance. Precision can be seen as a measure of exactness or quality, whereas recall is a measure of completeness or quantity. Recall is nothing but the true positive rate for the class [56]. • Precision = diagonal element/sum of relevant column. • F-measures = 2 * precision * recall/(precision + recall) In this paper, we have used WEKA (Waikato environment for knowledge analysis) tool for comparison of NaïveBayes, J48, LMT and SimpleLogistic algorithm and calculating efficiency based on accuracy regarding correct and incorrect instances generated.
5 Results and Discussion In our research we used four classification algorithms. J48, LMT, NaïveBayes, and SimpleLogistic applied to our data [12]. In this section we compare the classification accuracy results of the four algorithms in order to choose the best between them (Table 5) at the top shows the accuracy of each algorithm. Table 5. A comparative study of classification algorithms. Classification Algorithms J48 LMT NaïveBayes SimpleLogistic
Accuracy 80,46% 75,79% 71,87% 80,12%
128
R. Ihya et al.
The performance of classification algorithm is usually examined by evaluating the accuracy of the classification. It is evident from the (Table 6) that J48 has the highest classification accuracy (80.46%) where 251 instances have been classified correctly and 5 instances have been classified incorrectly. The Second highest classification accuracy for SimpleLogistic algorithm is (80.12%). Moreover the LMT and showed a classification accuracy of (75.7813%). TheNaïveBayes algorithm results in lowest classification accuracy which is (71.87%) among the five algorithms. So the J48 outperforms the LMT, NaïveBayes, and SimpleLogistic in terms of classification accuracy. In addition to obtaining the best results, it offers very low computing times ( b} giving rise to two new leaves of this tree. This process is repeated at each leaf of the decision tree until the objects in each final leaf belong to same decision class or the permitted maximum number of cuts is reached. The best cut ck at the leaf Lk is chosen so that it should allow distinguish the maximum of Xk objects. 3.2
Attribute Reduction Methods
The attribute reduction or feature selection [32] is another data preprocessing stage required by some data mining methods, especially in classification problems, to reduce the dimensionality of processed data sets. It refers to a way of reducing the total conditional attribute set of a database to a subset of indispensable ones by removing redundant, superfluous and poor-information features and keeping all of the relevant information and interesting knowledge of this database intact [32]. Therefore, the usage of the resultant subset of indispensable attributes allows [7,13–15,32,33,37]: • Preserving the indiscernibility relation IN D(A) and so the partition U/A; • Eradicating redundant information and their sources that can engender random and illusory dependencies with the decision attribute and thus to screw up and distort the learned model; • Improving time and space complexities of classification algorithms; • Simplifying intelligent system design and implementation, especially in the industrial field (fewer needed measurement points necessarily imply a reduced cost of implementation); • Reducing the total number of outliers and missing values in input data sets. Generally, for every data set, several such reduced subsets can be generated. However, the desired target is to find the minimal ones called Reducts or Decision Reducts [7,13,37]. Besides, the set of all indispensable conditional attributes of A is denominated the Core of A and expressed by Core(A) = ∩Red(A) where Red(A) is the set of all reducts of A [25]. Determining a minimal reduct through computing and evaluating all possible reducts of A is NP-hard [7] problem. Luckily, there are some interesting heuristics permitting to find Approximate Reducts in often reasonable time.
Rough Set Supervised Machine Learning Approaches: Survey and App
411
In the following, we discuss the most reputed feature selection algorithms based on Rough Sets. 3.2.1 Rough Set Attribute Reduction Elaborated by Shen and Chouchoulas [33], the Rough Set Attribute Reduction (RSAR) or QuickReduct is an heuristic that determines a conditional attribute set reduction, which is called a super-reduct, based on dependency degree concept of RST. It consists in iteratively evaluating the improvement of dependency degree γR∪{a} (D) (see its definition given by (3)) after adding each remaining conditional attribute a to the current reduct candidate R ⊆ A, where D = {d}. The attribute which maximizes this degree is chosen to be added to the current reduct, thus giving a new reduct candidate. This process continues until the dependency degree of the decision attribute on the resulting reduct is equal to that calculated from the global set of condition attributes, i.e. γA (D). 3.2.2
Greedy Heuristic for Computing Decision Reducts and Approximate Decision Reducts The Greedy Heuristic for Computing Decision Reducts and Approximate Decision Reducts (GHCADR) is developed to select the most relevant conditional attributes of a decision system and filter out those that are dispensable, vulnerable to data disturbances and that can introduce random dependencies with decision attributes [7,13,15,32,37]. For this goal, it is relied on a generalization of reduct concept called the Approximate Decision Reduct in order to relax the requirement of preserving all dependencies between conditional attributes and decision ones. Thus, it allows to deal with random perturbations in data sets and to evade choosing features that have no impact on decision. The new generalized concept of the approximate reduct is defined as follows: Given a decision table DT = (U, A ∪ {d}) and an attribute quality measure Qd : 2A −→ IR such that it is increasing in inclusion sense, a subset R ⊆ A is named a (Qd , )-approximate decision reduct with respect to an approximation threshold ∈ [0, 1[, iff: (a) Qd (R) ≥ (1 − )Qd (A)
(6)
(b) There is no proper subset R R which verifies the previous inequality. Following a quasi-similar process to RSAR and using a quality measure Qd instead of dependency degree, this method produces an appropriate approximate reduct for each data set presented in its input. The stopping criterion of its iterative process is the condition given by (6) [15]. Among the functions Qd that can be used to assess attribute subset quality, we can mention [13,15]: • The entropy: HDT (X) = −
w∈VX
PDT (w) · log2 (PDT (w))
(7)
412
A. Hadrani et al.
• The Gini-index:
PDT (w) · (1 − PDT (w))
(8)
Disc(X) = {(u, u ) ∈ U 2 : X(u) = X(u )}
(9)
Gini(X) =
w∈VX
• The discernibility:
where X ⊆ A and for every w ∈ VX , PDT (w) is the probability of satisfying (X, w) by DT objects. 3.2.3 Greedy Heuristic for Determining Superreduct Based on RST Compared to the previous general heuristic, this method corresponds to the particular case = 0 which leads to return to the particular concept ”reduct” of the general concept ”approximate reduct”. Thus, the Greedy Heuristic for Determining Superreduct based on RST (GHDSR) is only a variant of GHCADR corresponding to this case ( = 0) [7,13,32,37]. 3.2.4 Dynamically Adjusted Approximate Reducts Heuristic ´ ezak in 2014 [15,32], the Dynamically Adjusted Proposed by Janusz and Sl¸ Approximate Reducts heuristic (DAAR) is a variant of GHCADR. It aims to decrease further chances of inclosing superfluous attributes in a reduct by relying on a new concept called random probes. Given an attribute a ∈ A, a random probe for a, which is denoted a ˆ, is an attribute artificially generated from Va such its associated random variable has the same distribution as the one associated with a, but it is independent of this later as well as of that corresponding to the decision attribute d. A probe a ˆ can be simply generated by randomly permuting a values. Compared to GHCADR and using the random probes, DAAR modifies the stopping condition by calculating automatically the correct approximation threshold needed for approximate reduct determination. Taking into account a subset of already chosen conditional attributes, random probes are employed to calculate a conditional probability which evaluates the chance of a new added attribute to increase the quality measure Qd of this subset relatively to its corresponding random probe. Consequently, a remaining attribute a is finally selected to be added to the current candidate reduct R if it verifies a})) ≥ prel , where prel ∈ [0, 1[ is an input parameter. P (Qd (R ∪ {a}) ≥ Qd (R ∪ {ˆ a})) can be estimated The conditional probability P (Qd (R ∪ {a}) ≥ Qd (R ∪ {ˆ by a subroutine that receives as input (in addition to DT , Qd , R and a) the additional parameter nP robes representing the number of probes generated for a. Using an iterative process and initializing a global variable Count to zero, this subroutine proceeds, in each iteration, to generate a random probe a ˆ and a}). If Qd (R∪{a}) ≥ Qd (R∪ compute the two measures Qd (R∪{a}) and Qd (R∪{ˆ {ˆ a}), the value of Count is increased by 1. This process is repeated nP ropes times and then the subroutine returns the estimated value of the conditional Count a})) as the quotient nP probability P (Qd (R ∪ {a}) ≥ Qd (R ∪ {ˆ robes .
Rough Set Supervised Machine Learning Approaches: Survey and App
3.3
413
Rule Induction Methods
Decision Rule Induction (RI) is one of the oldest and most effective techniques of machine learning. At the onset of data mining, this later has frequently been likened to the techniques of predictive rule induction to distinguish it from statistics and data analysis. Moreover, rule induction methods present very efficient and advantageous tools for extracting useful knowledge from data. They generate predictive models that are powerful, comprehensible, easy to implement, and do not require any previous knowledge or skill in statistics and data mining for their interpretation. In this paper, we focus on the supervised learning context considering only the following methods of decision rule induction. 3.3.1 Algorithm Quasi-optimal (AQ) Developed by Ryszard S. Michalski [20], AQ, which is also called the Algorithm of synthesis of Quasi-minimal covers, is a global inductive learning method that retrieves a predictive rule for every decision class from a given input data set. It constitutes the core of an entire family of AQi systems (AQ1,...,AQ21) that continues to expand to improve the prediction precision and the generated rule simplicity [3,36]. The general form of the AQ generated rules is as follows: if coveri then predict classi , where coveri , that constitutes the rule condition part, is a disjunction of complexes each one is a conjunction of selectors and every one of these later is a condition on an attribute with respect to a set of modalities. Given a class Ci of the training set T S ⊆ U , AQ proceeds to find an adequate coveri in the following way. It starts to select randomly a seed, which is a positive example (i.e. an example from Ci ) not yet covered by coveri , and generates a set of all possible complexes (called star) that cover this seed and none from T S − Ci . Then, AQ chooses the best complex from this star according to the criterion defined by the user and adds it to coveri . In general, the best complex is chosen so that it should cover the maximum number of positive examples and contain the minimum of selectors. This step is repeated until all positive examples are covered by coveri . Hence, AQ utilizes a flexible strategy to search quasi-minimal covers that allow to induce quasi-deterministic rules from data sets. However, it is sensitive to noisy data and produces rules depending on the specific selected seeds. 3.3.2 CN2 Proposed by Clark and Niblett [4], CN 2 is a variant of AQ combining the best characteristics of AQ and ID3 [4,29], namely the intrinsic pruning techniques of ID3 and the predictive rule generation of form “if complexi then predict classi ”, where complexi has a definition identical to that given in paragraph 3.3.1. To generate decision rules from a given training subset designated by T S, CN 2 proceeds in iterative manner to populate a rule list denoted List of Rules and initialized to the empty list. At each iteration, this algorithm seeks for the best complex (already symbolized by complexi ) that covers the maximal number of examples of a unique class Ci and the minimal one of the other classes (i.e. T S − Ci ). Then, CN 2 deletes all examples covered by complexi from T S and
414
A. Hadrani et al.
adds the engendered rule (e.i. “if complexi then predict classi ”) to the end of List of Rules. This process is repeated until no suitable complex can be found or that T S is empty. At CN 2 output, we get an ordered list of predictive rules ending by a default one. This later simply foresees the most common class in T S for every new example not satisfying any other induced rule before that default one in List of Rules. Unlike AQ, CN 2 doesn’t depend on any particular example and can search for decision rules in a larger space by taking advantage of its intrinsic pruning mechanism. Further, this method can deal with noisy data and induces noise-resistant rules which are probabilistic (non deterministic) ones. Even so, the produced ordered rules sacrifice an understanding degree since a single rule interpretation depends on the other rules that precede it in the generated list. 3.3.3 Learning from Examples Module, Version 2 (LEM2) Suggested by Grzymala-Busse [11], LEM 2 is a rule induction algorithm based on a local covering strategy. It is a main component of the second version of the system called Learning from Examples using RS (LERS) appeared in 1997 [11]. To describe LEM 2 algorithm, we first have to define its specific concepts as follows. Let T S ⊆ U be a training set and Ci a concept of T S which can be a class represented by a decision-value pair (d, ci ), its lower or upper rough approximation with respect to a given subset N ⊆ A [11]. For every attributevalue pair p = (a, v) ∈ A × Va , we call a block of p, symbolized by [p], the subset expressed as: [p] = [(a, v)] = {x ∈ T S | a(x) = v}. Given a set P of attribute[p] ⊆ Ci . P is named value pairs, we say that Ci depends on P , iff: ∅ = [P ] = p∈P
a minimal complex of Ci iff: Ci depends on P and there is no P P such that Ci depends on P . Let P be a collection of nonempty sets of attribute-value pairs such that P = ∅. P is called local covering of Ci iff P satisfies: (i) every element of P is a minimal complex of Ci ; (ii) [P ] = Ci ; P ∈P
(iii) P is minimal (i.e. P has the minimal cardinality). Given a concept Ci of T S, LEM 2 induces a decision rule by determining a local covering Pi , suitable for Ci , using the next process. In step (1), this method computes the set P (Ci ) of all possible attribute-value pairs pertinent for the current set Ci as follows: P (Ci ) = {p | [p] ∩ Ci = ∅}. In step (2.1), LEM 2 selects, from P (Ci ), the best pair pbest maximizing |[p] ∩ Ci|, having the minimal cardinality if a tie occurs and met first if another tie occurs. Then, in step (2.2), pbest is added to the current candidate P of a minimal complex of Ci and the sets Ci and P (Ci ) are updated as follows: Ci = [pbest ] ∩ Ci and P (Ci ) = {p | [p] ∩ Ci = ∅} − P . The partial sequence of steps (2.1) and (2.2), which can be represented as step (2), is repeated until the condition [P ] ⊆ Ci is satisfied. Once this last condition is verified, P is pruned, in step (3), to obtain a minimal complex of Ci then added to the current candidate covering denoted P. Afterwards, all the covered examples by P are removed from Ci at step (4)
Rough Set Supervised Machine Learning Approaches: Survey and App
415
and the total step sequence, i.e. steps (1), (2), (3) and (4), is iterated till we get Ci = ∅. Finally, the obtained covering is pruned to get a local one, viz. the targeted set Pi . Decision rules induced from lower approximations of concepts are called certain, deterministic, exact or discriminant rules. Whereas, the rules generated from their upper approximations are named possible or approximate ones. 3.3.4 Indiscernibility Based Rule (IBR) Based on RST, IBR is an inductive learning algorithm allowing to generate decision rules from indiscernibility classes (i.e. elementary sets) defined by a given subset of conditional attributes [24,32]. Given a learning set T S (discretized of course) and a predetermined condition attribute reduction R, this algorithm starts to compute the indiscernibility relation IN D(R) and its equivalence classes [x]R as defined in Sect. 2. Each equivalence class [x]R can be represented by a set P of attribute-value [p] where pairs p = (a, v), where (a, v) ∈ R × Va , as follows: [x]R = [P ] = p∈P [p] = {y ∈ T S | a(y) = v}. Then, for every elementary set [x ]R = [(ai , vi )], 1≤i≤k
IBR induces the rule “if (a1 (x) = v1 ) ∧ ... ∧ (ak (x) = vk ) then predict dj ”, where {a1 , a2 , ..., ak } ⊆ R and dj is the most prevalent decision class on [x ]R .
4
Experiments: Description, Analysis and Discussion
In this section, we aim to evaluate experimentally the above studied approaches. For this purpose, we take the community-acquired meningitis classification as application field. Besides, we intend to support experts and medical practicians in their discrimination of meningitis categories by providing a smart application that facilitates this task based on a suitable, effective and comprehensible machine learning method. Indeed, meningitis is an inflammation of the protective membranes of the brain and spinal cord, which are called the meninges, caused by an infection of the fluid surrounding them named Cerebrospinal Fluid (CSF) [1]. This infection is usually of viral or bacterial origin. For viral or aseptic meningitis, patients can be cured within a few weeks without any treatment needed. However, bacterial meningitis is fatal since it can quickly lead to serious complications, such as deafness and mental deficiency, or even death if not screened and treated urgently [1,35]. For that reason, doctors are faced with the challenge of providing adequate and rapid treatment, especially for urgent cases of acute bacterial meningitis, within one hour of the patient’s hospitalization. In fact, in addition to the clinical presentation parameters or symptoms such as severe headache or cephalalgia, stiff neck, high fever, convulsions and photophobia, doctors usually and consistently rely on the results of CSF biological examination to diagnose this disease. These results are composed of the CSF macroscopic appearance (clear, turbid, hematic or purulent), its cytological characteristics indicating the predominant cells type in CSF (Lymphocytes or
416
A. Hadrani et al.
Polymorphonuclear Neutrophils), its biochemical parameters providing dosages or concentrations of some CSF components such as protein and glucose, the direct microbiological test (or Gram stain) result and the CSF microbial culture one. The dilemma with these above five different categories of CSF biological analysis results is to find a trade-off between their related discriminatory powers and the delay needed to obtain them. On the one hand, the first four ones can be attained in less than one hour, but their discriminatory powers can be reduced in some situations as in the case of a patient with bacterial meningitis who had an inappropriate and insufficient antibiotic therapy before hospitalization (case of a decapitated bacterial meningitis). On the other hand, the fifth result of this analysis, which is associated with CSF culture, makes it possible to determine the germ that caused bacterial meningitis and thus confirm the exact meningitis type, however the time to get it may exceed 48 h [1,35]. 4.1
Used Database Description
In this study, many experiments were performed on two medical databases related to the pathology of community-acquired meningitis [28]. Gathered from the individual files of the meningitis cases of the archive of the public health and epidemiological surveillance service of Chaouia Ouardigha regional health ex-direction and confronted with the result archive of the bacteriological cytological and chemical tests of the medical analysis laboratory of Hassan II hospital of Settat-Morocco, the original database covers patients, from Settat, hospitalized for meningitis between December 31st , 2005 and December 16th , 2016. This raw database contains 467 records each one is described by 28 conditional attributes in addition to the decision attribute ”TYPE MENINGITE”. Also, it’s characterized by: • a mixture of attribute types: 10 features are quantitative and 19 others are categorical. Here, It ought be noted that, apart from the decision attribute, the attributes that determine the exact germ causing meningitis in bacterial cases, such as the attribute related to CSF culture result, have been removed from this database for the reasons explained at the beginning of Sect. 4 as well as for the purpose of generating an efficient and reliable model for meningitis categories’ discrimination without the need for such late-obtained microbial analysis results; • a large number of missing values randomly distributed on certain attributes and objects: the total number of complete records is 310 which constitute the reduced database used in the following; • a number of noisy objects mainly due to missing values and errors of description and classification related to certain external and internal factors; • a variety of meningitis types given by decision classes of ”TYPE MENINGITE” (i.e. meningitis type) and illustrated with Fig. 1: the number of reduced database classes is 6. It should be noted that the meningitis classes used in this reduced database (see Fig. 1) are groupings of meningitis types standardized by the World
Rough Set Supervised Machine Learning Approaches: Survey and App
417
Health Organization. These groupings are determined by the Moroccan Health Ministry for professional reasons related to the specificities of this pathology at the national level and how to react according to the cases [1]. These classes can be specified as follows: • Haemophilus Meningitis (HM): class of meningitis caused by Haemophilus influenzae type b; • Pneumococcal Meningitis (PM): class of meningitis caused by Streptococcus Pneumoniae bacteria; • Confirmed Meningococcal Meningitis (CMM): meningitis class certainly caused by the bacterium Neisseria Meningitidis (i.e. Meningococcus); • Probable Meningococcal Meningitis (PMM): class of meningitis that are possibly due to Neisseria Meningitidis; • Probale Bacterial Meningitis (PBM): class of bacterial meningitis whose germs are not determined due to the bacteria decapitation effect; • Lymphocytic Meningitis (LM): class of meningitis with lymphocyte predominance. It should be emphasized that, in this reduced database, more than 99% of LM cases are viral.
Fig. 1. The distribution of the reduced database objects according to meningitis classes.
Moreover, and among the 467 cases of meningitis reported in this period, Settat province has known 10 cases of meningitis rarely recorded in this province, namely 1 case of Pyocyanic Meningitis , 2 cases of Proteus Mirabilis Meningitis, 1 case of Staphylococcus Aureus Meningitis, 1 case of Meningoencephalitis, 1 case of Post-Traumatic Meningitis and 4 cases of Tuberculous Meningitis. Nine of them are eliminated due to incompleteness of records, as previously mentioned, and the tenth case is also discarded for its rarity to avoid unbalancing the database classes and thus skewing the results. Finally, the experiments that detailed below are realized using this reduced database designated by RDB and
418
A. Hadrani et al.
another database derived from RDB and named DDB. The latter is a copy of the first one in which the 6 classes are abbreviated into 2 parent classes V and B: V is the viral meningitis class representing LM and B is the bacterial meningitis class gathering the subclasses HM, PM, CMM, PMM and PBM. 4.2
Test Environment
The experiments described below are performed using a laptop with a 2.8 GHz Intel® Core TM i7−2640M processor, 8 GB of RAM and a Windows 7 operating system. In addition and concerning the studied methods (see Sect. 3), we use their R implementations given by the “RoughSets” version 1.3-0 library [32]. Furthermore, the PREDICT function of “RoughSets” is utilized to foresee the decision class for each new object based on a given rules set. Besides, all the executed simulations have been performed with R software version 3.5.3. 4.3
Realized Experiments
As earlier mentioned, we study three decision rule induction approaches involving RST. Each of these approaches is modular and in which every module (i.e. stage or phase) can be one of the possible methods detailed in Sect. 3. Thus, for every approach, we will consider many possible combinations of the available methods as depicted in Table 1. To evaluate the classification accuracies of these combinations, we use the stratified tenfold cross-validation repeated 10 times. The validation results are given by Figs. 2 and 3. It should be underlined that, in a preliminary phase, each experiment or test was repeated several times to optimize different parameters’ values of used methods. This optimization is done for maximizing the prediction quality of the generated rule set. In our case, the optimal values of these parameters are those mentioned in Table 1. 4.4
Results and Discussion
From Figs. 2 and 3, we can notice that: • For the 1st approach: the average error rate varies between 1.94% (result of the experiments 0 01, 0 02, 0 03, 0 07, 0 08, 0 09 and 0 24) and 24% (result of the experiment 0 11) with a standard deviation (SD) of 6.23% in the case of DDB while this rate increases from 9.68% (result of the tests 1 01 and 1 02) to 64.32% (result of the test 1 21) with a SD of 22.31% in the case of RDB. • For the 2nd approach: the average prediction error of generated rule sets varies between 3.13% (result of the experiment 0 32) and 7.06% (result of the experiment 0 33) with a SD of 2.09% in the case of DDB whereas this error increases from 19.1% (result of the test 1 32) to 29.32% (result of the test 1 31) with a SD of 5.24% in the case of RDB.
Rough Set Supervised Machine Learning Approaches: Survey and App
419
• For the 3rd approach: the average classification error rate passes from 3.03% (result of the experiment 0 35) to 7.97% (result of the experiment 0 39) with a SD of 1.78% in the case of DDB while this rate grows from 10.1% (result of the test 1 35) to 22.29% (result of the test 1 37) with a SD of 4.41% in the case of RDB. From the previous analysis of induced rules’ performance in term of classification accuracy, we can deduce that, in the cases of DDB and RDB, both method combinations X 01 and X 02 (see Table 1), which correspond to the 1st approach, generate decision rule sets with the best predictive power. Moreover, to analyze the performances of these approaches in term of quality and compactness of the generated decision rule sets, we present, as example, in Table 2 those produced by the experiments 0 01, 0 02, 1 01 and 1 02 in the 1st iteration of their 1st stratified cross-validations. In this Table, R DGS, P PNN, NO ORGANISM, supS and L stand for the Result of the Direct Gram Stain, the Proportion of the PolyNuclear Neutrophils, No Organism detected by gram stain, the number of objects supporting the matched rule and Laplace’s estimate of the corresponding rule confidence [32] respectively. From Table 2 and the experts’ view point, we can also deduced that both combinations X 01 and X 02 allow inducing global and deterministic decision rule sets whose cardinality is minimal. Each of these generated rules is based on a conditional attribute subset (i.e. a reduction) that is relevant to discern different decision classes of Meningitis. These interesting results concerning X 01 and X 02 are certainly due to the following three factors: 1. Unlike the 2nd and 3rd approaches, the 1st one integrates a very important stage corresponding to feature selection and allowing to remove all the superfluous attributes that can produce random dependencies with decision attribute and consequently reduce discriminative and predictive power of the generated rule set. Among the studied methods for this stage, we can deduce that GHCADR is the best one provided that: • Its approximation threshold is adjusted to its optimal value depending on the application field; • The entropy or the Gini-index is used as a quality measure in it. 2. The ability of the IBR algorithm to generate global and discriminant decision rules from indiscernibility classes, once a good reduction is determined for the used data set. 3. The adaptation of the EWID method and the optimal value of its parameter k (i.e. k = 2) to the most indispensable continuous features of the two databases RDB and DDB, particularly the proportions of the polynuclear neutrophils and lymphocytes in Cerebrospinal Fluid (CFC) [1,5].
420
A. Hadrani et al.
Table 1. The different possible method combinations for the three studied modular approachesa . Approach
1st one
2nd one
3rd one
a
Test no Possible combinations
FMS parameter values
Discretization Feature method selection method (FSM)
RI method
X 01
EWID(k = 2)
GHCADR(Qd ,)
IBR
(Gini,0.25)
X 02
EWID(k = 2)
GHCADR(Qd ,)
IBR
(HDT ,0.25)
X 03
EWID(k = 2)
GHCADR(Qd ,)
IBR
(Disc,0.25)
X 04
EFID(k = 3)
GHCADR(Qd ,)
IBR
(Gini,0.25 (0.3) for X = 0 (1))
X 05
EFID(k = 3)
GHCADR(Qd ,)
IBR
(HDT ,0.3)
X 06
EFID(k = 3)
GHCADR(Qd ,)
IBR
(Disc,0.25 (0.06) for X = 0 (1))
X 07
SDBLD
GHCADR(Qd ,)
IBR
(Gini,0.25 (0.35) for X = 0 (1))
X 08
SDBLD
GHCADR(Qd ,)
IBR
(HDT ,0.3)
X 09
SDBLD
GHCADR(Qd ,)
IBR
(Disc,0.25 (0.1) for X = 0 (1))
X 10
EWID(k=2)
RSAR
IBR
−
X 11
EFID(k=3)
RSAR
IBR
−
X 12
SDBLD
RSAR
IBR
−
X 13
EWID(k = 2)
GHDSR(Qd )
IBR
HDT
X 14
EWID(k = 2)
GHDSR(Qd )
IBR
Gini
X 15
EWID(k = 2)
GHDSR(Qd )
IBR
Disc
X 16
EFID(k = 3)
GHDSR(Qd )
IBR
HDT
X 17
EFID(k = 3)
GHDSR(Qd )
IBR
Gini
X 18
EFID(k = 3)
GHDSR(Qd )
IBR
Disc
X 19
SDBLD
GHDSR(Qd )
IBR
HDT
X 20
SDBLD
GHDSR(Qd )
IBR
Gini
X 21
SDBLD
GHDSR(Qd )
IBR
Disc
X 22
EWID(k = 2)
DAAR(Qd )
IBR
Gini
X 23
EWID(k = 2)
DAAR(Qd )
IBR
HDT
X 24
EWID(k = 2)
DAAR(Qd )
IBR
Disc
X 25
EFID(k = 3)
DAAR(Qd )
IBR
Gini
X 26
EFID(k = 3)
DAAR(Qd )
IBR
HDT
X 27
EFID(k = 3)
DAAR(Qd )
IBR
Disc
X 28
SDBLD
DAAR(Qd )
IBR
Gini
X 29
SDBLD
DAAR(Qd )
IBR
HDT
X 30
SDBLD
DAAR(Qd )
IBR
Disc
X 31
SDBLD
−
LEM2
−
X 32
EWID(k = 2)
−
LEM2
−
X 33
EFID(k = 3)
−
LEM2
−
X 34
SDBLD
−
AQ
−
X 35
EWID(k = 2)
−
AQ
−
X 36
EFID(k = 3)
−
AQ
−
X 37
SDBLD
−
CN2
−
X 38
EWID(k = 2)
−
CN2
−
X 39
EFID(k = 3)
−
CN2
−
X is set to 0 for the experiments applied on DDB and 1 for those performed with RDB.
Rough Set Supervised Machine Learning Approaches: Survey and App
421
Fig. 2. Average classification error rates resulting from validation of the decision rule sets induced by 39 experiments (i.e. the 0 j where j ∈ {01, 02, ..., 39}) in case of DDB.
Fig. 3. Average classification error rates resulting from validation of the decision rule sets induced by 39 experiments (i.e. the 1 j where j ∈ {01, 02, ..., 39}) in case of RDB. Table 2. Decision rule sets examples induced by experiments 0 01, 0 02, 1 01 and 1 02. Test no Decision Rule Set 0 01 or
”IF P PNN is (50.5, Inf] THEN is B; (supS = 179; L = 0.99)”
0 02
”IF P PNN is [-Inf,50.5] THEN is V; (supS = 100; L = 0.94)” ”IF P PNN is (50.5, Inf] and R DGS is GRAM-NEGATIVE COCCOBACILLI THEN is HM; (supS = 8; L = 0.64)” ”IF P PNN is (50.5, Inf] and R DGS is GRAM-NEGATIVE BACILLI THEN is PBM; (supS = 1; L = 0.29)”
1 01
”IF P PNN is (50.5, Inf] and R DGS is GRAM-NEGATIVE DIPLOCOCCI THEN is CMM;(supS = 42; L = 0.71)”
or
”IF P PNN is (50.5, Inf] and R DGS is GRAM-POSITIVE DIPLOCOCCI THEN is PM; (supS = 37;L = 0.86)”
1 02
”IF P PNN is (50.5, Inf] and R DGS is NO ORGANISM THEN is PBM;(supS = 91; L = 0.82)” ”IF P PNN is [-Inf,50.5] and R DGS is GRAM-NEGATIVE DIPLOCOCCI THEN is PMM;(supS = 1; L = 0.29)” ”IF P PNN is [-Inf,50.5] and R DGS is GRAM-POSITIVE DIPLOCOCCI THEN is PM; (supS = 1; L = 0.29)” ”IF P PNN is [-Inf,50.5] and R DGS is NO ORGANISM THEN is LM; (supS = 98; L = 0.93)”
422
5
A. Hadrani et al.
Conclusion
In this work, we studied three decision rules induction approaches involving Rough Sets. These three approaches are modular including the real-valued attribute discretization step, the feature selection one required only by the 1st approach and the decision rule induction stage. The availability of several interesting methods for each stage allowed us to evaluate and compare various possible models using two real meningitis databases one of them is derived from the other. The validation results of these different models showed that IBR algorithm outperforms LEM2, AQ and CN2 methods in term of classification accuracy, quality and compactness of induced decision rules provided that it is combined with GHCADR and EWID methods and the additional parameters of these last two techniques are well adjusted depending on the application field.
References 1. Minist`ere de la Sant´e, Maroc. Guide de la lutte contre les m´eningites bact´eriennes communautaires (2010) 2. Bazan, J.G., Nguyen, H.S., Nguyen, S.H., Synak, P., Wr´ oblewski, J.: Rough set algorithms in classification problem. In: Kacprzyk, J., Polkowski, L., Tsumoto, S., Lin, T.Y. (eds.) Rough Set Methods and Applications, vol. 56, pp. 49–88. PhysicaVerlag HD, Heidelberg (2000) 3. Cervone, G., Panait, L., Michalski, R.: The development of the AQ20 learning system and initial experiments. In: Klopotek, M.A., Michalewicz, M., Wierzcho´ n, S.T. (eds.) Intelligent Information Systems 2001, Advances in Intelligent and Soft Computing, vol. 10, pp. 13–29. Physica-Verlag, Heidelberg (2001) 4. Clark, P., Niblett, T.: The CN2 induction algorithm. Mach. Learn. 3(4), 261–283 (1989) 5. Coll`ege des universitaires de Maladies Infectieuses et Tropicales: Maladies infectieuses et tropicales, ecn.pilly 2018 - 5`eme ´edition edn. ALINEA Plus, 8, rue Froidevaux - Paris (2017) 6. Dash, R., Paramguru, R.L., Dash, R.: Comparative analysis of supervised and unsupervised discretization techniques. Int. J. Adv. Sci. Technol. 2(3), 29–37 (2011) 7. Slezak, D.: Approximate entropy reducts. Fundamenta Informaticae 53(3–4), 365– 390 (2002) 8. Dougherty, J., Kohavi, R., Sahami, M.: Supervised and unsupervised discretization of continuous features. In: Proceedings of the Twelfth International Conference on International Conference on Machine Learning, ICML 1995, pp. 194–202. Morgan Kaufmann Publishers Inc., San Francisco (1995) 9. Dubois, D., Prade, H.: Rough fuzzy sets and fuzzy rough sets. Int. J. Gen. Syst. 17(2–3), 191–209 (1990) 10. Fayyad, U., Irani, K.: Multi-interval discretization of continuous-valued attributes for classification learning. In: Proceedings of the 13th International Joint Conference on Artificial Intelligence, vol. 2, pp. 1022–1027. Morgan Kaufmann, Chambery (1993) 11. Grzymala-Busse, J.W.: A new version of the rule induction system LERS. Fundamenta Informaticae 31(1), 27–39 (1997)
Rough Set Supervised Machine Learning Approaches: Survey and App
423
12. Øhrn, A.: ROSETTA Technical Reference Manual (2001). http://bioinf.icm.uu.se/ rosetta/materials/manual.pdf 13. Janusz, A., Stawicki, S.: Applications of approximate reducts to the feature selection problem. In: Proceedings of the 6th International Conference on Rough Sets and Knowledge Technology, RSKT 2011, pp. 45–50. Springer-Verlag, Heidelberg (2011) ´ ezak, D.: Utilization of attribute clustering methods for scalable 14. Janusz, A., Sl¸ computation of reducts from high-dimensional data. In: Proceedings of the 2012 Federated Conference on Computer Science and Information Systems (FedCSIS), pp. 295–302. IEEE, Wroclaw (2012) ´ ezak, D.: Random probes in computation and assessment of approx15. Janusz, A., Sl¸ imate reducts. In: Kryszkiewicz, M., Cornelis, C., Ciucci, D., Medina-Moreno, J., Motoda, H., Ra´s, Z.W. (eds.) Rough Sets and Intelligent Systems Paradigms, vol. 8537, pp. 53–64. Springer International Publishing, Cham (2014) 16. Jiang, F., Sui, Y.: A novel approach for discretization of continuous attributes in rough set theory. Knowl.-Based Syst. 73, 324–334 (2015) 17. Jiang, F., Zhao, Z., Ge, Y.: A supervised and multivariate discretization algorithm for rough sets. In: Yu, J., Greco, S., Lingras, P., Wang, G., Skowron, A. (eds.) Rough Set and Knowledge Technology, RSKT 2010. Lecture Notes in Computer Science, vol. 6401, pp. 596–603. Springer, Heidelberg (2010) 18. Liu, H., Hussain, F., Tan, C.L., Dash, M.: Discretization: an enabling technique. Data Min. Knowl. Disc. 6, 393–423 (2002) 19. Liu, H., Setiono, R.: Chi2: feature selection and discretization of numeric attributes. In: Proceedings of the Seventh International Conference on Tools with Artificial Intelligence, pp. 388–391. IEEE, Herndon (1995) 20. Michalski, R.S., Kaufman, K., Wnek, J.: The AQ family of learning programs: a review of recent developments and an exemplary application. Reports of Machine Learning and Inference Laboratory MLI 91-11, School of Information Technology and Engineering, George Mason University, Fairfax, VA (1991) 21. Nguyen, H.S.: Discretization problem for rough sets methods. In: Polkowski, L., Skowron, A. (eds.) Rough Sets and Current Trends in Computing. Lecture Notes in Computer Science, vol. 1424, pp. 545–552. Springer, Berlin Heidelberg (1998) 22. Nguyen, H.S.: On efficient handling of continuous attributes in large data bases. Fundamenta Informaticae 48(1), 61–81 (2001) 23. Øhrn, A., Ohno-Machado, L., Rowland, T.: Building manageable rough set classifiers. In: Proceedings of AMIA Annual Fall Symposium, Orlando, USA, pp. 543–547 (1998) 24. Pawlak, Z.: Rough sets. Int. J. Comput. Inform. Sci. 11(5), 341–356 (1982) 25. Pawlak, Z.: Rough Sets: Theoretical Aspects of Reasoning about Data. Springer, Netherlands, Dordrecht (1991) 26. Pawlak, Z., Skowron, A.: Rough membership functions: a tool for reasoning with uncertainty. Banach Cent. Publ. 28(1), 135–150 (1993) 27. Polkowski, L., Skowron, A.: Rough mereology: a new paradigm for approximate reasoning. Int. J. Approximate Reasoning 15(4), 333–365 (1996) 28. Pottecher, T., Balabaud-Pichon, V.: M´eningites nosocomiales de l’adulte. Annales Fran¸caises d’Anesth´esie et de R´eanimation 18(5), 558–566 (1999) 29. Quinlan, J.R.: Induction of decision trees. Mach. Learn. 1(1), 81–106 (1986) 30. Quinlan, J.R.: C4.5: Programs for Machine Learning. Morgan Kaufmann Publishers Inc., San Francisco (1993) 31. Radzikowska, A.M., Kerre, E.E.: A comparative study of fuzzy rough sets. Fuzzy Sets Syst. 126(2), 137–155 (2002)
424
A. Hadrani et al.
´ ezak, D., Ben´ıtez, 32. Riza, L.S., Janusz, A., Bergmeir, C., Cornelis, C., Herrera, F., Sl¸ J.M.: Implementing algorithms of rough set theory and fuzzy rough set theory in the R package “RoughSets”. Inf. Sci. 287, 68–89 (2014) 33. Shen, Q., Chouchoulas, A.: A modular approach to generating fuzzy rules with reduced attributes for the monitoring of complex systems. Eng. Appl. Artif. Intell. 13(3), 263–278 (2000) 34. Stefanowski, J., Wilk, S.: Rough sets for handling imbalanced data: combining filtering and rule-based classifiers. Fundamenta Informaticae 72(1–3), 379–391 (2006) 35. Viallon, A., Bothelo-Nevers, E., Zeni, F.: Clinical decision rules for acute bacterial meningitis: current insights. Open Access Emerg. Med. 8, 7–16 (2016) 36. Wojtusiak, J., Michalski, R.S., Kaufman, K.A., Pietrzykowski, J.: The AQ21 natural induction program for pattern discovery: initial version and its novel features. In: Proceedings of the 18th IEEE International Conference on Tools with Artificial Intelligence (ICTAI 2006), Arlington, VA, USA, pp. 523–526 (2006) 37. Wr´ oblewski, J.: Ensembles of classifiers based on approximate reducts. Fundamenta Informaticae 47(3–4), 351–360 (2001) 38. Yao, Y.: Decision-theoretic rough set models. In: Yao, J., Lingras, P., Wu, W.Z., ´ ezak, D. (eds.) Rough Sets and Knowledge TechnolSzczuka, M., Cercone, N.J., Sl¸ ogy. Lecture Notes in Computer Science, pp. 1–12. Springer, Heidelberg (2007) 39. Yao, Y., Wong, S.: A decision theoretic framework for approximating concepts. Int. J. Man Mach. Stud. 37(6), 793–809 (1992) 40. Zadeh, L.A.: Fuzzy sets. Inf. Control 8(3), 338–353 (1965) 41. Zeeman, E.C.: The topology of the brain and visual perception. In: Fort Jr., M.K. (ed.) Topology of 3-Manifolds and related topics, pp. 240–256. Prentice Hall, Englewood Cliffs (1962) 42. Zhang, Q., Xie, Q., Wang, G.: A survey on rough set theory and its applications. CAAI Trans. Intell. Technol. 1(4), 323–333 (2016) 43. Ziarko, W.: Variable precision rough set model. J. Comput. Syst. Sci. 46(1), 39–59 (1993)
Knee Functional Telerehabilitation System for Inclusive Smart Cities Based on Assistive IoT Technologies Mohamed El Fezazi(&) , Mounaim Aqil, Atman Jbari, and Abdelilah Jilbab Electronic Systems Sensors and Nanobiotechnologies, Higher School of Technical Education, Mohammed V University in Rabat, Rabat, Morocco [email protected], [email protected], [email protected], [email protected]
Abstract. In the current smart cities, assistive IoT technologies are employed in several services to improve the physical and social functioning of persons with disabilities (PwD) and elderly people. This book chapter highlights the exploitation of these technologies to build a home-based smart rehabilitation system for a knee injury. This system aims to reduce patient mobility to attend the rehabilitation session, in order to achieve good accessibility and proper management of transport, improving thus smart cities’ characteristics in terms of smart living, smart mobility, and smart healthcare. The architectural model of the proposed system is designed in a manner to ensure the telemonitoring function of the rehabilitation process, and the remote access to feedback information for both the practitioners and patients. It consists of three main components: measurement instrumentation, processing and transmission unit, and cloud data storage and visualization. Furthermore, the results obtained in the experimental test showed that measurements error percentages were lower than 2%, and the data displayed in the cloud are significant, indicating that the accuracy and reliability of the developed system are satisfactory, as well as telerehabilitation engineering has prospects in Inclusive smart city applications. Keywords: IoT Data acquisition Knee telerehabilitation
Smart cities Assistive technologies
1 Introduction Smart cities have become among the essential keys to promoting sustainable development, putting information and communication technologies (ICTs) at the service of inhabitants, in order to improve the quality of life and achieve higher development levels [1]. The Internet of Things (IoT) plays a critical role in building smart cities and supporting comprehensive urban information systems. It offers smart solutions for cities in terms of governance, economic growth, environmental sustainability, quality of life, transportation, power, and water usage [2]. The basis of the IoT is the connectivity of physical objects using the framework and protocols of networks. These © Springer Nature Switzerland AG 2020 M. Ben Ahmed et al. (Eds.): SCA 2019, LNITI, pp. 425–439, 2020. https://doi.org/10.1007/978-3-030-37629-1_31
426
M. El Fezazi et al.
objects apply specific technologies working more efficiently and effectively in an environment of many low power devices, operating in locations not previously connected, and often generating significant amounts of data over time [3]. For this purpose, the IoT requires customized systems within a framework that will ensure both the expectations of reliable connectivity at the lower levels, while processing and applying the generated data into meaningful information and actions. Therefore, several factors and challenges are to be taken into consideration for adopting and implementing IoT in smart cities. Following are some major of these challenges [2, 4]: – Planning: before any change in city characteristics, it is essential to understand inhabitants’ needs, determine the areas that require improvements. and make a master plan based on a city’s needs in order to improve better facilities for residents and avoid wasting funds because of repeated and redundant construction. – Costs and Quality: among the most important reasons for the success of smart city projects, is the Choice between low costs versus high quality in term of hardware and software infrastructure. – Security and Privacy: IoT based smart cities solutions raise concern for security and privacy of citizen’s information. This sensed data from physical spaces contain granular details about the people living in those environments. Hence, the general security and privacy requirements include ensuring confidentiality and integrity, implementing information security levels and risk assessment systems, improving the network monitoring capabilities, and strengthening network management. – IoT challenges: IoT technology presents some issues. Citing for example heterogeneity in terms of communication protocols, data formats, and technologies. Reliability from the different views of service, system, and data. And the large scale of information and big data, that requires suitable storage and computational capability collected at a high rate. Indeed, IoT provides several applications for smart cities such as environment monitoring, transportation and traffic management, smart healthcare, smart grid, smart homes. Each of these applications carries objectives that contribute to offering a highquality living environment [2]. In the case of the healthcare domain, IoT technologies have incorporated many advantages in smart cities focused on three important aspects: – The first is tracking of people and objects including patients, staff and ambulance, aims to monitor the status of patients in a clinic or hospital and control the availability of the ambulance, blood products, and different organs for transplantation in order to provide better and faster work-flow in the hospital. – The second is patient identification, seeks to decrease the risk of mistake for the prevention of getting wrong drugs, doses, and procedures. – The third aspect is automatic data gathering and sensing helps to save time for data processing and preventing human errors. Through sensor devices, diagnosing patient conditions, providing real-time information on patient health indicators such as prescription compliance by the patient is implemented. By using bio-signal monitoring, the patient condition is investigated through heterogeneous wireless access-based methods to enable for getting the patient data anywhere [3].
Knee Functional Telerehabilitation System for Inclusive Smart Cities
427
In this field, assistive technologies are a vital tool for any physically challenged person. They provide assistive, adaptive, and rehabilitation services, used to increase, maintain, or improve functional capabilities of persons with disabilities (PwD) to face the urban space in an autonomous and independent way, and contribute to the achievement of easy access to services and knowledge, user-oriented, and personalized. This is known as inclusive smart cities [5]. The aim of this work is to design a smart rehabilitation system for knee injury based on IoT in order to improve smart city characteristics, in terms of smart living, devoting special attention to health conditions for residents which are key to a city’s development. And in terms of smart mobility, planning the rehabilitation sessions in an efficient manner towards good accessibility and proper management of transport reducing somehow carbon dioxide emissions and pollution. The rest of this chapter is structured as follows. Section 2 presents an overview of IoT functions, architectures, and frameworks. Section 3 explains the methodological approach followed in system realization. Then, Sect. 4 describes the evaluation test and the experimental results. Finally, we conclude the paper in Sect. 5.
2 Related Work 2.1
IoT Functions and Architectures
The functional elements of the IoT system may differ from one point of view to another. In general, these elements can be classified into two essential areas: (a) core subsystems and (b) optional subsystems. a. Core subsystems of an IoT Architecture: At the core an IoT application consists of the following components: – IoT devices: including sensors, actuators, and embedded communication hardware [6]. These devices help in interacting with the physical environment, providing connectivity options for sending and receiving data [7]. – Communication protocols: protocols and standards are applied to ensure the interoperability of networking. In the IoT environment, devices are required to be very small in size for implementation, therefore they must operate on battery, and run on low memory and CPU power with a limited network communication capability. Because of these, IoT protocols are required to provide communications between resource-constrained devices within the IoT Low-Power and Lossy Networks [7]. – Middleware: summarizes storage and computing tools for data analytics [6]. It provides an Application Programming Interface (API) for communication, data management, computation, security, and privacy [7]. – User Interface: aims to visualize telemetry data and facilitate device management. It can be delivered on a wide array of device types, in native applications, and browsers [6].
428
M. El Fezazi et al.
b. Optional subsystems of an IoT Architecture: In addition to the core subsystems, many IoT applications will include further subsystems such as: – Data transformation: allows restructuring, combination, or transformation of telemetry data sent from devices. This manipulation can include protocol transformation, combining data points, and more [8]. – Machine Learning (ML) Subsystem: Enables systems to learn from data and experiences and to act without being explicitly programmed. and therefore, allowing predictive algorithms to be executed over historical telemetry data [9]. IoT is not a single technology; rather it is an agglomeration of various technologies that work together in tandem [7]. Therefore, in a smart IoT system, it is important to determine architecture properly, identify element technologies to configure the architecture, and stipulate the specifications required for the element technologies at an early stage of the system. Consequently, tasks can be easily divided, and redundant functions can be eliminated. Hence, many organizations are trying to define an architecture for systems with an IoT vision. In [7] and [10] an overview of different architectures has been proposed: a. Three-Layer Architecture:
Fig. 1. IoT architecture (A: three layers) (B: five layers) [7].
The three layers are (as shown in Fig. 1): – The first is the perception layer, comprises a wide range of endpoint devices that measure physical variables using a specific method. – Communications and connectivity are concentrated in the second layer namely the network layer. The most important function of this level is reliable, timely information transmission to specific servers in the cloud or public services using non-cellular and cellular communication technologies. The implementation of this layer should not require a different network—it relies on existing networks. – The third is the application layer, where information interpretation, reporting, and control occurs.
Knee Functional Telerehabilitation System for Inclusive Smart Cities
429
b. Five-Layer Architecture: This architecture is organized into five layers as shown in Fig. 1 namely, perception, transport, processing, application, and business layers. The role of the perception and application layers is the same as the three-layer architecture. Following the function of the remaining layers: – The transport layer defines services to segment, transfer, and reassemble the data for individual communications between end devices. – The processing layer provides four kinds of services: information repositories, databases, servers, computing, and storage. It employs many technologies such as cloud computing and big data processing modules. – The business layer It provides checking and managing the whole IoT system. c. Service-Oriented Architecture (SOA): In SOA, there are four layers, namely, perception layer, network layer, service layer, and application layer. The function of layers is similar as the previous architecture models except for the service layer. This service layer is further divided into two layers called as service management sublayer and service composition sublayer. It provides services to the application layer by organizing communication and managing data. d. Fog-to-Cloud Architecture (F2C): In Cloud-based architecture, data processing is done in a large centralized fashion by cloud computers. Lately, there is a move towards another system architecture, namely, fog computing, where a part of the data processing and analytics performed at the sensors and network gateways. 2.2
IoT Platform and Framework
IoT platform and Framework are defined as a hardware and software toolkits to develop IoT applications according to a certain style or method [11]. Hardware items include embedded systems and boards. These latter provide device control, which includes monitoring of devices, security, and firmware updates. Data acquisition, that encompasses management and transformation at different layers of the IoT. And application development, that comprises analytics, event-driven logic, visualization, and application programming [12]. Selecting the optimum development board is a very important factor in the success of IoT applications. Typically, three things must be considered in that respect, namely, computing specifications, development environment, and communication standards. Fundamentally, account must be taken for processor/microcontroller, clock speed, GPIO, ADC/DAC, connectivity (Wi-Fi, Bluetooth, or Ethernet), communication (I2C, UART, and SPI), and the Costs-Quality of the development board to be selected [12]. In a software aspect, the IoT platform is defined as the middleware layer responsible for consuming data from sensors and devices and applying this data into meaningful information and actions. It is often a cloud infrastructure that offers databases, data and device management, security and storage, protocol translation, and programming framework through which the hardware platform can connect and consume these cloud-based services [12, 13].
430
M. El Fezazi et al.
According to the above, the cloud IoT platform is the basic infrastructure for the implementation of IoT projects. The most important factors to consider for choosing the optimum platform are the instruments provided, specifically, the IDE for programming, the API available for access of data and notifications, and communication protocols. Furthermore, the efficiency of data management tools, such as desktop and mobile applications, dashboards. Then, pricing, community support, and available documentation [12].
3 Material and Method The smart rehabilitation system proposed in this paper is a home-based telemonitoring system of knee physical rehabilitation exercises, premised upon an IoT architecture. In fact, the architectural model of the system environment in this study is suggested in order to ensure real-time feedback and adherence, and therapist supervision as shown in Fig. 2. It consists of three main components: data acquisition wherein knee angle measurement is carried out, to monitor the movement while patients do rehabilitation exercises, data aggregation and transmission using IoT development board, and data storage and visualization using cloud platform. The details of each component are described in the following sections.
Fig. 2. Architectural model of the smart rehabilitation system.
3.1
Knee Angle Measurement
The knee angle is mainly measured in physical rehabilitation to assess knee joint function. In fact, the consistency of knee angle measurement over a period provides insights into changes and improvements of the joint flexion. Numerous methods exist to measure it, including but not limited to goniometry, inertial sensors, and 3D motion capture, etc. [14]. For this work, the knee angle is measured using electro-goniometer device uses a potentiometer as the sensing element, which changes its output voltage
Knee Functional Telerehabilitation System for Inclusive Smart Cities
431
with the angular motion of the joint that causes its rotation. For calibration, the potentiometer is attached directly to a manual goniometer as shown in Fig. 3. Moving the manual goniometer from one known position to another while recording data from the potentiometer positions will provide a voltage measure equivalent to an actual angular displacement. From these measurements, a calibration coefficient is computed [15].
Fig. 3. A manual goniometer.
3.2
Processing and Transmission Unit
The main processing and transmission unit is an IoT development board with a Wi-Fi module. It allows the processing of data and at the same time transmission over the WiFi network which is an obvious choice for IoT connectivity because in-building Wi-Fi coverage is now almost ubiquitous. Wi-Fi operates over an internationally approved frequency band of 2.4 GHz. It offers a max data rate of 100 Gbps and a max range of 1000 m [16]. The knee angle is captured using a trim potentiometer interfaced to the Analog-toDigital Converter pin of the IoT development board. The raw analog signal is read from the pin connected to the sensor. The reading is then processed and returned as a value representing the rotation angle in degrees. Indeed, the reading raw signal is then converted into a voltage reading by multiplying it by the input voltage range and dividing it by the upper boundary of the reading’s range. The final value, in degrees, is calculated by multiplying the reading voltage by the angular range of the sensor (MaxAngle) and then dividing it by the input voltage range according to the following equations: Sensor Voltage ¼ Angle Value ¼
Sensor Value Input Voltage Range 2n
ð1Þ
Sensor Voltage Max Angle Input Voltage Range
ð2Þ
Where n is the resolution of the Analog-to-Digital Converter.
432
M. El Fezazi et al.
In terms of data transmission, two execution modes are suggested [17]: – In the first mode as shown in Fig. 4, the data is acquired and forwarded continuously and displayed to the practitioner in real-time state. This mode executes in the following sequence of steps: reading data from the sensor, connecting to the Wi-Fi network, establishing a connection with the cloud IoT platform, and posting data to the cloud server.
Fig. 4. Flowchart of the continuously transmission mode algorithm.
– In the second mode as shown in Fig. 5, the monitored data is mainly stored at the patient side and is transmitted on practitioner demand. The difference in this mode is that the collected data are first aggregated and then transmitted whenever required by the practitioner. Indeed, the system collects data continuously and stores it in the IoT development board buffer memory. When the transmission demand happens, connecting to the internet via the Wi-Fi network is established, and a sequence of aggregated data is sent to the cloud IoT platform.
Knee Functional Telerehabilitation System for Inclusive Smart Cities
433
Fig. 5. Flowchart of the on-demand transmission mode algorithm.
3.3
Cloud Data Logging and Visualization
The telemonitoring function is established using the Microsoft Azure IoT Suite cloud platform, which is a collection of tools and services for IoT. It offers a solution to three key pillars: connecting and scaling, telemetry patterns, and big data [18]. Indeed, The IoT development board is utilized to post data on Azure IoT Hub and storing them in a table using Azure stream analytics. Azure IoT Hub is a solution for connect and control. It offers massive scalability, secure two-way communication and command control to a device level. It can be viewed as a high-scale gateway that acts as an enabler and manager of all bidirectional communication to and from devices, as shown in Fig. 6 [19]. In order to send data to Azure IoT Hub, the IoT development board is directly connected using IoT Hub-supported protocols like HTTP. Thus, at this stage, they are two essential steps. The first is creating an IoT Hub by configuring some settings including messaging settings, namely, configuration settings pertaining to cloud-todevice and device-to-cloud message communication. Operations monitoring settings
434
M. El Fezazi et al.
Fig. 6. The role of IoT Hub [19].
that contains configuration settings pertaining to the type of information and events monitored by IoT Hub. And other options such as: diagnostics settings, shared access policies, pricing and scale [19]. The second step is device registering, that concerned about adding and connecting our IoT development board to created IoT hub in order to ensure device-to-cloud communication [20]. Azure stream analytics is the event processing solution. It offers the processing, analytics, and handling of massive amounts of real-time data. At this point in the process, the data from the devices are sitting in Azure IoT Hub. Then, we create Stream analytics job to save the data in a storage account using the table storage service [21]. For real-time data monitoring a cloud dashboard allowing visualization of the transmitted information is implemented based on Power BI Desktop, which is a cloudbased dashboard and analytics tool. It provides secure access of data, as well as secure and live communication with data sources. By combining the services offered by Microsoft, namely, azure IoT hub, azure stream analytics, and Power BI, we create an intermediate link between the measurement module and the service visualization [22].
4 Experimental Test and Results 4.1
Prototype Implementation
For the experimental test, a prototype of the proposed system is implemented in order to acquire data concerning the flexion-extension involving knee joints movement as shown in Fig. 7. The prototype consists of a trimmer potentiometer with two armatures one is attached to one segment of the joint and the other to the adjacent segment. Any angular motion of the joint causes the potentiometer to rotate and therefore change its output voltage [15]. The trimmer potentiometer is connected to the Analog-to-Digital Converter pin of the IoT development board that connects to the available Wi-Fi network, to perform HTTP requests to our azure IoT hub. The Power BI platform is configurated to generate visualization from data source.
Knee Functional Telerehabilitation System for Inclusive Smart Cities
435
Fig. 7. The prototype of the developed system.
4.2
Test of Measurement Accuracy
A manual goniometer is attached to the prototype to evaluate its accuracy. It is used to combine eight measurements in a predetermined way as a reference for comparison with the data displayed in the cloud. We considered angles from 0°, which correspond to the full extension of the joint to 140°, which is the maximum flexion angle of the knee joint [23]. Table 1 and Fig. 8 show the results and error percentages, which is calculated by the following equation: %Error ¼
jMeasured value Reference j 100 Reference
ð3Þ
Where Reference is the angle value given by the manual goniometer, and Measured value is the angle value obtained using the prototype board. Table 1. Angles measured with the prototype compared to the reference. Reference (degrees) Measured value (degrees) % Error 0 0.26 – 20 20.32 1.6 40 39.59 1.02 60 60.18 0.3 80 79.97 0.03 100 101.88 1.88 120 120.88 0.73 140 139.35 0.46
436
M. El Fezazi et al.
Fig. 8. Angles measured with the prototype compared to the reference.
We note that error percentages were lower than 2%. It’s due to the different sources including the mechanical slack between the attachments of the prototype with the manual goniometer and electric noise in the circuitry [23]. 4.3
Results of Evaluation Test
In order to test the performance of the developed system, the data of knee flexion/extension exercise is collected from a subject using the implemented prototype. The knee angle measurement using the electro-goniometer require to identify anatomical body landmarks as presented in Fig. 9. The electro-goniometer axis is placed on the lateral epicondyle of the femur, while the stationary arm of the electrogoniometer is lined up with the greater trochanter at the hip, and the moveable arm of the electro-goniometer is lined up with the lateral malleolus at the fibula [24].
Fig. 9. Placement of the prototype to measure knee joint angle.
Knee Functional Telerehabilitation System for Inclusive Smart Cities
437
Fig. 10. Screenshot of the knee angle data posted in power BI platform.
The data displayed in power BI dashboard shown in Fig. 10. The knee angle is displayed both by a Gauge to know the minimum and the maximum of knee flexion/extension easily and graph to save all posted data.
5 Conclusion In line with the current development of information and communication technologies regarding smart cities building, this work focuses primarily on how assistive technologies can improve smart city characteristics in terms of smart healthcare, smart living, and smart mobility. In fact, a knee functional telerehabilitation system is developed through assistive IoT technologies, allowing the telemonitoring function of the rehabilitation process. This IoT based proposed system is used to acquire in realtime parameters for Knee kinematics assessment and transmit them to the practitioner in order to ensure real-time feedback and therapist supervision. The measured parameter is the knee joint angle. It is obtained using electro-goniometer device uses a potentiometer as the sensing element. The IoT development board provides data processing and at the same time transmission to Cloud platform in order to establish data logging and visualization. The obtained results show that measurements accuracy of the developed system is satisfactory if considering the cost and quality of the sensors and materials used in prototype implementing. And the experimental test indicates that the system can be developed by integrating some additional improvements including processing and transmission methods resting on artificial intelligence and cloud computing technologies in terms of hardware, software, and infrastructure. Furthermore, telerehabilitation engineering has prospects in Inclusive smart city applications.
438
M. El Fezazi et al.
References 1. Escamilla, S.S., Plaza, C.P., Flores, U.S.: Smart cities and sustainable development. a case study. In: Peris-Ortiz, M., Bennett, D., Pérez-Bustamante, Y.D. (eds.) Sustainable Smart Cities. Innovation, Technology, and Knowledge Management. Springer, Cham (2017). https://doi.org/10.1007/978-3-319-40895-8_5 2. Ejaz, W., Anpalagan, A.: Internet of Things for smart cities: overview and key challenges. In: Internet of Things for Smart Cities. SpringerBriefs in Electrical and Computer Engineering. Springer, Cham (2019). https://doi.org/10.1007/978-3-319-95037-2_1 3. Talari, S., Shafie-khah, M., Siano, P., Loia, V., Tommasetti, A., Catalão, J.P.S.: A review of smart cities based on the Internet of Things concept. Energies 10, 421 (2017). https://doi.org/ 10.3390/en10040421 4. Arasteh, H., et al.: Iot-based smart cities: a survey. In: 2016 IEEE 16th International Conference on Environment and Electrical Engineering (EEEIC), Florence, pp. 1–6 (2016). https://doi.org/10.1109/eeeic.2016.7555867 5. de Oliveira Neto, J.S.: Inclusive smart cities: theory and tools to improve the experience of people with disabilities in urban spaces. Université Paris-Saclay, Universidade de São Paulo, Brésil (2018) 6. Gubbi, J., Buyya, R., Marusic, S., Palaniswami, M.: Internet of Things (IoT): a vision, architectural elements, and future directions. Future Gener. Comput. Syst. 29(7), 1645–1660 (2013). https://doi.org/10.1016/j.future.2013.01.010 7. Sethi, P., Sarangi, S.R.: Internet of Things: architectures, protocols, and applications. J. Electr. Comput. Eng. 2017, 25 (2017). https://doi.org/10.1155/2017/9324035 8. Datta, S.K., Bonnet, C., Ferreira Da Costa, R.P., Härri, J.: DataTweet: an architecture enabling data-centric IoT services. In: 2016 IEEE Region 10 Symposium (TENSYMP), Bali, pp. 343–348 (2016). https://doi.org/10.1109/TENCONSpring.2016.7519430 9. Mahdavinejad, M.S., Rezvan, M., Barekatain, M., Adibi, P., Barnaghi, P., Sheth, A.P.: Machine learning for internet of things data analysis: a survey. Digit. Commun. Netw. 4(3), 161–175 (2018) 10. Aswale, P., Shukla, A., Bharati, P., Bharambe, S., Palve, S.: An overview of Internet of Things: architecture, protocols and challenges. In: Satapathy, S., Joshi, A. (eds.) Information and Communication Technology for Intelligent Systems. Smart Innovation, Systems and Technologies, vol. 106. Springer, Singapore (2019). https://doi.org/10.1007/978-981-131742-2_29 11. Rahman, L.F., Ozcelebi, T., Lukkien, J.J.: Choosing your IoT programming framework: architectural aspects. In: 2016 IEEE 4th International Conference on Future Internet of Things and Cloud (FiCloud), Vienna, pp. 293–300 (2016). https://doi.org/10.1109/FiCloud. 2016.49 12. Singh, K.J., Kapoor, D.S.: Create your own Internet of Things: a survey of IoT platforms. IEEE Consum. Electron. Mag. 6(2), 57–68 (2017). https://doi.org/10.1109/MCE.2016. 2640718 13. Khajenasiri, I., Estebsari, A., Verhelst, M., Gielen, G.: A review on Internet of Things solutions for intelligent energy control in buildings for smart city applications. Energy Procedia 111, 770–779 (2017). https://doi.org/10.1016/j.egypro.2017.03.239 14. Ana, P., Vânia, G., Ines, S.: Joint angles tracking for rehabilitation at home using inertial sensors: a feasibility study (2017). https://doi.org/10.1145/3154862.3154888 15. Douglas, R., Caldwell, G.E., Joseph, H., Gary, K., Saunders, N.W.: Research Methods in Biomechanics: Second edition (eBook) (2013)
Knee Functional Telerehabilitation System for Inclusive Smart Cities
439
16. Cerullo, G., Mazzeo, G., Papale, G., Ragucci, B., Sgaglione, L.: IoT and sensor networks security. In: Ficco, M., Palmieri, F. (eds.) Intelligent Data-Centric Systems, Security and Resilience in Intelligent Data-Centric Systems and Communication Networks. Academic Press, pp. 77–101 (2018) https://doi.org/10.1016/B978-0-12-811373-8.00004-5 17. Li, C., Xiangpei, H., Zhang, L.: The IoT-based heart disease monitoring system for pervasive healthcare service. Procedia Comput. Sci. 112, 2328–2334 (2017). https://doi.org/ 10.1016/j.procs.2017.08.265 18. Ammar, M., Russello, G., Crispo, B.: Internet of Things: a survey on the security of IoT frameworks. J. Inf. Secur. Appl. 38, 8–27 (2018). https://doi.org/10.1016/j.jisa.2017.11.002 19. Klein, S.: Azure IoT hub. In: IoT Solutions in Microsoft’s Azure IoT Suite. Apress, Berkeley (2017). https://doi.org/10.1007/978-1-4842-2143-3_3 20. Klein, S.: Ingesting data with azure IoT hub. In: IoT Solutions in Microsoft’s Azure IoT Suite. Apress, Berkeley (2017). https://doi.org/10.1007/978-1-4842-2143-3_4 21. Klein, S.D.: Azure stream analytics. In: IoT Solutions in Microsoft’s Azure IoT Suite. Apress, Berkeley (2017). https://doi.org/10.1007/978-1-4842-2143-3_5 22. Klein, S.: Real-time insights and reporting on big data. In: IoT Solutions in Microsoft’s Azure IoT Suite. Apress, Berkeley (2017). https://doi.org/10.1007/978-1-4842-2143-3_13 23. Mercado-Aguirre, I.M., Contreras-Ortiz, S.H.: Design and construction of a wearable wireless electrogoniometer for joint angle measurements in sports. In: Torres, I., Bustamante, J., Sierra, D. (eds.) VII Latin American Congress on Biomedical Engineering CLAIB 2016, Bucaramanga, Santander, Colombia, 26th–28th October 2016. IFMBE Proceedings, vol. 60. Springer, Singapore (2017). https://doi.org/10.1007/978-981-10-4086-3_98 24. Milanese, S., Gordon, S., Buettner, P., Flavell, C., Ruston, S., Coe, D., O’Sullivan, W., McCormack, S.: Reliability and concurrent validity of knee angle measurement: smart phone app versus universal goniometer used by experienced and novice clinicians. Manual Ther. 19 (6), 569–574 (2014). https://doi.org/10.1016/j.math.2014.05.009
Configuring MediBoard HIS for Usability in Hospital Procedures Youssef Bouidi(&), Mostafa Azzouzi Idrissi, and Noureddine Rais Laboratory of Applied Physics, Computer Science and Statistics (LPAIS), University of Sidi Mohammed Ben Abdallah (USMBA), Fez, Morocco [email protected], [email protected], [email protected]
Abstract. Hospital complexity required the use of an information system (IS) to ensure the proper management of the various processes of the health facility. The choice of the IS presented a problematic considering the different types and qualities of ISs. To choose one, we completed a comparative study of nine open source ISs (OSHIS). We used the DeLone & McLean IS quality evaluation model and the SQALE method to evaluate the source code. The implementation using the SonarQube platform allowed us to choose MediBoard which scored a minimal technical debt by 42.16%. The installation of MediBoard was not immediate, we conducted a maintenance intervention to complete the usability of this HIS. We used ISO 12207: software life cycle standard, and ISO 14764: software maintenance standard. The implementation of these standards allowed us to have a complete version with 63 modules. Keywords: HIS MediBoard SQALE Quality Maintenance Life cycle Standard Usability
1 Introduction The organization and traditional management of health facilities suffer from various problems caused by their complexity. Therefore, it is necessary to use an information system (IS) to help these institutions control their different processes. The World Health Organization (WHO) confirmed the lack of resources and investments and budget as well as the lack of a health sector management strategy [1]. However, this sector needs an IS that manages its different institutions and organizations. Indeed, the IT market offers several solutions: commercial SI and free IS and complete IS and IS incomplete and IS open source. But the adaptation of the IS to our needs requires maintenance work. From there, we had the idea to use an open source solution. Currently, there are several OSHIS that differ in their qualities. Indeed, there is a margin of difference in their services, functional coverage, types and licenses. For this reason, we conducted a comparative study of OSHIS and chose MediBoard as the best suited to our needs.
© Springer Nature Switzerland AG 2020 M. Ben Ahmed et al. (Eds.): SCA 2019, LNITI, pp. 440–453, 2020. https://doi.org/10.1007/978-3-030-37629-1_32
Configuring MediBoard HIS for Usability in Hospital Procedures
441
However, the use of this OSHIS was not immediate. During the installation, we faced different problems that can be seen in two types: 1. Installation problems. Indeed, MediBoard offers a distributed installation procedure that requires both computer skills and hospital expertise. According to the current procedure, the installation of this OSHIS remains blocked to the main module which manages the installation of the other modules [3]. So far, MediBoard officially proposes version 0.5 published in 2014 [3]. This version is not compatible with current environments and requires maintenance so that it can be used in our contexts. 2. Configuration problems. In fact, the current installation procedure of MediBoard is organized in several processes included in the installation manager [3]. These processes require skills and computer expertise. Then, the configurations of the modules are based on the dependencies of the installations, hence the need for health expertise in the orchestration of these modules. The lack of procedural and organizational standardization and coordination among the various contributors has had the following consequences: • The lack of a compatible and complete and recent version. • A complete installation difficulty. • An impossibility of configuration. As a result, the use of this currently available version is not immediate for most users in the community. To solve this problem, we have done work to maintain the security information system and its use. The goal of our approach is to solve all the obstacles related to the organization of the system and the maintenance processes. We have been able to standardize the installation and configuration of this system. In this article, we begin by presenting the OSHISs compared and the environments used as well as the standards used. In the method section, we used the DeLone & McLean IS quality model [4] and the SQALE method for technical quality based on the technical debt calculation [5] using the Sonar platform as the SQALE implementation. And, to make MediBoard usable, we used the ISO 12207 and ISO 14764 standards to apply maintenance processes. In the results and discussions section, we presented our results, accompanied by interpretations and explanations. Finally, we conclude with a global and final synthesis and then declare our future work.
2 Materials In our study, we used the Linux operating system, Apache HTTP Server, PHP and the MySQL database management system. We used also the standard of life cycle of information systems: ISO 15288, the standard of life cycle of software: IEEE 12207 and standards of maintenance: ISO14764 and IEEE 1219.
442
2.1
Y. Bouidi et al.
MediBoard
MediBoard is an open source hospital information system. It was developed by the OpenXtrem under GPL 2 license [3]. In a multilayer modular web architecture, using languages include PHP, XML, XHTML, JavaScript, CSS, Smarty and PEAR [3]. It is a software solution that enables the management of all activities of all levels of health institutions (MediBoard, 2014). Currently MediBoard is officially version 0.5 since 2014. This version, which is compatible with PHP 5, is multiplatform. It has a modular organization with 63 modules [3]. 2.2
OpenMRS
It is a collaborative project designed to manage healthcare, especially, in developing countries. Created by Paul Biondich and Burke Mamlin from the Indiana University School of Medicine in 2004 [6]. 2.3
OpenEMR
It is an Electronic Health Record for a medical practice management. It was originally developed by Syntech organization in 2001. Now, the system is on version 5 since 2017 [7]. 2.4
HospitalOS
It is a HIS and a research/development project designed to every small sized hospital. It was created and developed for Thailand community. It is now until version 3.9 [8]. 2.5
Care2x
It is an integrated HIS started as “Care 2002“project in 2002. The first official release was until version 1.1 in 2004. The last stable release was in 2012 until version 2.6.29 [9]. 2.6
OpenHospital
It was developed by Informatici Senza Frontiere, in collaboration with students of Volterra San Donà di Piave Technical high school, in 2005. This HIS is at its seventh release [10]. 2.7
MedinTux
It was initiated by the doctor Roland Sevin since 10 years. It is distributed under the CeCiLL V2 license which is equivalent to the GPL license. It was originally written for French emergency services [11].
Configuring MediBoard HIS for Usability in Hospital Procedures
2.8
443
HOSxP
It is a HIS used in over 70 hospitals in Thailand. It was called KSK HDBMS. Its development started in 1999 by Suratemekul to be continued by employers of its company Bangkok Medical Software. It is distributed under a GPL license and free only for its Primary Health Care Unit version [12]. 2.9
PatientOS
It is a HIS for small hospitals and clinics. It is a web-based application under the GPL license [13]. 2.10
Linux
Firstly, we used Linux, which is an open source operating system, to install the HIS MediBoard. Secondly, to give us the possibility to work in a totally open source environment. Among Linux operating systems, we chose Debian. It is a multilingual with a stable modular monolithic kernel. Debian is founded by Ian Murdock under a free license. Its first version was 16 August 1993. Currently it is at version 9.3 since 9 December 2017 [14]. In fact, Linux is our operating system (OS) suitable for preparing the installation environment of MediBoard. This OS gave us an opportunity to install the prerequisites of MediBoard, a control on rights to read and write on various file of MediBoard and a flexibility in migration between PHP versions. 2.11
Apache
The client-server architecture of MediBoard requires an http server that uses the PHP interpreter [2]. Apache HTTP Server is a free and multi-platform, written by the C language. The first version was in 1995. Currently it is in version 2.4.29. Apache has several features such as the interpretation of the PHP error handling using the log [15]. 2.12
PHP
We installed PHP considering that is the developing language of MediBoard [3]. The Hypertext Preprocessor is an open source programming language used to generate dynamic web pages using the http server. The first version was published in 1994 by Rasmus Lerdorf. PHP is written using the C language, it is multi-platform, imperative, object-oriented and operates on a client-server architecture. Currently it is in version 7.2.2 from 1 February 2018 [16]. 2.13
MySQL
MediBoard needs a database to be operational. The owners of this system have used MySQL [3] which is a relational database management system developed by MySQL AB and Oracle. The first version was launched in 1995. MySQL is multiplatform developed using C/C++ languages and it is currently in its 5.7.20 release. It has two licenses, one of which is free [17].
444
Y. Bouidi et al.
2.14
IEEE 12207
It is an international standard for software life cycle processes. Introduced for the first time in 1995, it aims to be a primary standard defining the necessary processes for the development and maintenance of software systems, including the results and activities of each process [18]. This standard was published by the International Standards Organization in the field of information technology systems [19]. This is a response to a critical need for management and software design [19]. This standard establishes a constructed architecture of a set of processes and interrelations based on two basic principles [20]: • Modularity: The processes are modular; they are more coherent and loosely coupled as far as possible. This enables the distribution and consolidation processes as required [20]. • Responsibility: A process is responsible for a part of the life cycle of the software. This concept allows to study the life cycle functions as subjects and treat them separately [20]. These processes are grouped into three main classes: • The primary processes are the main drivers of the software life cycle. They are the acquisition, supply, development, operation and maintenance [19]. • The support processes include documentation, configuration management, quality assurance, the joint review, audit, verification, validation and troubleshooting [19]. • Organizational processes are management, infrastructure improvement and training [19] (Fig. 1).
Fig. 1. IEEE 12207 process software lifecycle [18].
Configuring MediBoard HIS for Usability in Hospital Procedures
2.15
445
ISO 14764
He described the management of the maintenance process described in ISO 12207, including the amendments. It also establishes definitions for the different types of maintenance. The scope of ISO 14764 includes maintaining multiple software products with the same maintenance resources [18] (Table 1). Table 1. The maintenance process ISO 14764 [18] Maintenance process 1. Implementation process 2. Analysis of problems and changes 3. Implementation of the change 4. Review maintenance/acceptance 5. Migration 6. Removing software
3 Methods 3.1
The DeLone & McLean Model
The comparative study concerns OSHISs qualities, so we have evaluated them to select the best one to use. The DeLone & McLean model offers three dimensions of IS quality evaluation. Initially, in 1992, this model defined only the system and information quality as the two only dimensions that judged the IS quality [4] (Fig. 2).
Fig. 2. The first model of DeLone & McLean [4]
An easy access, a short response time and practical tools for users determine the system quality, which contributes to a more efficient work. The information quality produced is determined by information accuracy, accessibility, completeness and reliability.
446
Y. Bouidi et al.
However, this model cannot judge the IS quality without use, examination of user satisfaction and verification of impact on the concerned organization. For this reason, this model was improved in 2003 by adding a third dimension concerning quality of service [4] (Fig. 3).
Fig. 3. The improved model of DeLone & McLean [4]
3.2
The SQALE Method
Technically speaking, SQALE method is a method that measures technical quality using technical debt concept. The SQALE method is based on the technical debt concept, which consists of measuring the quality indices of the technical IS characteristics after having analyzed them. The quality indicators present these measures by defining the technical debt that characterizes each of IS characteristics [21] (Fig. 4).
Fig. 4. The characteristics of SQALE quality [21]
Configuring MediBoard HIS for Usability in Hospital Procedures
447
According to Jean-Louis Letouzey [21], the SQALE analysis model performs two main tasks: the first applies rules to standardize measures by transforming them into costs and the second sets rules to aggregate these standardized values. The SQALE method defines the cost aggregation rules either in the quality model tree or in the artefact hierarchy of source code. The SQALE method defines four indicators related to the quality characteristics, allowing a highly synthesized representation of the SI quality [21]. The SQALE rating consists to produce a derived measure or an ordinal scale subdivided in five levels from A (Green) to E (Red). The Kiviat diagram consists to present the SQALE evaluation in concentric areas and targeting the quality of each project according to its values. It presents in the same diagram all compared projects characteristics ranking according to the quality model. The SQALE pyramid helps to make appropriate decisions with considering the dependence between quality characteristics model and the SI life cycle. The fourth indicator is the SQALE Debt Map, which represents the artefacts of the assessment scope drawn on two dimensions: the first is the technical debt (SQI) and the second is the business impact (SBII) [21]. 3.3
Comparative Study Implementations
To implement this method, we used the SonarQub platform [22]. It has a scanner that analyzes the source code to determine the characteristics and attribute them their technical debt scores. This platform respects the Client-Server architecture. In the server, the SQALE method is implemented to evaluate the source codes and to send to the Client the measures to be presented in indices form and indicators that allow us the subtraction of the according technical debt to each characteristic. Concerning system and information quality measurements, there are several models that implement the DeLone & McLean model. The implementations DeLone & McLean (2003, 2004) and Holsapple & Lee-Post (2006) have confirmed the criteria that allow the system and information evaluation. These judgments must be recovered from the final users of IS (Fig. 5). In our study, we have examined the technical side to compare the SIHOSs and judge their service qualities. After loading the source code in the SonarQub platform, we analyzed each of the measurements retrieved by the Sonar server that implements the SQALE method. Subsequently, we recovered the measures and indicators that we will present in the next section and discuss them to confirm the choice of the good OSHIS. On the other hand, to examine the OSHISs activity, we have compared the two criteria retrieved directly from the Open Source community [24]. 3.4
Maintenance of MediBoard
In this part of our method we have applied a maintenance of HIS MediBoard. In fact, according to the standard procedure of ISO 14764, we have identified the problem, then we have established a causal analysis and a design of our solution and finally we proposed our implementation.
448
Y. Bouidi et al.
Fig. 5. Holsapple & Lee-Post model [23]
3.4.1 Identifying the Problem In MediBoard, installation performed using the wizard is an initial installation requiring a second step performed by the administrator. Indeed, after the administrator authentication, the system is configured in three non-updated modules: • Administration. • Establishment. • Permissions. Dependencies blocked the modules updates to have a full installation of MediBoard. Indeed, it requires that the administration module is updated. But, due to the cross dependencies, the update of this module was blocked in a version which did not allow the display of the rest of modules. 3.4.2 Analysis and Design To solve the problem of dependencies, we have marked our intervention on two sides. The first one is global, when we modified the installations procedure. The second is local, when we treated the modules installation. In fact, the installation procedure initially follows two parts: an installation guided by the MediBoard wizard and a module installation carried out by the administrator. But this procedure makes the system installation so complicated because of the impossibility to run the second part without obtaining the list of modules by updating the administration module. In
Configuring MediBoard HIS for Usability in Hospital Procedures
449
addition, the installation of modules requires some expertise in medical and computer fields to identify the modules which can be installed and dependencies. So, the administrator will be confronted by a major problem of compatibility of modules versions. For these reasons, we decided to change the traditional MediBoard installation procedure to a non-subdivided procedure by integrating functionalities of both parts in one. This procedure will ensure a complete installation of the system. That is mean a kernel accompanied by all modules. Using an architectural point of view, the modification we have made involves the two following steps: • Completing the database. • Finalizing the configuration. 3.4.3 Implementing Change Our intervention in the implementation code of our solution, was consisting in two parts: first the creation of the database and secondly the configuration. In fact, we used two different methods. For the database, we replaced the original SQL file with the last one to avoid the partial construction of the database and to exempt the administrator from managing the database. But the new SQL file was the result of adding the SQL queries included in the module’s installation files to the original SQL code. Regarding the configuration, we proceeded by addition: we kept the default configuration, we added the final configuration, we managed all installations and updates related to the database. This method helps the administrator to get rid of update issues and keeps the system homogeneous.
4 Results and Discussions 4.1
Comparative Study Results
The method mentioned in the previous section allowed us to have a set of results that prove the technical potential of MediBoard among other SIHOSs. The measurements obtained from the SonarQub platform were Reliability, Security, Maintainability, Complexity, Documentation and Rules Compliance. Technically speaking, the SQALE method implementation, SonarQub, has determined six characteristics that have been evaluated for MediBoard (respectively OpenEMR, OpenMRS, OpenHospital, HospitalOS, PatientOS, Care2x, MedinTux and HOSxP) with a debt of 42.16% (respectively 53.23%, 54.5%, 65%, 66.1%, 65.2%, 56.96%, 52.13% and 75.5%) for the SQALE Quality of Technical Support. In Table 2, we present the details of the obtained results recovered from the SonarQub for each SIHOSs (Fig. 6). Indeed, the implementation of the SQALE method by the SonarQub platform has allowed us to compare the OSHISs technically. We compared the average of the six measures marked for each system. This average presents the quality debt rate of the technical support SQALE. So, based on the results, we chose the OSHIS MediBoard which scored a minimum debt of 42.16%.
450
Y. Bouidi et al. Table 2. Classification of OSHISs according to their SQALE quality debts
SIHOSs
Reliability Security Maintainability RC
Complexity Documentation SQ
MediBoard MedinTux OpenEMR OpenMRS Care2x OpenHospital PatientOS HospitalOS HOSxP
36.4% 60.6% 62% 58.4% 56% 64.6% 73.4% 74.2% 92%
2.4% 23.2% 4.2% 20.6% 38.8% 13.6% 48.6% 41.4% 60.2%
24.8% 43.2% 42.8% 65.4% 80.6% 62.4% 62.8% 56.4% 84%
21.4% 27% 25% 26.8% 24.4% 74.6% 43.4% 63.6% 41.4%
70.8% 80.2% 89% 58.4% 83% 95.4% 72.8% 65.4% 77.4%
97.2% 78.6% 96.4% 97.4% 59% 79.4% 90.2% 95.6% 98%
42.16% 52.13% 53.23% 54.5% 56.96% 65% 65.2% 66.1% 75.5%
5 4 3 2 1 0
Reliability
Security
maintainability
RC
Complexity
DocumentaƟon
Fig. 6. The technical debts of the SQALE characteristics of the OSHISs
These results have proved the technical potential of MediBoard. Indeed, its minimal debt with a complexity of 2.4%, a maintainability of 21.4% and a reliability of 36.4 and a security of 24.8%; We conclude that this OSHIS insured the technical support, i.e. MediBoard has won the third-dimension challenge, which directly affects the IS service quality according to the DeLone & McLean model. This technological aspect in the IS presents a basis for the rest of the dimensions. Indeed, the technical test allowed us to examine implicitly the other axes that constitute the IS. The reliability and security of data proves the MediBoard advantage of health information. Because hospital information is sensitive and requires a degree of confidence and protection, the examined characteristics prove the power of the MediBoard quality in front of the rest of the OSHISs. Not only the information quality, but also MediBoard has won the system evaluation challenge by the technical examination of complexity and maintainability. This OSHIS proves its strength by a minimal technical debt in the most complicated difficulty of the health system and the management of its institutions. In other words, MediBoard marks its advantage by reducing the complexity of the system.
Configuring MediBoard HIS for Usability in Hospital Procedures
4.2
451
MediBoard Maintenance Results
By applying the maintenance steps mentioned in the ISO 14764 standard, we have successfully implemented MediBoard. Indeed, we were able to detect the worries that blocked the complete installation of the system and analyze them and propose solutions in order to solve them and finally we applied the modifications. We scored a very important transition in the installation of MediBoard. In fact, we managed to have a complete installation provided by the system itself (Figs. 7 and 8).
Fig. 7. MediBoard before implementation of our solution
Fig. 8. MediBoard after implementing our solution
452
Y. Bouidi et al.
5 Conclusion The adoption of open source has been a challenge for the management of health institutions despite the problems facing the health sector. Our study was more than a comparison between the chosen OSHIS; we have also proven the power and quality of these information systems and their services. In this study, we proved that MediBoard is the best system security information system. However, this confirmation is only the result of a technical evaluation and a statistical study. The MediBoard installation problem presents a gap in the system development for the open source community. MediBoard contributors must work under a continuous chain to evolve. In other words, all contributors must be based on the current state of the system to modify or add features. The only people responsible for the validation of contributions are the owners of the system. This must redefine the current system’s state. The current version of the system must be available and operational. So, the contributor can visualize and test the system before intervention and use. The problem of blocking the installation gives contributors the impression of non-functioning of the system. which imposes for them two choices: • Due to the difficulty of installation, the contributor decides to leave the project. • If the contributor can identify problems and solve them, he decides to keep solutions for himself and use them in business purposes. We were able to identify the cause of the problem and explain it and present a solution and implement it. The results of our solution have been great. We were able to completely install MediBoard and solve the problem of dependencies. We have provided a system with a full installation wizard that will not make the administrator accountable for installations or updates. After choosing MediBoard as the best OSHIS that meets our needs and introduce our installation work, we will discuss in future works our maintenance interventions and implementations of this system.
References 1. R. O. f. t. E. M. World Health Organization: Les fonctions essentielles de santé publique: de l’évaluation à l’action, Rabat (2016) 2. Hannah, K.J., Ball, M.J.: Strategic Information Management in Hospitals. An Introduction to Hospital Information Systems. Springer, New York (2004) 3. MediBoard (2014) 4. Delone, W.H., Mclean, E.R.: The DeLone and McLean model of information systems success: a ten-year update. J. Manage. Inf. Syst. 19, 9–30 (2003) 5. Letouzey, J.L.: Meaningful insights into your Technical Debt (2012) 6. OpenMRS LLC: Introducing OpenMRS 2.0 (2014) 7. OpenEMR Project: OpenEMR Project (2012) 8. Webster, P.C.: The rise of open-source electronic health records. Lancet 377, 1641–1642 (2011)
Configuring MediBoard HIS for Usability in Hospital Procedures 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24.
453
care2x (2013) Informatici Senza Frontiere: Open Hospital - The project (2014) MedinTux: Documentation MedinTux (2012) SourceForge: HOSxP (2002) SourceForge: PatientOS (2007) Debian.org: A propos de Debian (2017) The Apache Software Foundation: Apache HTTP Server Project (2017) Achour, M., Betz, F., Dovgal, A., Lopes, N., Magnusson, H., Richter, G., Seguy, D., Vrana, J., Manuel, P.H.P., Cowburn, P. (éds.) PHP Documentation Group (2018) Oracle: MySQL Documentation (2018) Croll, P.R.: A best practice approach for software maintenance - sustaining critical capability. In: 16th Annual Systems and Software Technology Conference (2004) Singh, R.: International standard ISO/IEC 12207 software life cycle processes. Softw. Process Improv. Pract. 2, 35–50 (1999) ISO/IEC 12207: Information Technology - Software life cycle processes (1995) Letouzey, J.-L.: The SQALE Method for Managing Technical Debt (2016) SonarSource (2017) Dorobat, I.: Models for measuring e-learning success in universities: a literature review. Informatica Economica 18, 77–90 (2014) Black Duck Software (2016). https://www.openhub.net/
Type 2 Diabetes Mellitus Prediction Model Based on Machine Learning Approach Othmane Daanouni1(&), Bouchaib Cherradi1,2(&), and Amal Tmiri1(&) 1
LaROSERI Laboratory, Chouaib Doukkali University, El Jadida, Morocco [email protected], [email protected], [email protected] 2 STICE Team, CRMEF Casablanca-Settat, El Jadida, Morocco
Abstract. A healthcare system using modern computing techniques is the highest explored area in healthcare research. Researchers in the field of computing and healthcare are persistently working together to make such systems more technology ready. Diabetes is considered as one of the deadliest and chronic diseases it leads to complications such as blindness, amputation and cardiovascular diseases in several countries and all of them are working to prevent this disease at early stage by diagnosing and predicting the symptoms of diabetes using several methods. The motive of this study is to compare the performance of some Machine Learning algorithms, used to predict type 2 diabetes diseases. In this paper, we apply and evaluate four Machine Learning algorithms (Decision Tree, K-Nearest Neighbors, Artificial Neural Network and Deep Neural Network) to predict patients with or without type 2 diabetes mellitus. These techniques have been trained and tested on two diabetes databases: The first obtained from Frankfurt hospital (Germany), and the second is the wellknown Pima Indian dataset. These datasets contain the same features composed of mixed data; risk factors and some clinical data. The performances of the experimented algorithms have been evaluated in both the cases i.e. dataset with noisy data (before pre-processing/some data with missing values) and dataset set without noisy data (after pre-processing). The results compared using different similarity metrics like Accuracy, Sensitivity, and Specificity and ROC (Receiver Operating Curve) gives best performance with respect to state of the art. Keywords: Diabetes diseases Machine learning Deep learning Computer aided diagnosis (CAD) Prediction systems
1 Introduction Due to its continuously increasing occurrence, more and more families are influenced by diabetes mellitus. Most diabetics know little about their health quality or the risk factors they face before diagnosis. The adults suffering from diabetes leads to dangerous consequences complications such as blindness, amputation and cardiovascular diseases [1, 2]. In medicine, the diagnosis of diabetes is according to many risk factors and clinical data [3] such as age, fasting blood glucose, glucose tolerance and random blood glucose levels [4]. In other hands, Diabetic Diet for diabetics is simply a © Springer Nature Switzerland AG 2020 M. Ben Ahmed et al. (Eds.): SCA 2019, LNITI, pp. 454–469, 2020. https://doi.org/10.1007/978-3-030-37629-1_33
Type 2 Diabetes Mellitus Prediction Model
455
balanced healthy diet, which is vital for diabetic treatment [5]. The earlier diabetes diagnosis is obtained, the much easier we can control it. Machine Learning is concerned with the development of algorithms and techniques that allow computers to learn and gain intelligence based on past experience. It is a branch of Artificial Intelligence (AI) and is closely related to statistics. By learning it means that the system is able to identify and understand the input data so that it can make decisions and predictions based on it [6]. In recent years, several decision support systems for human diseases diagnosis using computational techniques namely artificial neural network (ANN), fuzzy logic [7], neuro-fuzzy, machine learning are developed. For example, the authors in [8, 9] used traditional machine learning algorithms such as linear kernel support vector machine (SVM-linear), radial basis function (RBF) kernel support vector machine, knearest neighbor (k-NN), artificial neural network (ANN) for diabetic patients’ classification. Actually, machine learning methods are widely used in predicting different diseases using clinical data [10], and they get promising results [11]. In this research, dedicated to the study and evaluation of some machine learning techniques applied to predicting patients with/without type 2 diabetes disease, we experimented four algorithms namely: decision tree (DT), K-nearest neighbor (KNN), artificial neural network (ANN) and a new deep neural network (DeepNN) architecture to predict diabetes. This paper is organized as follows: Sect. 2 illustrate other relevant related works on various classification techniques used for the prediction of different human diseases and we focused on type 2 diabetes diseases. Section 3 describes the used Datasets structure and different machine learning algorithms. Implementation, results and discussion are given in Sect. 4. Section 5 concludes this paper and present some perspectives at this work.
2 Some Related Work This section, we review a few important works that are closely related to diabetesaffected patient’s classification issue. Kaur et al. [8] Studied the performance of five different models based upon linear kernel support vector machine (SVM-linear), radial basis kernel support vector machine (SVM-RBF), k-nearest neighbour (KNN), artificial neural network (ANN) and multifactor dimensionality reduction (MDR) algorithms to detect diabetes in female patients. They conclude that SVM-linear and K-NN are two best models for patient classification. Kandhasamy et al. [9] compared four prediction models (Decision Tree J48, KNN, Random Forest, SVM) using University of California Irvine(UCI) machine learning data repository and 8 important attributes under two use cases; before and after pre-processing the dataset. The study concludes that after pre-processing and removing the noisy data, the dataset had more accurate result compared whit the case before preprocessing. Sisodia et al. [12] introduced a design model, which can detect diabetes at an early stage using three machine learning classification algorithms namely Decision Tree, SVM and Naïve Bayes. The experiments are performed on Pima Indian diabetes database and the performance of these three algorithms are calculated using various
456
O. Daanouni et al.
measures like accuracy, precession, recall. Result obtained show that Naïve Bayes outperforms with highest accuracy of 76.30% compared to the two other algorithms. Meng et al. [13] Introduced and compared three prediction models for diabetes or prediabetes incidence using demographics characteristics and lifestyle risk factor. Results indicated that the decision tree model performed best on classification accuracy.
3 Materials and Methods In this section, we describe the approach we adopted in order to discriminate between patients affected from diabetes and not affected ones. 3.1
Diabetics Database Description
In this section, we give a brief overview on the used datasets. Table 1 summarize the datasets features description and value range. Table 1. Diabetes dataset attributes Attribute
Description 1 Pregnancies Number of times pregnant 2 Glucose Plasma glucose concentration (mg/dl) 3 BloodPressure Diastolic blood pressure (mm Hg) 4 SkinThickness Skin fold thickness (mm) 5 Insulin serum insulin (mu U/ml) 6 BMI BMI (kg/m2) 7 DiabetesPedigreeFunction Diabetes pedigree 8 Age Age in years 9 Outcome 0 Not diabetic 1 diabetic
Range 0–17 0–199
0–122
0–99 0–846 0–67.1 0.078–2.42 21–81 0 1
First dataset was obtained from Frankfurt hospital, Germany [14] (dataset 1) and the second dataset is the Pima Indian dataset [15] (dataset 2). These datasets are consisting of two parts: the healthy and the diabetic patients. The first dataset has 2000 instance and the second data has 768 of patient data with 8 attributes/features and one output giving the label/outcome of the considered patient (0: Not diabetic, 1: Diabetic).
Type 2 Diabetes Mellitus Prediction Model
457
Since these datasets contain the real labels, they are adaptable to supervised Machine Learning techniques. 3.2
Machine Learning Algorithms for Prediction
In this sub-section, we present some elements on the considered Machine leaning algorithms. In fact, the following algorithms are considered for comparison analysis of predicting diabetes diseases from the above presented datasets. K-Nearest Neighbour’s Algorithm (KNN) K-Nearest Neighbours is a simple algorithm, it’s a lazy, non-parametric and instancebased learning algorithm, KNN is applied to find out the class, to which new unlabelled object belongs. For this, a ‘k’ is decided (where k is number of neighbour’s elements to be considered) which is generally odd and the distance between the data points that are nearest to the objects is calculated by the ways like Euclidian’s distance, Hamming distance, Manhattan distance or Minkowski distance. The following Fig. 1 illustrate an example for binary classification.
Fig. 1. Predicting new example using KNN with two class.
As shown in the Fig. 1 above, the value of k (positive integer) is determined by inspecting the data set. Cross-validation is a method used to retrospectively determine a good k value by using an independent data set to validate the k. In this study, we use 10 cross-validations and we have taken the values (k = 2) because it produces the best results. Decision Tree Algorithm (DT) Decision tree is the most powerful and popular tool for classification and prediction. A Decision tree is a flowchart like tree structure, where each internal node denotes a test on an attribute, each branch represents an outcome of the test, and each leaf node (terminal node) holds a class label. Figure 2 illustrate the application of Decision tree technique to classify a person as fit or unfit based of two attributes (age and food).
458
O. Daanouni et al.
Fig. 2. Illustration of decision tree algorithm
Artificial Neural Network An Artificial Neural Network learning algorithm, is a computational learning system that uses a network of functions to understand and translate a data input of one form into a desired output, usually in another form. The concept of the artificial neural network was inspired by human biology and the way neurons of the human brain function together to understand inputs from human senses.
Fig. 3. Example of neural network with one hidden layer.
In this example below (Fig. 3), the layers of functions between the input and the output are what make up the neural network. Deep Neural Network A deep neural network is a neural network with a certain level of complexity, a neural network with more than two layers. Deep learning is learning multiple levels of representation and abstraction, helps to understand the data such as images, audio and text.
Type 2 Diabetes Mellitus Prediction Model
459
Fig. 4. Deep neural network with three hidden layers.
Deep learning can be expensive and require massive datasets to train itself on. Since, in deep learning, more the neurons (cells in hidden layers) are, the more features it creates, and correspondingly it needs more data to train on. Figure 4 present a simple deep neural network with three hidden layers. 3.3
Proposed Predictive Model
Fig. 5. Proposed model diagram.
The model diagram in Fig. 5 describe the main elements of the proposed predicting system. The flowchart illustrates the steps of the research conducted to construct the diabetes diseases predicting models. The system is based on training and evaluating
460
O. Daanouni et al.
four Machine Learning algorithms, the KNN, Decision Tree, ANN and DeepNN on part of diabetes dataset to construct the best models for each algorithm; before using them for test data. We selected 80% of the data for the training process, while 20% is selected for the testing process. Except for ANN and Deep NN, we dived the data into three-part as follow 70% for training and 15% for validation while last 15% of data is selected for the testing process. we used two scenarios for analyzing the algorithms. One is without pre-processing and other with pre-processing. With the pre-processing technique we used a linear interpolation, in this technique the zero values in our dataset have been treated as missing values except in the attribute 1 “Number of Pregnancy” and all other have been replaced by linear interpolation. The best model in four classifiers is selected based on the highest accuracy. In KNN algorithm, we varied K from 1 to 30, we used a Euclidian distance, and in order to pick the optimal model we perform a 10-fold cross validation. In decision tree, we used an Optimize Hyperparameters to minimize cross-validation loss. In ANN algorithm, a feed-forward was used with 8 input and 1 output. The sigmoid function was used for the ANN activation function. In DeepNN algorithm we used a cascade forward network that is similar to feed-forward network, but include a connection from the input and every previous layer to following layers, we used a three hidden layer with 20 neurons in 2 first hidden layers and 10 neurons for the last hidden layer. 3.4
Performance Evaluation Methods
In this sub section, we present some well-known metrics usually used to evaluate performance of machine algorithms used in both classification and prediction. Confusion Matrix for Binary Prediction To construct the confusion matrix, the predicted output/labels are the class that is predicted by the classifier, and the actual outputs is the class that is given in the data set (Table 2). Table 2. Confusion matrix for binary prediction. Actual outputs/labels (ground truth) Patient with disease (True/1) Patient without disease (False/0)
• • • •
Predicted output/labels Patient with disease (Positive/1) TP FP
Patient without disease (Negative/0 FN TN
True Positive (TP): number of correctly classified patients as diabetic. True Negative (TN): number of correctly classified patients as not diabetic. False Negative (FN): number of patients incorrectly predicted as not diabetic. False Positive (FP): number of patients incorrectly predicted as diabetic.
Type 2 Diabetes Mellitus Prediction Model
461
Performance Evaluation Measures In this work, we used both a confusion matrix to appraise the performance of the four machine-learning algorithms and four evaluation metrics (accuracy, sensitivity, specificity and ROC (Receiver Operating Curve). The classification accuracy measures the proportion of cases correctly classified. Sensitivity measures the fraction of positive cases that are classified as positive. Specificity measures the fraction of negative cases that are classified as negative [16], ROC (Receiver Operating Curve) It is one of the most important evaluation metrics for checking any classification model’s performance. It is also written as AUROC (Area under the Receiver Operating Characteristics). The following equations define the used similarity metrics. Specificity ¼
TN TN þ FP
ð1Þ
Sensitivity ¼
TP FN þ TP
ð2Þ
TP þ TN TP þ TN þ FP þ FN
ð3Þ
Accuracy ¼
4 Results and Discussion 4.1
Results Evaluation and Comparison
In this sub-section, we evaluate and compare the performance of the four supervised machine learning algorithms using test data to evaluate the quality of owner pretrained model using various similarity measures under two scenarios: • Scenario 1: Dataset with noisy data/data with missing values/original dataset (before pre-processing). • Scenario 2: Dataset without noisy data/data with estimated missing values (after pre-processing). Testing Results on Dataset1 Here we present the results relatively to the application of the four models, generated in the training phase (1800 patients from dataset1), on testing data (200 patients from the same dataset 1). Dataset Without Pre-processing (Scenario 1) In this part of the work, the four proposed predicting techniques trained models are applied to classify the 200 testing patients in two cluster: diabetic’s patients and patients without diabetes disease. To do this, the predicting trained models are executed on original testing data and the results are recorded.
462
O. Daanouni et al.
Here, a comparison is made to find which classifier is the best for classification when the dataset is noisy. Table 3 presents testing results in terms of TP, TN, FP and FN values. Table 3. Confusion matrix of diabetes diseases prediction using machine-learning algorithms tested on 200 patients from dataset1 (Without pre-processing). TN FN TP FP ANN 118 16 60 6 KNN (K = 2) 134 0 57 9 DeepNN 131 3 65 1 Decision tree 131 6 60 3
From Table 4 below, we observe that the DeepNN model achieves prediction more accurately than other three predictor’s models. Another finding from this table is that KNN model achieves better accuracy when the k value equal 2. Table 4. Machine learning algorithms tested on 200 patients from dataset1 (Without preprocessing). ANN Accuracy (%) 89.0 Sensitivity (%) 90.91 Specificity (%) 88.06
Deep NN 98 98.48 97.76
KNN (k = 2) DT 95.5 95.50 100 90.91 93.71 97.76
Dataset After Pre-processing (Scenario 2) The pre-processing of data for estimating missing values in original data set is done with linear interpolation method. Zero values in our dataset have been treated as missing values except in the attribute 1 “Number of Pregnancy” and all others have been replaced using pre-processing stage. In the pre-processing stage, we used linear interpolation. The interpolated value at a missing value point is based on linear interpolation of the values at neighboring grid points in each respective dimension. Requires at least 2 points. Linear interpolation involves estimating a new value by connecting two adjacent known values with a straight line. If the two known values are (x1, y1) and (x2, y2), then the y value for some point x is (Fig. 6): y ¼ y1 þ ð x x 1 Þ
y2 y1 x2 x1
ð6Þ
In the following Table 5, we present the new results in terms of TP, TN, FP and FN values. From the Table 6 we observe that the accuracy of DeepNN are enhanced and the trained model achieve best accuracy among other algorithms and it’s above 99.5%.
Type 2 Diabetes Mellitus Prediction Model
463
Fig. 6. Linear interpolation is a straight line fit between two data points. Table 5. Confusion matrix of diabetes diseases prediction using machine-learning algorithms tested on 200 patients from dataset1 (after pre-processing). TN FN TP FP ANN 108 26 58 8 KNN K = 2 134 0 60 6 Deep ANN 134 0 65 1 Decision tree 127 5 61 7
Table 6. Machine learning algorithms tested on 200 patients from dataset1 (after preprocessing). ANN Accuracy (%) 83 Sensitivity (%) 87.88 Specificity (%) 80.60
Deep NN 99.5 98.48 100
KNN (k = 2) DT 97 94 100 92.42 95.71 94.78
Testing Results on Dataset2 In this subsection, we test the best pretrained models with Pima Indian dataset. In this test, we apply the best pretrained models that get high accuracy among the two scenarios (with/without pre-processing) in training phase. So, the testing data is now consisting of the 768 patients in Pima Indian dataset (dataset2). Test Results on Dataset2 Without Pre-processing Here, a comparison is made to find which machine-learning algorithm gives the best prediction when the dataset is noisy. Table 7 presents testing results in terms of TP, TN, FP and FN values. In Table 8 below, it’s clear that the decision Tree algorithm gives the best results. Therefore, the Decision Tree machine learning classifier can predict the risk of having diabetes disease with more accuracy compared to the other classifiers. Test results on Dataset2 after pre-processing
464
O. Daanouni et al.
Table 7. Confusion matrix of diabetes diseases prediction using machine-learning algorithms tested on 768 patients. ANN KNN K=2 DeepNN Decision tree
TN 434 469 405 491
FN 66 31 95 22
TP 240 189 217 246
FP 28 79 51 9
Table 8. Machine learning methods tested on 768 patients. ANN Accuracy (%) 87.76 Sensitivity (%) 89.55 Specificity (%) 86.80
DeepNN 80.99 80.97 81.00
KNN (k = 2) DT 85.68 95.96 85.91 91.79 85.58 98.20
Table 9. Confusion matrix of diabetes diseases prediction using machine-learning algorithms tested on 768 patients. ANN KNN K=2 DeepNN Decision tree
TN FN TP FP 399 101 210 58 495 5 254 14 483 17 257 11 491 22 246 9
Table 10. Machine learning methods tested on 768 patients. ANN Deep NN Accuracy (%) 79.30 96.35 Sensitivity (%) 78.36 95.90 Specificity (%) 79.80 96.60
KNN (k=2) 97.53 98.07 97.25
DT 95.96 91.79 98.20
The pre-processing of data for estimating missing values in original Pima Indian dataset set is also done with linear interpolation method. In the following Table 9, we present the new results in confusion matrix for binary prediction. According to Table 10 we found that KNN (with k=2) has a better performance than the three other algorithms do, and the accuracy is of 97.53%. Performance Comparison Using ROC (Receiver Operating Curve) We have evaluated and compared performance of different models using Receiver operating characteristic (ROC) curve. ROC is a well-known tool to visualize the performance of a binary classifier algorithm [17]. It is a plot of true positive rate (Sensitivity) against false positive rate (Specificity) as the threshold for assigning observations are varied to a particular class. The area under the ROC curve
Type 2 Diabetes Mellitus Prediction Model
465
(AUC) refers to the probability that a classifier will rank a positive instance higher than a negative one, (AUC) value of a classifier may lie between 0.5 and 1. Values below 0.50 indicated for a set of random data, which could not distinguish between true and false. An optimal classifier has the value of the area under the curve (AUC) near 1.0. If it is near 0.5 then this value is comparable to random guessing [18]. ROC Area for Testing Results on Dataset1 Here a comparison of the ROC performance of the two scenarios (without and with preprocessing) on Dataset1 is shown in Table 11. Table 11. Performance comparison of classification Algorithms using ROC Area KNN (K=2) DT ANN DNN Without pre-processing 0.9318 0.9434 0.8948 0.9812 With pre-processing 0.9545 0.936 0.8424 0.9924
Fig. 7. Roc curve for DNN model (with preprocessing).
As shown in Table 11 and Fig. 7 the highest value of AUC (0.9924) indicates that DNN is optimal classifier for diabetic dataset. ROC Area for Testing Results on Dataset2 Table 12. Performance comparison of classification Algorithms using ROC Area KNN (K=2) DT ANN DNN Without pre-processing 0.8216 0.95 0.8818 0.8099 With pre-processing 0.9689 0.95 0.7908 0.9625
466
O. Daanouni et al.
Fig. 8. Roc curve for KNN model (with preprocessing)
In this section, we compared the performance of our pre-trained models using Pima Indians diabetes dataset using ROC metric. Table 12 and Fig. 8 shows the highest values obtained with KNN (k = 2) and DNN. Implementation Setup and Computational Time Analysis All Simulations were implemented using MATLAB R2018b. The design of the language makes it possible to write a powerful program in a few lines. Table 13. Comparison of Computational time of 4 employed method in seconds Without pre-processing With pre-processing KNN 6.56 s 7.26 s DT 47.32 s 33.61 s ANN 0.57 s 0.30 s DeepNN 9.46 s 7.62 s
The simulation carried out on a computer with Core I7-7500U (2.7 GHz processor) and 16 GB Ram. As shown in Table 13 the computational time with pre-processing is faster in the most cases compared to without pre-processing. Compared Result with Related Work In Table 14, we present a comparative overview of some relevant works on predicting diabetes disease using Machine Learning techniques and comparing it with algorithms result from Table 4 in our study. According to the above experiments finding suitable attributes, classifier and data mining method are very interesting.
Type 2 Diabetes Mellitus Prediction Model
467
Table 14. Comparison of our results with some related work. Authors/year
Predictive models
Harleen Kaur -SVM linear et al. (2018) -RBK SVM [8] -KNN -ANN -MDR J. Pradeep -Decision Kandhasamy Tree J48 et al. (2015) -KNN(K=5) [9] -SVM -Random Forest Sisodia -Naïve Deepti et al. Bayes (2018) [12] -SVM -Decision Tree Xue-Hui -Logistic Meng et al. regression (2013) [13] -ANN -Decision Tree C5.0 Our study -KNN (without(K = 2) processing) -Decision Tree -ANN -DeepNN
Used database
Performance metric Accuracy Sensitivity/recall Specificity Precision (TPR*)
Pima Indian diabetes 89 84 88 86 83 UCI diabetes 73.82 datasets 73.17 73.34 71.74
87 83 90 88 87 59.7 53.73 53.84 53.81
N/A N/A N/A N/A N/A 81.4 83.6 73.39 80.4
88 85 87 85 82 N/A N/A N/A N/A
Pima Indian diabetes 76.30 65.10 73.82
76.3 65.1 73.8
N/A N/A N/A
75.9 42.4 73.5
standard 76.54 72.59 questionnaire from two communities in 76.97 Guangzhou at China
79.40 79.40 78.11
73.54 65.47 75.78
N/A N/A N/A
Frankfurt hospital, Germany
100 90.91 90.91 98.48
93.71 97.76 88.06 97.76
N/A N/A N/A N/A
95.5 95.5 89 98
*TPR: True Positive Rate
4.2
Discussion
In the present study, the recent literature was reviewed concerning applications of machine learning methods in Diabetes research. We conduct two experiments that aim to improve predicting diabetics patients using four machine learning algorithms KNN, Decision Tree, ANN and DeepNN. The first scenario consists of applying machine learning on dataset records that do contain missing values (without pre-processing). Here, the studies conclude that the DeepNN achieve higher accuracy of 98% than the other three Machine Learning algorithms. In the second scenario, (Dataset achieved with estimating missing values); we have a more accurate result when compared to the first study. We observe also that the accuracies of DeepNN are enhanced and achieve the best values among the other three algorithms it is about 99.5%. To push the research furthermore, and provide more evidence to demonstrate the prediction accuracy of our proposed models, we applied the model in a new diabetes dataset, it’s a Pima Indian Dataset under the same scenarios as before, the results in
468
O. Daanouni et al.
Table 8 shows that the decision tree get the best result without pre-processing, because decision tree makes no assumptions on relationships between features, inherently “throws away” the input features that it doesn’t find useful, whereas a neural net or deep neural net will use them all unless you do some feature selection as a preprocessing step. Table 10 shown the accuracy of Machine Learning algorithms enhanced after preprocessing. KNN and DeepNN achieve the highest accuracy of 97.53% and 96.35% respectively. In this case, KNN provides the best result because the classes are binomial and are quite separable, also KNN generate a highly convoluted decision boundary as it driving by the raw training data itself. Further, it can be seen from Table 12 that AUC value of KNN and DNN model are 0.9689 and 0.9625 respectively, such a high value of AUC indicates that both KNN and DNN are optimal classifiers to find that whether the patient is diabetic or not.
5 Conclusion and Perspectives One of the important real-world medical problems is the detection of diabetes at an early stage, in that purpose we proposed a support diagnostic system based on the comparison of four prediction algorithms models for predicting diabetes under two different scenarios. Experimental results determine the adequacy of the designed system and the result shows that the pre-processing phase is relevant in term of precision and reducing computational time furthermore deep neural network (DeepNN) prove his potential to get higher accuracy. This study may assist future researchers in choosing the subset features from features spaces using feature selection methods and using other machine learning technique such as data augmentation to enhance precision models.
References 1. Morrish, N.J., Wang, S.-L., Stevens, L.K., Fuller, J.H., Keen, H., and WMS Group: Mortality and causes of death in the WHO multinational study of vascular disease in diabetes. Diabetologia 44(2), S14 (2001) 2. World Health Organization: Global health risks: mortality and burden of disease attributable to selected major risks. World Health Organization, Geneva (2009) 3. Federal Bureau of Prisons Management of Diabetes Clinical Practice Guidelines, June 2012 4. Zou, Q., Qu, K., Ju, Y., Tang, H., Luo, Y., Yin, D.: Predicting diabetes mellitus with machine learning techniques. Front. Genet. 9, 515 (2018) 5. Ali, I.M.A.: Knowledge acquisition for an expert system for diabetic. In: Proceedings of the Mediterranean Symposium on Smart City Applications, pp. 747–758 (2017) 6. Sun, S.: A survey of multi-view machine learning. Neural Comput. Appl. 23(7–8), 2031– 2038 (2013) 7. Terrada, O., Cherradi, B., Raihani, A., Bouattane, O.: A fuzzy medical diagnostic support system for cardiovascular diseases diagnosis using risk factors. In: 2018 International Conference on Electronics, Control, Optimization and Computer Science (ICECOCS), pp. 1–6 (2018)
Type 2 Diabetes Mellitus Prediction Model
469
8. Kaur, H., Kumari, V.: Predictive modelling and analytics for diabetes using a machine learning approach. Appl. Comput. Inform. (2018). https://doi.org/10.1016/j.aci.2018.12.004 9. Kandhasamy, J.P., Balamurali, S.: Performance analysis of classifier models to predict diabetes mellitus. Procedia Comput. Sci. 47, 45–51 (2015) 10. Terrada, O., Raihani, A., Bouattane, O., Cherradi, B.: Fuzzy cardiovascular diagnosis system using clinical data. In: 2018 4th International Conference on Optimization and Applications (ICOA), pp. 1–4 (2018) 11. Terrada, O., Cherradi, B., Raihani, A., Bouattane, O.: Classification and prediction of atherosclerosis diseases using machine learning algorithms. In: 2019 5th International Conference on Optimization and Applications (ICOA), pp. 1–5 (2019) 12. Sisodia, D., Sisodia, D.S.: Prediction of diabetes using classification algorithms. Procedia Comput. Sci. 132, 1578–1585 (2018) 13. Meng, X.-H., Huang, Y.-X., Rao, D.-P., Zhang, Q., Liu, Q.: Comparison of three data mining models for predicting diabetes or prediabetes by risk factors. Kaohsiung J. Med. Sci. 29(2), 93–99 (2013) 14. https://www.kaggle.com/johndasilva/diabetes 15. https://www.kaggle.com/uciml/pima-indians-diabetes-database 16. Ravikumar, P., et al.: Prevalence and risk factors of diabetes in a community-based study in North India: the Chandigarh Urban Diabetes Study (CUDS). Diabetes Metab. 37(3), 216– 221 (2011) 17. Ware, L.B., et al.: Biomarkers of lung epithelial injury and inflammation distinguish severe sepsis patients with acute respiratory distress syndrome. Crit. Care 17(5), R253 (2013) 18. Rice, M.E., Harris, G.T.: Comparing effect sizes in follow-up studies: ROC area, Cohen’s d, and r. Law Hum Behav. 29(5), 615–620 (2005)
Hybrid Method for Breast Cancer Diagnosis Using Voting Technique and Three Classifiers Hajar Saoud1(&), Abderrahim Ghadi1, and Mohamed Ghailani2 1
LIST Laboratory, University of Abdelmalek Essaadi (UAE), Tangier, Morocco [email protected], [email protected] 2 LabTIC Laboratory, University of Abdelmalek Essaadi (UAE), Tangier, Morocco [email protected]
Abstract. Breast cancer is one of the most dangerous types of cancer in women sector; it infects one woman from eight during her life and one woman from thirty die and the rate keeps increasing. The early prediction of breast cancer can make a difference and reduce the rate of mortalities, but the process of diagnosis is difficult due to the varying types of breast cancer and due to its different symptoms. So, the proposition of decision-making solution to reduce the danger of this phenomenon has become a primordial need. Machine learning techniques have proved their performance in this domain. In previous work we tested the performance of several machine learning algorithms in the classification of breast cancer such as Bayesian Networks (BN), Support Vector Machine (SVM) and k Nearest Neighbor (KNN). In this work, we will combine those classifiers using the voting technique to produce better solution using Wisconsin breast cancer dataset and WEKA tool. Keywords: Breast cancer WEKA
Diagnosis Voting technique Classification
1 Introduction Breast cancer is a hard disease and its diagnosis is sometime difficult, the patient should pass through several tests starting with clinical examination ending with extracting and analyzing biological simples of breast cancer, the proposition of decision-making solution here has become a primordial need to reduce the process of diagnosis and also to reduce the rate of mortalities. In this paper, we tried to propose a solution for breast cancer diagnosis using machine learning due to their performance in the medical field. In previous work we tried to classify breast cancer using several classifiers such as Bayes Network (BN), Support Vector Machine (SVM), k-nearest neighbors algorithm (Knn), Artificial Neural Network (ANN), Decision Tree (C4.5) and Logistic Regression in [1] the higher accuracies are given by Bayes Network (BN) and Support Vector Machine (SVM) 97.28%. Then we tried to improve those accuracy in [2] by using the technique of feature selection Best First, the accuracy of Bayes Network (BN) has increased to 97.42% but the accuracy of Support Vector Machine (SVM) has decreased
© Springer Nature Switzerland AG 2020 M. Ben Ahmed et al. (Eds.): SCA 2019, LNITI, pp. 470–482, 2020. https://doi.org/10.1007/978-3-030-37629-1_34
Hybrid Method for Breast Cancer Diagnosis
471
to 95.279%. So, we should search for others solutions that can improve more the accuracy of classification of breast cancer. The objective of this work is to improve the accuracy of breast cancer classification using voting technique that aim to combine between classifiers. First, we did a combination between Bayes Network (BN) and Support Vector Machine (SVM) but there is no improvement. Consequently, we added K Nearest Neighbors algorithm (Knn) and the accuracy of classification has improved. The rest of this paper is structured as follows. Part two is a presentation of breast cancer. Part three gives a vision about similar research. Part four is a theoretic presentation of machine learning algorithms. Part five give the definition of voting technique. In part six we will explain our proposed approach. Part seven shows the experiments performed by WEKA software on Wisconsin breast cancer dataset and results of these experiments and finally conclusion and perspectives in part eight.
2 Breast Cancer Breast cancer can be defined as an abnormal production of cells in the breast that form in the form of cancerous masses, these masses are called tumors. Cancer cells can stay in the breast these types of cancer are called non-invasive, they lead to healing and do not produce metastatic cases. The other type of breast cancer is called invasive. These are dangerous type of cancers that can spread to the other organs of the body and can lead to metastatic cases. 2.1
Types of Breast Cancer
The types of breast cancers are invasive and non-invasive, Ductal Carcinoma In Situ is a non-invasive type the others are invasive [3] (Table 1): Table 1. Types of breast cancer. Type Ductal Carcinoma In Situ Ductal carcinoma Lobular carcinoma
Inflammatory carcinoma Paget’s disease Other carcinomas
Description Ductal Carcinoma In Situ is the most common type of breast cancer in the non-invasive cancer category in women. As the name suggests, it is formed inside the breast lactation channels This type of cancer is also formed in the lactation ducts, but cancer cells pass through the canal wall In this type of cancer the cancerous cells appear in the lobules grouped in the lobes. Afterwards, they cross the wall of the lobules and disperse in the surrounding tissues Is a rare type of cancer that is known by a breast that can turn red, swollen and hot Is also a rare type of cancer that is manifested by a small nipple wound that does not heal (medullary, colloidal or mucinous, tubular, papillary). Are the rarest types of breast cancer
472
2.2
H. Saoud et al.
Diagnosis of Breast Cancer
The process of the diagnosis of breast cancer is difficult due to the varying types and symptoms of breast cancer and also the patient should pass through several steps [4] starting with physical examination, it is a palpation of the breast that can determine the signs of appearance of cancer. The next step is medical imaging, it allows the detection of tumor masses and also it provides details on the clinical examination, there are several types of medical imaging among them: Mammography, Ultrasonography and MRI, The choice of one of these techniques is made according to the case of the patient. A diagnosis can only be decided after having studied biological samples at the microscopic level of the lesions that appeared in the medical imaging, the choice of the sampling method is according to the characteristics of the lesion, the exciting techniques are Aspiration or Cytological Puncture, Biopsy and Macrobiopsia (Fig. 1).
Fig. 1. Process of diagnosis of breast cancer.
The image obtained from the microscopic level will be studded at the same time with others images and features. So, the proposition of decision making solution will be an interesting thing to reduce the number of steps of the diagnosis also to avoid any error in the diagnosis. The machine learning techniques will be powerful tools due to their performance in the domain of medicine.
3 Related Works Several approaches are proposed in the domain of cancer diagnosis and also for others diseases using machine learning algorithms, voting technique and others techniques like bagging, stacking and boosting. In this paragraph we will cite same of them:
Hybrid Method for Breast Cancer Diagnosis
473
Khuriwal and Mishra in [5] they proposed an adaptive ensemble voting method using Artificial Neural Network (ANN) and Logistic Regression (LR), the database used is Wisconsin Breast Cancer database. They achieved 98.50% in accuracy. Kumar et al. in [6] they compared the performance of machine learning techniques in the classification of breast cancer using Wisconsin Breast Cancer database then they combined those techniques using voting technique. The three techniques tested in this research are Naïve Bayes, SVM and j48. Latha and Jeeva in [7] they examined the ensemble algorithms bagging, boosting, stacking and majority voting for prediction of the heart disease using Cleveland heart dataset from the UCI machine learning repository. Leon et al. in [8] they analyzed the influence of several voting methods on the performance of K Nearest Neighbor and Naïve Bayes algorithms used for datasets with different levels of difficulty. Rishika and Sowjanya in [9] they aim to compare the performance of Decision Tree, Neural Network and Naive Bayes, then they tried to combine between them using stacking approach. Sri Bala and Rajya Lakshmi in [10] they implemented four models Adaboosting, bagging and stacking or blending on preliminary classifiers to improve the accuracy of the classification of breast cancer. So, the totals of built models are 12.
4 Machine Learning Algorithms The machine learning techniques that we will see in this paper are Bayesian Network (BN), Support Vector Machines (SVM) and k-nearest neighbors algorithm (Knn). We will examine each algorithm separately than we will combine between them to improve the accuracy of classification of the breast cancer. 4.1
Bayesian Network (BN)
Bayesian Network [11], also called (Bayesian belief network), is directed acyclic graph (DAG) composed of nodes and edges, the nodes represent variables and edges represent the probabilistic dependencies between those variables. Bayesian Network combines principles of statistics, graph theory, probability theory and computer science. 4.2
Support Vector Machines (SVM)
Support Vector Machines is supervised learning model, which is always known by the notion of hyperplane, this hyperplane is a line that divide a plan into two spaces each space represent a class. Taking training data the Support Vector Machines well search an optimal hyperplane that will separate the data into two dimensional spaces.
474
4.3
H. Saoud et al.
K Nearest Neighbors Algorithm (KNN)
The k-nearest neighbors classifier is a supervised machine learning algorithm that can be used in both classification and regression. The k-nearest neighbors classifier capture the idea of similarity (called also distance). So, the principle of k-nearest neighbors that it calculates the distance between a given test tuple and others tuples to search the K closest tuples, these tuples are named (k nearest neighbors).
5 Voting Classifier Technique Voting classifiers is a technique used in classification; it aims to combine between classifiers to improve the accuracy of classification. The principle of voting technique that each machine learning technique gives classification or output then the vote of those outputs will be taken as classification (Fig. 2).
Fig. 2. Voting classifiers technique.
If we take the example of 3 classifiers C1, C2 and C3 the prediction of each classifier successively will be P1, P2 and P3. The final prediction will be: PF = mode {P1, P2, P3}.
6 Proposed Method In our proposed method we will improve the accuracy of the classification of the three machine learning algorithms Bayes Network (BN) and Support Vector Machine (SVM) and K Nearest Neighbors algorithm (KNN) by using the voting technique, that aim to combine between them to improve the accuracy of classification. Figure 3
Hybrid Method for Breast Cancer Diagnosis
475
represents the process of the proposed method first we choose Wisconsin breast cancer dataset, then we did the pre-processing of data to eliminate missing data and finally we passed to the step of classification.
Fig. 3. Process of proposed method.
7 Experimentation and Results 7.1
Description of the Dataset
The database that we used in this research is the Wisconsin breast cancer dataset available in UCI machine learning repository [12]. It contains 699 records (458 benign tumors and 241 malignant tumors). It is composed of 11 variables 10 predictor variables and one result variable that shows whether the tumor is benign or malignant. The predictive attributes vary between 0 and 10. The value 0 corresponds to the normal state and the value 10 corresponds to the most abnormal state. The table above presents the description of the 11 attributes of the Wisconsin breast cancer dataset (Table 2):
476
H. Saoud et al. Table 2. Attributes of WBCD.
Attributes Id Clump thickness Uniformity of cell size Uniformity of cell shape Marginal adhesion Single epithelial cell size Bare nuclei Bland chromatin Normal nucleoli Mitoses Class
7.2
Description A code for the identification of each line The benign cells are grouped in monolayers whereas the cancer cells are grouped in multilayers The size of the cancer cells The shape of the cancer cells Cancer cells can lose their tights; this is a sign of malignant cancer Single Epithelial Cell Size The nuclei are not surrounded by the rest of the cell in benign tumors Cancer cells have coarse chromatin In cancer cells, the nucleoli are transforming into protuberant, but the nucleoli are small Cell growth If the cancer is a benign tumor or malignant tumor
WEKA Tool
The tool that we used to apply the machine learning algorithms on the breast cancer database is WEKA [13], because WEKA is a collection of open source machine learning algorithms, which allows realizing the tasks of data mining to solve real world problems. It contains tools for data preprocessing, classification, regression, grouping, and association rules. Also it offers an environment to develop new models. 7.3
K-Fold Cross-validation
To evaluate the performance of machine learning algorithms based on breast cancer data we used the K-fold cross validation test method. This method aims to divide the database in two sets, the training data to run the model and the testing data to evaluate the performance of the model. This is the most used method in the evaluation of machine learning techniques (Fig. 4).
Fig. 4. Process of K-fold Cross-validation.
Hybrid Method for Breast Cancer Diagnosis
7.4
477
Confusion Matrix
Confusion matrix gives the possibility to evaluate the performance of each classifier by calculating its Accuracy, Sensitivity and Specificity. It contains information about real classifications or (current) and predicted (Table 3): Table 3. Confusion matrix. Predicted benign Predicted malignant Actual benign TP (true positives) FN (false negatives) Actual malignant FP (false positives) TN (true negatives)
TP: the cases predicted as benign tumors, they are in fact benign tumors. TN: the cases predicted as malignant tumors, they are in fact malignant tumors. FP: the cases predicted as benign tumors but in the reality they are malignant tumors. FN: the cases predicted as malignant tumors but in the reality they are benign tumors. From the confusion matrix we can calculate: TP + TN TP + FP + TN + FN TP • Sensitivity ¼ TP þ FN TN • Specificity ¼ TN þ FP • accuracy ¼
7.5
Bayesian Network (BN)
The accuracy obtained by Bayesian Network (BN) is 97.28%, 680 from 699 are well classified instances and 19 are incorrectly classified instance that represent the 2.71%. Table 4 represents the confusion matrix of Bayesian Network (Figs. 5, 6 and Table 5): Table 4. Confusion matrix of BN. Predicted benign Predicted malignant Actual benign 442 16 Actual malignant 3 238
Table 5. Results of BN. TP rate FP rate Precision Recall F-measure ROC area benign 0,965 0,012 0,993 0,965 0,979 0,991 malignant 0,988 0,035 0,937 0,988 0,962 0,991
478
H. Saoud et al.
Fig. 5. ROC curve of benign for BN.
7.6
Fig. 6. ROC curve of malignant for BN.
Support Vector Machines (SVM)
The accuracy obtained by Support Vector Machines (SVM) is 97.28% using the Puk as kernel function, 680 from 699 are well classified instances and 19 are incorrectly classified instance that represent the 2.71%, the same results as Bayesian Network (BN). Table 6 represents the confusion matrix of Support Vector Machines (SVM) (Figs. 7, 8 and Table 7): Table 6. Confusion matrix of SVM. Predicted benign Predicted malignant Actual benign 442 16 Actual malignant 3 238
Table 7. Results of SVM. TP rate FP rate Precision Recall F-measure ROC area Benign 0,967 0,017 0,991 0,967 0,979 0,975 Malignant 0,983 0,033 0,940 0,983 0,961 0,975
Fig. 7. ROC curve of benign for SVM.
Fig. 8. ROC curve of malignant for SVM.
Hybrid Method for Breast Cancer Diagnosis
7.7
479
BN-SM
The accuracy obtained by BN-SVM is 96.99% there is no improvement, 678 from 699 are well classified instances and 21 are incorrectly classified instance that represent the 3%. Table 8 represents the confusion matrix of BN-SVM (Figs. 9, 10 and Table 9): Table 8. Confusion matrix of BN-SVM. Predicted benign Predicted malignant Actual benign 445 13 Actual malignant 20 221
Table 9. Results of BN-SVM. TP rate FP rate Precision Recall F-measure ROC area Benign 0,974 0,037 0,980 0,974 0,977 0,991 Malignant 0,963 0,026 0,951 0,963 0,957 0,991
Fig. 9. ROC curve of benign for BN-SVN.
7.8
Fig. 10. ROC curve of malignant for BN-SVM.
K Nearest Neighbors Algorithm (KNN)
The accuracy obtained by k-nearest neighbors algorithm (Knn) is 95.27%, 666 from 699 are well classified instances and 30 are incorrectly classified instance that represent the 4.72%. Table 10 represents the confusion matrix of k-nearest neighbors (KNN) (Figs. 11, 12 and Table 11): Table 10. Confusion matrix of KNN. Predicted benign Predicted malignant Actual benign 445 13 Actual malignant 20 221
480
H. Saoud et al. Table 11. Results of KNN. TP rate FP rate Precision Recall F-measure ROC area Benign 0,972 0,083 0,957 0,972 0,964 0,974 Malignant 0,917 0,028 0,944 0,917 0,931 0,974
Fig. 11. ROC curve of benign for KNN.
7.9
Fig. 12. ROC curve of malignant for KNN.
BN-SVM-KNN
The accuracy obtained by the proposed combination of the three algorithms by voting techniques is 97.56%, 682 from 699 are well classified instances and 17 are incorrectly classified instance that represent the 2.43%. Table 12 represents the confusion matrix of BN-SVM-KNN (Figs 13, 14 and Table 13): Table 12. Confusion matrix of BN-SVM-KNN. Predicted benign Predicted malignant Actual benign 445 13 Actual malignant 4 237
Table 13. Results of BN-SVM-KNN. TP rate FP rate Precision Recall F-measure ROC area Benign 0,972 0,017 0,991 0,972 0,981 0,990 Malignant 0,983 0,028 0,948 0,983 0,965 0,990
Hybrid Method for Breast Cancer Diagnosis
Fig. 13. ROC curve of benign for BN-SVNKNN.
481
Fig. 14. ROC curve of malignant for BNSVM-KNN.
Table 14 resumes the obtained results by each algorithm: Table 14. Results of all models. BN SVM BN-SVM KNN BN-SVM-KNN
Accuracy 97.28% 97.28% 96.99% 95.27% 97.56%,
Well classified instance Wrong classified instance Time taken 680 19 0.02 s 680 19 0.22 s 678 21 0.09 s 666 30 0s 682 17 0.03 s
8 Conclusion To conclude, in this paper we tried to classify breast cancer into its two types benign or malignant using machine learning algorithm and the voting technique. First we examined each algorithm, Bayes Network (BN), Support Vector Machine (SVM) and k-nearest neighbors algorithm (KNN) separately then we tried to combine between them to improve the accuracy of the classification of breast cancer using the voting technique the accuracy produced 97.56%. The database of breast cancer in which the algorithms are tested is Wisconsin breast cancer dataset available in UCI machine learning repository using the WEKA tool. Acknowledgement. H. Saoud acknowledges financial support for this research from the “Centre National pour la Recherche Scientifique et Technique” CNRST, Morocco.
References 1. Saoud, H., et al.: Application of data mining classification algorithms for breast cancer diagnosis. In: Proceedings of the 3rd International Conference on Smart City Applications SCA 2018, pp. 1–7. ACM Press, Tetouan (2018). https://doi.org/10.1145/3286606.3286861
482
H. Saoud et al.
2. Saoud, H., et al.: Using feature selection techniques to improve the accuracy of breast cancer classification. In: Ben Ahmed, M., et al. (ed.) Innovations in Smart Cities Applications, edn. 2. pp. 307–315. Springer International Publishing, Cham (2019). https://doi.org/10.1007/ 978-3-030-11196-0_28 3. Le cancer du sein. https://www.passeportsante.net/fr/Maux/Problemes/Fiche.aspx?doc=can cer_sein_pm. Accessed 19 Oct 2019 4. Le diagnostic. https://rubanrose.org/cancer-du-sein/depistage-diagnostics/diagnostic. Accessed 19 Oct 2019 5. Khuriwal, N., Mishra, N.: Breast cancer diagnosis using adaptive voting ensemble machine learning algorithm. In: 2018 IEEMA Engineer Infinite Conference (eTechNxT), pp. 1–5. IEEE, New Delhi (2018). https://doi.org/10.1109/ETECHNXT.2018.8385355 6. Kumar, U.K., et al.: Prediction of breast cancer using voting classifier technique. In: 2017 IEEE International Conference on Smart Technologies and Management for Computing, Communication, Controls, Energy and Materials (ICSTM), pp. 108–114. IEEE, Chennai (2017). https://doi.org/10.1109/ICSTM.2017.8089135 7. Latha, C.B.C., Jeeva, S.C.: Improving the accuracy of prediction of heart disease risk based on ensemble classification techniques. Inform. Med. Unlocked 16, 100203 (2019). https:// doi.org/10.1016/j.imu.2019.100203 8. Leon, F., et al.: Evaluating the effect of voting methods on ensemble-based classification. In: 2017 IEEE International Conference on INnovations in Intelligent SysTems and Applications (INISTA), pp. 1–6. IEEE, Gdynia (2017). https://doi.org/10.1109/INISTA.2017. 8001122 9. Rishika, V., Sowjanya, A.M.: Prediction of breast cancer using stacking ensemble approach 11 10. Int. J. Adv. Res. Comput. Sci. Softw. Eng 11. Mahmood, A.: Structure learning of causal bayesian networks: a survey 6 12. UCI Machine Learning Repository: Breast Cancer Wisconsin (Original) Data Set. https:// archive.ics.uci.edu/ml/datasets/breast+cancer+wisconsin+(original). Accessed 19 Oct 2019 13. Weka 3 - Data Mining with Open Source Machine Learning Software in Java. https://www. cs.waikato.ac.nz/ml/weka/. Accessed 19 Oct 2019 14. Saoud, H., Ghadi, A., Ghailani, M.: Analysis of evolutionary trends of incidence and mortality by cancers. In: Ben Ahmed, M., Boudhir, A. (eds.) Innovations in Smart Cities and Applications. SCAMS 2017. Lecture Notes in Networks and Systems, vol. 37. Springer, Cham (2018)
Visual Question Answering System for Identifying Medical Images Attributes Afrae Bghiel(&), Yousra Dahdouh(&), Imane Allaouzi(&), Mohamed Ben Ahmed(&), and Abdelhakim Anouar Boudhir(&) Abdelmalek Essaâdi University, Faculty of Sciences and Techniques, Tangier, Morocco [email protected], [email protected], [email protected], [email protected], [email protected]
Abstract. In this paper, we propose a new Visual Question Answering (VQA) System which is able to give an answer in natural language about an image and a natural language question. In our System, we have employed a Convolutional Neural Network (CNN) for visual input processing and Recurrent Neural Network (RNN) in Encoder-Decoder model which consists of an LSTM to encode sequential input which is an assembling between image and question vectors, and another LSTM to decode the states for predicting target answers in output, this one will be generated in natural language using greedy search algorithm. Keywords: Computer vision CNN RNN Natural language Encoder-Decoder LSTM Word embedding Word2vect Visual Question Answering Greedy search
1 Introduction In the last few years, the number of published papers related to Deep Learning have exploded. Both the academic world and the industry are pushing forward to speed up the developments and the research in this area [1]. The reason is that deep learning has shown a great performance solving a lot of problems that were previously tackled by more classic machine learning algorithms and it has also opened the door to more complex tasks such as image recognition [2, 3], machine translation [4], image captioning [5], and visual question answering [6, 7]. For medical images, innovations from AI, deep learning and big data development should enable them to continue integration into the clinical care, to be more visible and available to prescribing doctors and patients [8]. where medicine hasn’t based anymore on the single doctor-patient relationship because patients now have access to health data through gateways, and that’s involves the help for understanding conditions concerning those health data. Thus, we have treated visual question answering problem using medical images, in the context of deep learning, computer vision and natural language, where we have used pre-trained CNN for images features extraction, pre-trained word2vector to extract © Springer Nature Switzerland AG 2020 M. Ben Ahmed et al. (Eds.): SCA 2019, LNITI, pp. 483–492, 2020. https://doi.org/10.1007/978-3-030-37629-1_35
484
A. Bghiel et al.
features from preprocessed textual inputs, and RNN of LSTM type, as encoder and decoder for sequential data in input, in the goal to obtain probabilities distributions in output as answer. Finally, we have used greedy search algorithm for answer generation based on this distribution. The contributions of this paper are summarized as follows. Section 2, presents the most relevant work. Section 3 gives a detailed description of the proposed system. Validation results for our model and final test results are presented in Sect. 4. Finally, the paper is concluded in Sect. 5.
2 Related Works Visual Question Answering is a new problem, sophisticated algorithms for VQA are already being deployed. Most existing papers on VQA have used long-short-termmemory (LSTM) neural networks [9]. [10–13] all used LSTM networks to encode the question and combined the question encoding with image features from a deep convolutional neural networks (CNN). In [10], separate LSTMs were used for the question and answer, but they had a shared word embedding layer, and CNN image features are fused at the end of the LSTM. Their model was able to output more than one-word answers or lists, and it could generate coherent sentences. And in [14] Allaouzi and all dealt the visual question answering problem in medical domain with the task as a multilabel classification problem. They extracted image features using VGG16 and word embedding of the question and feed it to a BiLSTM network to extract question features. Then concatenated question features and image features and fed them to a decision tree classifier. Zhou and Kang and all propose in [15] to use some image enhancement methods like clipping and questions preprocessing methods like lemmatization; they have used Inception-Resnetv2 model (CNN) to extract image features, and have used Bi-LSTM model (RNN) to encode the questions. Finally, they have concatenated the coded questions with the image features to generate the answers. In [24], Authors have explored transfer learning on image channel to extract meaningful features from the medical images, where they present a novel approach of utilizing Embedding based topic modeling for transfer learning. Second, they implemented co-attention mechanism integrated with Multi-modal Factorized Bilinear Pooling (MFB) and Multi-modal Factorized High-order Pooling (MFH) to medical VQA. The system proposed in this paper is shows superior performance for handling answers, Encoder Decoder approach to merge the visual and textual information for answering prediction. Where we propose to use pre-trained CNN as Resnet50 to gain time and extract image’s features; and for textual inputs we parse and encode every sentences first, then we pass the result to pre-trained word2vec to obtain vocabulary’s words vectors. Thus, we combine between sentence’s matrix and repeated image’s vector to have an input to the encoder as matrix which is an LSTM, then the last hidden state and cell state redirected to another LSTM which is a decoder to obtain answer’s probabilities results; and finally we generate answers with greedy search algorithm.
Visual Question Answering System for Identifying Medical Images Attributes
485
3 The Proposed Model Visual Representation
Textual Representation
Image
Question
Preprocessing
ResNet50
Padding
Extracting FC 1000[29]
Word Embedding (Word2Vec)
Flatten & Repeat Multiply
Encoder
Decoder
LSTM Last hidden & cell state
LSTM Probabilities about vocabulary words for answer generation
Answer
Fig. 1. VQA model architecture.
Our proposed model for VQA consists of two Artificial Intelligence branches, computer vision for visual input processing, and Natural Language Processing for textual input processing. To extract features from textual input or inputted question, we have pre-processed this one; first, we have gone through building our own vocabulary from
486
A. Bghiel et al.
questions and answers tokens, then we have encoded every pre-processed question and answer and we have padded them to the same length. The Padded questions/answers have passed to pre-trained embedding model -word2vect- for obtaining every word’s vector. We have chosen Word2Vec because it includes words vectors for a vocabulary of 3 million words and phrases they have formed on about 100 billion words from a Google News database. So, technically word2vec will also include medical words. For extracting prominent features from medical images, we have used the pre-trained Resnet50 network that is trained on more than a million images from the ImageNet database. The network is 50 layers deep and can classify images into 1000 object categories. In our system we extract the fully connected layer of this network to obtain an output vector of 1000 elements, the vector of the 1000 elements will pass to Flatten, and it is necessary to repeat this vector til question max length. Thus; we combine between image’s and question’s matrices, first, we have repeat image vector to have a matrix of the same length size as the input question to know the question is about which images area; so we have decided to use multiplication for assembling image model and question model; and this operation needs matrices of the same size; and like we have matrices of the same length but different heights; we have used commonly a Greatest Common Divisor (GCD) to reshape matrices to the same height. and multiplying the matrices to have a result passed on to Encoder Model which is an LSTM, and a Decoder Model which is also an LSTM to have finally probabilities distribution on all vocabulary words; and we generate answer using greedy search basing on this distribution (Fig. 1). 3.1
Image Representation
It has already been observed that features extracted from natural image using pretrained CNN, in our case, to extract the important features of medical images, we have used the pre-trained network Resnet50, it won 1st place in the ILSVRC 2015 classification competition with top-5 error rate of 3.57% [2].
Image
Resnet50 (Last Layer)
1000 Fig. 2. Image features extraction.
Visual Question Answering System for Identifying Medical Images Attributes
487
ResNet-50 is a convolutional neural network that is trained on more than a million images from the ImageNet database [16]. The network is 50 layers deep and can classify images into 1000 object categories. The network has an image input size of 224-by-224. The ResNet-50 model consists of 5 stages each with a convolution and Identity block. Each convolution block has 3 convolution layers and each identity block also has 3 convolution layers; is a powerful backbone model that is used very frequently in many computer vision tasks ResNet50 uses skip connection to add the output from an earlier layer to a later layer. This helps it mitigate the vanishing gradient problem [17]. We extract the fully connected layer of this network with its Softmax activation function to obtain an output vector of 1000 elements, the vector of the 1000 elements will pass to Flatten, and it is necessary to repeat this vector til question’s or answer’s max length (Fig. 2). 3.2
Question Representation
We have pre-processed the textual inputs; first we have gone through building our own vocabulary due to we have medical words, then we have encoded every question and answer pre-processed and we have padded them to the same length. Padded questions and padded answers have passed to pre-trained embedding model -word2vect- for obtaining every word’s vector of 300 elements (Fig. 3).
Question
Preprocessing
Word2Vect SkipGram
Encoding
300 18
Fig. 3. Question representation.
We have used Word2Vec [18] (Skip-gram) which is presented as one of the most advanced and important in Natural Language Processing (NLP) ‘s field. It includes words vectors for a vocabulary of 3 million words and phrases they have formed on about 100 billion words from a Google News database. 3.3
Merge
Multiplication can be used to combine two models, Not only it decreases the training time but it can also increase the robustness and the performance of the model; But there is a logic behind assembling or combining where this method requires two matrices of the same size to be inputted, so we must first go through the step of resizing each matrix by the Dense function using the Greatest Common Divisor (GCD), the size of
488
A. Bghiel et al.
each matrix must be (Question length, GCD (embed size (300), 1000)), and then we can use the function Multiply () according to Keras. Layers. Merge (Fig. 4).
Image Matrix
Question Matrix
Dense (QMaxlength,GCD(Emb_size,1000))
Dense (QMaxlength,GCD(Emb_size,1000))
Multiplication Fig. 4. Merging with multiplication.
3.4
Encoder Decoder Model
The output of the question model is multiplied by the output of the image model. The idea of using a repetition block for the image characteristics is to make the model receive the resulting vector of this fusion containing the image’s and the question’s information as a vector of the image and sequence. We then pass this merged output as input to LSTM, which is a particular type of RNN that can learn long-term dependencies, the LSTM model has three multiplicative units, input gate, output gate and forget gate. The input gate is used to memorize some information of the present, the output gate is used for output, and the forget gate is used to choose to forget some information of the past [19] (Fig. 5).
Fig. 5. Encoder-decoder architecture [25].
Visual Question Answering System for Identifying Medical Images Attributes
489
LSTMs are used for sequence modeling and long term dependency capture. We have configured the model to produce only the last hidden state and the cell state. The last ones will encode all the information in the sequence of question with the image information.
Fig. 6. LSTM unit [26].
The output of the encoder model corresponding to the last hidden state and cell state of the LSTM is transmitted as an initial state to another LSTM which is decoder. During training, the outputs of the LSTM decoder are probability distributions (on all vocabulary words) generated by the model for the next word of the sentence (Fig. 6). 3.5
Answer Generation
A greedy algorithm is a simple, intuitive algorithm that is used in optimization problems. The algorithm makes the optimal choice at each step as it attempts to find the overall optimal way to solve the entire problem. A greedy algorithm for an optimization problem always makes the choice that looks best at the moment and adds it to the current sub solution [20]. Examples already seen are Dijkstra’s shortest path algorithm and Prim/Kruskal’s MST algorithms. Steps: • Define a heuristic function h(x) to estimate the distance from the state to the objective. • From the current state, determine the search space (actions) for a step forward. • Select the action in the search space that minimizes the heuristic function. hðxÞ h ðxÞ
ð1Þ
490
A. Bghiel et al.
h(x): Estimate. h*(x): Actual. In our case, we have used greedy search algorithm for answer generation using the predicted probabilities distributions by trained visual question answering model (Decoder Output).
4 Results and Discussion 4.1
Evaluation and Results
The performance evaluation of the VQA System is a very important task, where we have used in this section two different metrics, accuracy and bleu score: BLEU Score is a metric used to evaluate a sentence generated in a sentence of reference, proposed by Kishore Papineni, et al. in 2002. A perfect match gives a score of 1.0, whereas a perfect discordance gives a score of 0.0. The Blue Score allows us to automatically calculate a score that measures the resulting phrase quality [21]. Accuracy is defined as the number of times that the model predicts the correct answer for inputs. We trained Our model conducting several experiments for different optimizers such RMSprop [22], Adam [22], and RAdam [23]; learning rates and other hyper parameters; where Our results are summarized in Table 1. Table 1. Results of our proposed model. Method Optimizer Blue score Multiplication Adam 0.031 RMSProp 0.15 RAdam 0.16 RMSProp 0.20 RAdam 0.24 RAdam 0.30 RMSProp 0.32
The best result of bleu score got accuracy of 28%. 4.2
Discussion
VQA is a relatively new field that requires a thorough understanding of images and text. Looking at the current research scenario in the field of deep learning, it can be expected that VQA systems will definitely improve over time, both in terms of optimality and accuracy. In deep learning has already shown an exponential increase in the performance of various computer vision and language models. Using these individual components to create a system that combines them will greatly improve the results obtained for tasks such as VQA.
Visual Question Answering System for Identifying Medical Images Attributes
491
Taking into account all existing systems, discussions are ongoing on several issues related to training techniques and model evaluation. As we used Encoder-Decoder approach where we have an LSTM Encoder which is usually used to encode a sequence, in our case the sequence is the fusion of the two vectors, image’s vector and question’s vector, the last state of cells and the last hidden layer are directed as the initial state of another LSTM that is the Decoder, in order to generate a probability distribution. There are other communes like the one that is based on the classification where the proposed approach considers the task of VQA as multi-label classification problem, where each label corresponds to a unique word in the answer dictionary that was built from the training set [14].
5 Conclusion This paper focuses on a deep learning based model for visual question answering in medical field, where we argue the need of vision research in medical characterization due to various research directions in the medical imaging applications that have not been fully explored, and can be solved by combining vision and language processing. This research aims to develop artificial intelligence algorithms that jointly reason over medical images and accompanying clinical text. The proposed research is fruitful in advancing healthcare by providing the importance of combining different models results to understand the question concern which image’s area exactly for answer generation. It shows how LSTM magnifies the ability of predicting answer’s probabilities distributions as Encoder and Decoder. Our research describes how our model is able to generate the best probable answer where we finally catch a best BLEU Score result of 0.32 and accuracy of 28%.
References 1. Issey Masuda Mora: Open-ended visual question-answering. arXiv:1610.02692v1 [cs.CL] 9 October 2016 2. He, K., Zhang, X., Ren, S., Sun, J.: Deep residual learning for image recognition. arXiv: 1512.03385v1 [cs.CV] 10 December 2015 3. Krizhevsky, A., Sutskever, I., Hinton, G.E.: ImageNet classification with deep convolutional neural networks. In: NIPS (2012) 4. Sutskever, I., Vinyals, O., Le, Q.V.: Sequence to sequence learning with neural networks. In: the 27th International Conference on Neural Information Processing Systems, vol. 2, pp. 3104–3112 (2014) 5. Vinyals, O., Toshev, A., Bengio, S., Erhan, D.: Show and tell: a neural image caption generator. In: CVPR (2015) 6. Yang, Z., He, X., Gao, J., Deng, L., Smola, A.: Stacked attention networks for image question answering. In: CVPR (2016) 7. Zhou, B., Tian, Y., Sukhbaatar, S., Szlam, A., Fergus, R.: Simple baseline for visual question answering. arXiv preprint arXiv:1512.02167 (2015)
492
A. Bghiel et al.
8. http://www.sfrnet.org/rc/org/sfrnet/htm/Article/2005/mie-20050128-00000008227/src/htm_ fullText/fr/Articles%20AFIB%20RSNA%202018.pdf 9. Kafle, K., Kanan, C.: Answer-type prediction for visual question answering. IEEE Xplore (2016) 10. Gao, H., Mao, J., Zhou, J., Huang, Z., Wang, L., Xu, W.: Are you talking to a machine? Dataset and methods for multilingual image question answering. In: NIPS (2015) 11. Malinowski, M., Rohrbach, M., Fritz, M.: Ask your neurons: a neural-based approach to answering questions about images. In: ICCV (2015) 12. Antol, S., Agrawal, A., Lu, J., Mitchell, M., Batra, D., Zitnick, C.L., Parikh, D.: VQA: visual question answering. In: ICCV (2015) 13. Ren, M., Kiros, R., Zemel, R.: Exploring models and data for image question answering. In: NIPS (2015) 14. Allaouzi, I., Benamrou, B., Benamrou, M., Ahmed, M.B.: Deep neural networks and decision tree classifier for visual question answering in the medical domain (2018) 15. Zhou, Y., Kang, X., Ren, F.: Employing inception-Resnet-v2 and Bi-LSTM for medical domain visual question answering (2018) 16. ImageNet. http://www.image-net.org 17. https://towardsdatascience.com/understanding-and-coding-a-resnet-in-keras446d7ff84d33 18. https://skymind.ai/wiki/word2vec 19. Xiao, Y., Yin, Y.: Hybrid LSTM neural network for short-term traffic flow prediction, 8 March 2019 20. https://home.cse.ust.hk/*dekai/271/notes/L14/L14.pdf 21. Papineni, K., Roukos, S., Ward, T., Zhu, W.-J.: Bleu: a method for automatic evaluation of machine translation, 17 September 2001 22. Ruder, S.: An overview of gradient descent optimization algorithms. arXiv:1609.04747v2 [cs. LG] 15 June 2017 23. https://medium.com/@lessw/new-state-ofthe-art-ai-optimizer-rectified-adamradam5d854730807b 24. Peng, Y., Liu, F., Rosen, M.P.: UMass at ImageCLEF medical visual question answering (Med-VQA) 2018 Task (2018) 25. https://towardsdatascience.com/time-series-forecasting-with-deep-stacked-unidirectionaland-bidirectional-lstms-de7c099bd918 26. https://ai.stackexchange.com/questions/6961/structure-of-lstm-rnns
New Generation of Networks and Systems for Smart Cities
Enhanced Mobile Network Stability Using Average Spatial Dependency Halim Berradi(B) , Ahmed Habbani, Chaimae Benjbara, Nada Mouchfiq, and Mohammed Souidi ENSIAS, Mohammed V University in Rabat, Rabat, Morocco [email protected]
Abstract. The mobility impact is a crucial element that can impact the stability of network performances in Mobile Ad hoc Networks (MANETs). Choosing a path with stability, security with an extended network lifetime is one of the critical points in routing protocols design. In this paper, the author suggests an enhanced algorithm based on a metric of mobility by discovering the modification of MPRs selection and incorporating the metric in the routing decision. The metric based on the spatial mobility of neighbor nodes named average spacial dependancy. The principal objective is to discover more performing MPRs that can improve the network performances even in a critical situation of the dynamic environment. The authors implement the proposed metric in OLSR protocols and get the results of the performance using a network simulator (NS3) under the Manhattan Grid mobility model. The obtained results prove important performance gains for the modified version. Besides, the proposed metric can be used as a new mechanism to increase network performances for protocols in MANETs. Keywords: IoT · MANET · Mobility environment · Manet · OLSR
1
· Spatial dependency · Smart
Introduction
The Mobile Ad Hoc Networks (MANETs) classified as a group of mobile wireless nodes communicating on popular wireless networks. It is a self-configured network of mobile nodes linked by wireless links. These wireless devices are mobile and continuously altering positions and do not have any stationary infrastructure. Therefore, the Mobile Ad hoc Networks can be reflected as an infrastructure-less-network where any node can transmit packets to another node deprived of using any base stations. In MANETs, each mobile node has a range of transmission within which signals produced are strong to permit other nodes to exchange the relevant data. Thus, Mobile nodes can communicate directly when they are within the transmission range of each other. The regular communication between two mobile nodes far away from each other is based on multi-hop communication, and this c Springer Nature Switzerland AG 2020 M. Ben Ahmed et al. (Eds.): SCA 2019, LNITI, pp. 495–509, 2020. https://doi.org/10.1007/978-3-030-37629-1_36
496
H. Berradi et al.
is the motivation behind the name of multi-hop wireless for Mobile Ad hoc networks. MANETs are greatly appropriate for practices correlated to distinctive outdoor events, communications in areas with no wireless substructure, tragedies or natural disasters, and military tactics. Hence, routing is one of the crucial challenges when mobile devices move freely in any direction, and links break occur, thus impact the quality of service performances. Consequently, MANETs offer numerous challenges that must be considered with intelligence, such as the topology discovery, mobility awareness of the node neighbors, the routing communication, the packets lost, the energy limitations, and the quality of service. Various routing protocols have been suggested for MANETs. These protocols can be classified into three diverse collections: proactive, reactive, and hybrid. In proactive routing protocols such as Destination- Sequenced DistanceVector (DSDV) [1] and Optimized Link State Routing Protocol (OLSR) [2], the establishment of the path to all target or parts of the network are defined at the start-up and preserved using constant path update control messages. In reactive protocols such as Ad hoc On-Demand Distance Vector (AODV) [3] and Dynamic Source Routing (DSR) [4], the routes established when the source requires them based on a route detection procedure. Hybrid routing protocols joined the basic properties of proactive and reactive protocols into one. Conversely, MANETs are involved generally in mobile nodes, which make routing more difficult as it must be completed in a mobility-adaptive and realtime. In wireless networks, OLSR protocol is an essential routing protocol since it is considered to be the primary routing protocol support Multipoint Relays (MPRs) [5,6] mechanism. This proposal improved the performance network based on MPRs nodes that can enhance the capacity of the network, considering its functionality [7,8]. The MPR considered the cluster-head of numerous neighbor nodes. Hence, our proposal tries to quantify the mobility neighbors’ value and use it in the MPRs election process. This paper proposes a characteristic function implemented in the MPRs Selection algorithm for OLSR protocol to offer an optimum process for mobile nodes to select stable MPRs. As projected in the simulation results, the metric is adaptive to a high and dense environment. Finally, the paper offers a linear distance-based spatial dependency as a metric for MPRs Selection named Average spacial dependency. The node mobility measured using the node GPS periodically. This measure will be the key to our proposal to calculated our characteristic function MPR. Hence, proactive protocols select the stables MPRs nodes, that impact the selection of optimal paths with a minimum delay, lost packets, to have more stability and durability in the network. The inspiration in our research is to modify and to enhance the MPRs selection in OLSR based on a characteristic metric for more performance in the network [9–11]. The results showed that the proposed algorithm improves the performances of the network, such as Packet loss ratio, delay, and throughput [12,13].
Enhanced Mobile Network Stability Using Average Spatial Dependency
497
The influence of these adaptations on the network performance under the Manhattan Grid Model has been evaluated using the NS-3 simulator. The rest of the paper prepared as follows. In Sect. 2, we have revised some related works in the literature related to improving MPRs selection. Section 4 explains the method adopted to define the metric and highlighted how this metric integrated into OLSR protocol. Simulation results assumed in Sect. 5. Lastly, Sect. 6 concludes this paper.
2
Related Work
Studies have shown the problems regarding OLSR routing by relating it with the MPR selection. Route performance can be enhanced by adding or amending some sort of criteria in the MPR mechanism, for example, by introducing the routing metrics. Selection of MPR plays a vital role regarding the lifetime of the links. So, the authors in [14] suggests a better variety of OLSR protocol named “DistOLSR”. It is done by inserting a new metric named “Remaining Time to Quite (RTTQ)”. It makes use of distance and radio scoop to approximate the lifetime of links between nodes and their neighboring. RT T Q(t + dt) =
Range − newdist ∗ time − newdif −dest newdif −dist
(1)
After the simulation, Dist-OLSR improved numerous metrics, namely MPRs lifetime, packet Delivered Ratio (PDR) and Average Throughput Traffic (ATT). Many mobility models, in turn, act on the connectivity of nodes within a MANET network. Thus, numerous studies are focused in this direction and suggesting appropriate solutions. Three examples are discussed below. 2.1
Mobility vs Predictive MPR Selection
The main objective of research [15] present a predictive mobility model named MPR Predictive Algorithm. The purpose of this model is to attain predictive routes, granting access to the mobile node via current and preceding MPRs. Therefore, the authors positioned the predictive routes in predicting table. Thus, decrease the packet loss, when we positioned the predictive routes in predicting tables, for the mobile nodes by fitting the perception of redundant MPRs. The main steps of this algorithm is presented in the Algorithm 1. In this paper [16], the authors propose select dispersive Multi-Point Relays algorithm using the node localization. The results show reduces the amount of traffic overhead. Thus, it gets more bandwidth for the smart environment’s applications. The same authors proposed in another paper [17] a new decentralized technique, called geographic forwarding rules (GFRs), to reduce the number of redundant broadcast TC messages by divided the network into virtual zones and avoid the duplicate re-transmissions between zones.
498
H. Berradi et al.
Algorithm 1. MPR-Predictive Algorithm 1: Begin 2: while 1 − hopneighborisdeleted do 3: Process comparison (1-hop neighbor, MRP selector set); 4: if match == true then 5: if thresholdmeets then 6: Generate TC, declaring self as M P RP rev for entry deleted and Willingness == LOW 7: else 8: Generate TC, declaring self as M P RP rev for entry deleted. 9: end if 10: else 11: No action 12: end if 13: end while 14: while 1 − hopneighborisadded do 15: process comparison (1-hop neighbor, 2-hop neighbor) 16: if match == true then 17: if computingN ode == M P R then 18: Generate TC, declaring self as M P RP rev for entry added. 19: else 20: Generate HELLO, informing neighbors, hence MPRs about dynamic node. 21: end if 22: else 23: Disconnected node, not possible to be traced from our predictive algorithm. 24: end if 25: end while
In VANETs, we have the same issue as the MANETs, Due to the dynamic nature of traffic environment, the unpredictable vehicle movement, and frequent network topology change. The routing process still a challenging process. In order to offer stable routes and establish good throughput in VANETs, the authors of [18] present optimization in route selection and route maintenance to get better stability of route and decrease overhead. In other study [19], the authors proposed a technique that pick out a reliable intermediate node by using the mobility information included in beacons to improve the routing problem. Other authors investigate in [20] how to jointly optimize Route Setup and channel assignment to mobile ad hoc cognitive networks by proposing algorithms that completely avoid the interference with primary nodes and minimize the conflict to cognitive nodes. The showed result demonstrate that the proposed algorithm significantly improves various network performance.
Enhanced Mobile Network Stability Using Average Spatial Dependency
2.2
499
Mobility Quantification for MultiPoint Relays Selection Algorithm
The authors [21] propose An algorithm introduced for MPR selection using the mobility rate (MR) that is further based on node’s relative speed. While selecting the MPR process, the mobility rate will serve as measure. So, it is mandatory for one node to keep the record of calculated rate of the other node. With this, the neighbor nodes will exchange their respective nodes using Hello message of OLSR. The main steps of this algorithm is presented in the Algorithm 2. As shown in simulation, authors present the improvement in the performance of network such as: received packet, packet loss, rate of packet delivery and transmitted level of packet, using the mobility concept.
Algorithm 2. Mobility Rate Algorithm. 1: Begin 2: For all nodes in N, calculate the mobility rate MR(y), where y is a member of N. 3: Add to the MPR set those nodes in N, which are the only nodes to provide reachability to a node in N2.
4: remove the nodes from N2 which are now covered by a node in the MPR set. While there exist nodes in N2 which are not covered by at least one node in the MPR set
5: Calculate the reachability for each node in N. 6: Select as MPR, the node with the less mobility rate(MR) among the nodes in N with nonzero reachability.
7: In case of multiple nodes providing the same reachability: select the node as MPR whose degree D(y) is greater.
8: Remove the nodes from N2 which are now covered by a node in the MPR set.
Where: – N: the subset of neighbors of the node. – N2: the set of 2-hop neighbors. – D(y): the degree is the number of symmetric neighbors of node y. The writers in [22] suggest user cooperative mobility unlike the past research about cooperative user mobility. The results show that the best position, to move to, is quite different from the intuitively best position on the way to the next-hop, which will be beneficial for both cooperative mobility users and ad-hoc networks. Some studies [23] suggested a technique to predict by implemented the deep learning to know the currently movement information of mobile node using it’s movement history. Mobility of nodes can be predict using the suggested system, through RSS values in dynamic environment.
500
2.3
H. Berradi et al.
Mobility Models Selection
Performance of MANET and routing protocol can be highly influenced by the mobility model. Research have shown the effect in the network performance which used selected mobility model. For example, the authors of the paper [24] showed the impact of mobility models on the performance of MANET through simulation. The simulation results proved that the performance of MANET varies across different mobility models, different speed, and pause time. In other research, [25] authors proposed two mobility models AVM and UPM to enhance the routing protocols in an emergency scenario. A comparison of mobility models between AVM, UPM, Waypoint, Brownian, pursue on emergency and rescue scenario, ensure the betterment of AVM and UPM models than other models. In this paper [26], authors have presented the effect of situations of different models on the QoS Metrics of two protocols: Dynamic Source Routing (DSR) and On-Demand Distance Vector (AODV). The results have shown that DSR cause better outcomes as compared to AODV, in the different situation.
3
Impact of the Mobility on the Network
Since the impact of node mobility in a MANET network and its role in the study of protocols, and applications of mobile networks [27], it has the character of the main simulation factor of such a network. In this section, we begin with defining the simulation setup, to measure the impact of the mobility on the standard protocols between three protocols (OLSR, AODV, and DSDV) in term of the lost packet, delay and the throughput. Next, we launch a series of NS3 simulations (Network Simulation 3), and finally, we evaluate the results. We established a network consists of 30 nodes in the network simulator NS3. We conducted several experiments that were distributed to 25 tests for 200 seconds. Ten nodes are randomly selected to be a source of CBR (Constant Bit Rate) To generate traffic in the network. Moreover, these selected nodes use UDP (User Datagram Protocol). Note that since the nodes are mobile, and we are in an arbitrary simulation environment, we repeat every simulation 25 times to achieve good simulation results. The entire node is moving randomly using “Random Waypoint Mobility” in the simulation. In the Table 1 below, we show our simulation parameters used during simulations: 3.1
Results and Discussions
To show the impact of the mobility on the three selected protocols (OLSR, AODV, and DSDV), we used two crucial metrics, which we consider significant in a mobile network namely lost packet and the end-to-end delay.
Enhanced Mobile Network Stability Using Average Spatial Dependency
501
Table 1. Simulation parameters. Parameters Values Modulation
802.11b
Nodes
30
Mobility model
RandomWayPointMobility
Simulation time 200 (s) Packet size
256 (bytes)
Protocols
OLSR, AODV, DSDV
Speed
[0,10,20,30,40,50,60,70,80,90,100 (m/s)]
Simulation area 5000 * 5000
Lost Packet. Packet loss is the failure of one or more send packets to achieve their destinations. [28] This can cause noticeable effects in all types of communications in the network. In Fig. 1a we plot the Lost Packet over speed between the three protocols, we observe that due to the mobility of the nodes, the lost packet increases when the speed is increasing. The AODV had more lost packet than the DSDV, and the OLSR had the less lost packet than the other protocols. [29] Moreover, this is a reasonable result because the AODV is a reactive protocol so that it will suffer from the break links more than the two other protocols. Delay-Sum. We get the Delay-Sum by calculating the cumulative of the delay of all the flow between the source, and the destination in the simulation. In Fig. 1b we notice the same observation as the lost packet, AODV has the highest delay sum, and OLSR has the smallest one.
(a) The lost packet over speed
(b) The delay sum over speed
Fig. 1. The impact of the mobility on OLSR, AODV, DSDV
502
H. Berradi et al.
Thus, OLSR is the best protocol against mobility between the three selected protocols. In our proposal, we will make OLSR work better than standard protocols by adding an intelligent character so that it can send the Hello message periodically.
4
The Proposed Model
In this model, mobility metric will be used as an influential criteria for selecting MPRs. Conversely, special procedure will be used in this model comparative to previous study in the Related Work. The average spatial dependency Average will based on relative speed and relative acceleration. ManhattanGrid mobility model will be used on the evaluating basis of neighboring mobility to assess the performance based on mobility metrics represented in an IMPORTANT framework [10]. Some important term are being introduced before defining metrics that will be used in this paper to improve the understanding: – N: Number of mobile node – T: Time stimulation – P: Number of Tuples Mobility model will be studied by the use of following metrics in our proposed study: – Average Degree of Spatial Dependence: This metric could be defined as the measuring of velocities similarity between two neighbor nodes. N N T i=1 j=1 t=1
Dspatial(i,j,t)
(2) P Where P represents the set of triples (i, j, t) such that Dspatial (i, j, t) = 0. – Average Relative Speed: This type of metric calculates the velocity deviation within two neighbor nodes. Dspatial =
N N T
RS(i, j, t)
i=1 j=1 t=1
(3) P Pattern of mobility is differentiated in relative velocity metric as function of relative motion. This metric has been proposed in [10]. – Average Relative Acceleration: Acceleration deviation has to be measured in this metric between two neighbor nodes. RS =
N N T
RA =
RA(i, j, t)
i=1 j=1 t=1
P
(4)
Where P represents the set of triples (i, j, t) such that RA(i, j, t) = 0 as mentioned in the framework [10].
Enhanced Mobile Network Stability Using Average Spatial Dependency
503
– Average Spatial Dependency (ASD): It can be define through formula below: N
ASD =
RS(i, j, t) × RA(i, j, t)
i=1
P
(5)
A high value of ASD represents a good mobility design relative to its neighbor, while acceleration and velocity will be strongly linked with each other. The node with high ASD value will be qualified as MPR. Thus, the ASD enhanced the network stability and life time. ASD make it suitable for most of the mobile application such as the smart cities. 4.1
Proposed Solution
The relative distribution of node neighbor for example relation with nodes, different angles between them, and also the distance between node, could offer most suitable environment for the improvement of MPR selection through algorithm [30,31]. Algorithm process 3 is defined below: Algorithm 3. Improved algorithm of node relay selection 1: 2: 3: 4: 5: 6: 7: 8: 9: 10: 11: 12: 13: 14: 15: 16: 17: 18: 19: 20: 21:
Initial Node i: Current note N (i): Set for the one hope neighbor nodes N2 (i): Set for the two hope neighbor nodes MPRset: Set of all members of N with willingness equal to WILL ALWAYS Begin Add to the MPR set with node in N which are (Only) nodes to offer reachability to node N2 Remove the nodes from N2 which are now covered by a node in the MPR set while there exist nodes in N2 which are not covered by at least one node in the MPR set do For each node in N, calculate the reachability Select as a MPR the node with highest willingness among the nodes in N with non-zero reachability if multiple choices then select the node which provides reachability to the maximum number of nodes in N2 else if multiple choices then select node with highest value of ASD, else if multiple choices then select node with highest degree, that is, number if 1-hop neighbors. end if Remove the nodes from N2 which are now covered by a node in the MPRset. end while return MPRset of node i
504
H. Berradi et al.
4.2
Implementation
OLSR is A well suited protocol for the mobile networks that have protective routing mechanism that works in the distribution area of MANETs. The basic functions of OLSR are: – Detection of links and neighborhoods achieved by transmitted a periodic HELLO messages between the nodes, generated at static time interval (every 2 s in the standard) – Selection of the MPRs to disseminate the topology information to other nodes. While there is a big disadvantage of this mechanism that it have no option about how to understand network mobility basically. On the basis of the mobility group, our model inserts the additional metrics ASD that take the neighbor mobility into consideration in the MPR selection process. Every node will measure it ASD using the relative velocity and the relative acceleration of the surrounding nodes. To allow each node to share its informations with it’s nearing nodes a feature is added to hello message format. Three fields that are added such as ASD (2 octet), Veloc (1 octet and Accl (1 octet) which showed in Fig. 2. We don’t add those fields in TC message, because we need just one hope and two hope neighbors to get the rate of node mobility.
(a) 1- OLSR STD
(b) 2- OLST MODIFIED
Fig. 2. The standard Hello message vs the modified HELLO message format
We saved position of neighbor and estimate ASD. In the next step, MPR computational protocol use the ASD in MPR selection process to choose the best stable MPR nodes.
5
Result Discussion
C++ commonly used in the discrete event stimulator known as NS-3. Simulator has been used in this model enhancement in the stimulation of several network scenarios. Most important proactive routing Protocol known as OLSR routing protocol has been implicated in MANETs to develop modification in previous model that will call as Modified-OLSR. To get precise output results, each scenario has examined twenty times. Nodes have been warmed up to 10 s to attain
Enhanced Mobile Network Stability Using Average Spatial Dependency
505
a constant state prior to transmitting the data. We have evaluated the mobility metrics adjacent to ManhattanGrid mobility model. We have generated the mobility model using the framework provided by the tool BonnMotion [32], and we used Network Simulator – 3 [33] to simulate the model. 5.1
Simulation Setups
We have generated the mobility model using the BonnMotion framework [32]. The maximum nodes speeds defined in this set of values are 0, 1, 2, 3, 4, 5, 10, 15, 20, 25, 30 m/s and the nodes number in this study are fixed as 30 nodes.
Table 2. The simulation parameters used for experiment. Parameters Values Nodes number
30
Mobility model
ManhattanGrid
Simulation time
200 s
Protocols
OLSR, OLSR-Modified
Speed
0,1,2,3,4,5,10,15,20,25,30 m/s
Transmission range
40 m
Area
1000 m × 1000 m
Start time
10 s
Stop time
200 s
Physical and MAC layer 802.11b
The Table 2 provides the simulation parameters which we have used for our experiment. For the evaluation of the proposed metric, following performance metric have been selected such as Packet loss, End-to-end delay, Throughput. 5.2
Performance Results
The results obtained from the simulation are explained here. Packet Loss. It has been proved that modified OLSR model represents an enhanced activity at lower and higher mobility as shown in Fig. 3, The lost a pocket value of standard OLSR is much greater than our proposal, and that is significantly proved our model. That is the reason that ASD selects the best static path with the best quality proved by the selection of algorithm used in this model.
506
H. Berradi et al.
End-to-End Delay. Standard OLSR has higher end to end delay value as compared to modified OLSR. While, algorithm used in the model selects stable MPR nodes so it proves that the modified OLSR is more suitable by the reduction of route errors in delay process as shown in Fig. 4. Throughput. Throughput Modified OLSR is greater than of the standard OLSR as shown in Fig. 5. This result since we added three fields on message hello packet.
Fig. 3. Packet loss vs Velocity
Fig. 4. End-to-end delay vs Velocity
Fig. 5. Throughput vs Velocity
6
Conclusion
In this paper, we enhanced the OLSR in mobile environment using the proposed ASD metric that have been explained in detail above. The main purpose of this research was to improve the MPR selection using the metric mobile related ASD in MANET networks. MPR selection process has been improved through the use of superior algorithms to achieve our goal. Studies performed the imitation using dissimilar scenarios to contrast the performance of the metric with the model ManhattanGrid mobility by the OLSR routing protocol. These modifications evaluated the investigational consequences and demonstrated the stability
Enhanced Mobile Network Stability Using Average Spatial Dependency
507
and binding of modified metric greater than the standard protocol for different speeds. In future, we plan to make more comparison with other works and study that will impact on of our proposal on the energy consumption with the proposed spatial dependence model in order to enhance MPR selection and to extend the network life.
References 1. Perkins, C.E., Bhagwat, P.: Highly dynamic destination-sequenced distance-vector routing (DSDV) for mobile computers. In: ACM SIGCOMM Computer Communication Review, vol. 24, pp. 234–244. ACM (1994) 2. Clausen, T., Jacquet, P., Adjih, C., Laouiti, A., Minet, P., Muhlethaler, P., Qayyum, A., Viennot, L.: Optimized link state routing protocol (OLSR) (2003) 3. Perkins, C.E., Royer, E.M.: Ad-hoc on-demand distance vector routing. In: Proceedings WMCSA 1999. Second IEEE Workshop on Mobile Computing Systems and Applications, pp. 90–100. IEEE (1999) 4. Johnson, D.B., Maltz, D.A., Broch, J., et al.: DSR: The dynamic source routing protocol for multi-hop wireless ad hoc networks. Ad hoc Networking 5(1), 39–172 (2001) 5. Busson, A., Mitton, N., Fleury, E.: Analysis of the multi-point relay selection in OLSR and implications. In: IFIP Annual Mediterranean Ad Hoc Networking Workshop, pp. 387–396. Springer (2005) 6. Conti, M., Giordano, S.: Mobile ad hoc networking: milestones, challenges, and new research directions. IEEE Commun. Mag. 52(1), 85–96 (2014) 7. Boushaba, A., Benabbou, A., Benabbou, R., Zahi, A., Oumsis, M.: Multi-point relay selection strategies to reduce topology control traffic for OLSR protocol in MANETs. J. Netw. Comput. Appl. 53, 91–102 (2015) 8. Kitasuka, T., Tagashira, S.: Finding more efficient multipoint relay set to reduce topology control traffic of OLSR. In: 2013 IEEE 14th International Symposium on “A World of Wireless, Mobile and Multimedia Networks” (WoWMoM), pp. 1–9. IEEE (2013) 9. Abed, A.K., Oz, G., Aybay, I.: Influence of mobility models on the performance of data dissemination and routing in wireless mobile ad hoc networks. Comput. Electr. Eng. 40(2), 319–329 (2014) 10. Bai, F., Sadagopan, N., Helmy, A.: IMPORTANT: a framework to systematically analyze the impact of Mobility on Performance of RouTing protocols for Adhoc NeTworks. In: EEE INFOCOM 2003. Twenty-second Annual Joint Conference of the IEEE Computer and Communications Societies (IEEE Cat. No. 03CH37428), vol. 2, pp. 825–835. IEEE (2003) 11. Cavalcanti, E.R., Spohn, M.A.: Enhancing OLSR protocol performance through improved detection of spatial dependence. In: 2014 IEEE Symposium on Computers and Communications (ISCC), pp. 1–6. IEEE (2014) 12. Nishiyama, H., Ngo, T., Ansari, N., Kato, N.: On minimizing the impact of mobility on topology control in mobile ad hoc networks. IEEE Trans. Wireless Commun. 11(3), 1158–1166 (2012) 13. Sarmah, N.: Performance analysis of mobile ad-hoc routing protocols by varying mobility, speed and network load (2014)
508
H. Berradi et al.
14. Ouacha, A., Lakki, N., El Abbadi, J., Habbani, A., Bouamoud, B., Elkoutbi, M.: Reliable MPR selection based on link lifetime-prediction method. In: 2013 10th IEEE International Conference on Networking, Sensing and Control (ICNSC), pp. 11–16. IEEE (2013) 15. Khan, A.Y., Rashid, S., Iqbal, A.: Mobility vs. predictive MPR selection for mobile ad hoc networks using OLSR. In: Proceedings of the IEEE Symposium on Emerging Technologies, pp. 52–57. IEEE (2005) 16. Souidi, M., Habbani, A., Berradi, H., Essaid, B.: Node localization to optimize the MPR selection in smart mobile communication. In: Proceedings of the 2017 International Conference on Smart Digital Environment, pp. 8–13. ACM (2017) 17. Souidi, M., Habbani, A., Berradi, H., El Mahdi, F.: Geographic forwarding rules to reduce broadcast redundancy in mobile ad hoc wireless networks. Pers. Ubiquit. Comput. 23, 765–775 (2018) 18. Pathak, C., Shrivastava, A., Jain, A.: Ad-hoc on demand distance vector routing protocol using Dijkastra’s algorithm (AODV-D) for high throughput in VANET (Vehicular Ad-hoc Network). In: 2016 11th International Conference on Industrial and Information Systems (ICIIS), pp. 355–359, December 2016 19. Park, K., Kim, H., Lee, S.: Mobility state based routing method in vehicular ad-hoc network. In: 2015 IEEE International Conference on Mobile Services, pp. 473–474, June 2015 20. Tang, F., Guo, M., Guo, S., Xu, C.: Mobility prediction based joint stable routing and channel assignment for mobile ad hoc cognitive networks. IEEE Trans. Parallel Distrib. Syst. 27(3), 789–802 (2016) 21. Amraoui, H., Habbani, A., Hajami, A.: Mobility quantification for MultiPoint Relays selection algorithm in mobile Ad hoc networks. In: 2016 5th International Conference on Multimedia Computing and Systems (ICMCS), pp. 278–283. IEEE (2016) 22. Luo, T., Murase, T.: User cooperative mobility for QoS improvement in ad-hoc networks. In: 2017 14th IEEE Annual Consumer Communications Networking Conference (CCNC), pp. 276–279, January 2017 23. Yayeh, Y., Lin, H., Berie, G., Adege, A.B., Yen, L., Jeng, S.: Mobility prediction in mobile ad-hoc network using deep learning. In: 2018 IEEE International Conference on Applied System Invention (ICASI), pp. 1203–1206, April 2018 24. Appiah, M.: The impact of mobility models on the performance of mobile ad hoc network (MANET). In: 2016 International Conference on Advances in Computing and Communication Engineering (ICACCE), pp. 208–213, November 2016 25. Tandazo, R.T., Carri´ on, S.C., Gonzalez, P.L., Cueva, K.M.: Evaluation of mobility models to mobile networks in emergency and rescue scenarios AVM, UPM, pursue, waypoint and brownian. In: 2015 10th Iberian Conference on Information Systems and Technologies (CISTI), pp. 1–6, June 2015 26. Sudhakar, T., Inbarani, H.H.: Spatial group mobility model scenarios formation in mobile ad hoc networks. In: 2017 International Conference on Computer Communication and Informatics (ICCCI), pp. 1–5, January 2017 27. Rezende, C., Boukerche, A., Pazzi, R.W., Rocha, B.P.S., Loureiro, A.A.F.: The impact of mobility on mobile ad hoc networks through the perspective of complex networks. J. Parallel Distrib. Comput. 71(9), 1189–1200 (2011) 28. He, R., Peng, H., Jiang, Q., Zhou, L., Zhu, J.: Performance analysis and threshold quantization of transformer differential protection under sampled value packets loss/delay. IEEE Access 7, 55698–55706 (2019)
Enhanced Mobile Network Stability Using Average Spatial Dependency
509
29. Nabou, A., Laanaoui, M.D., Ouzzif, M.: The effect of transmit power on manet routing protocols using AODV, DSDV, DSR and OLSR in NS3. In: International Conference on Advanced Intelligent Systems for Sustainable Development, pp. 274– 286. Springer (2018) 30. Helmy, A., Bai, F.: A survey of mobility models in wireless ad hoc networks (2006) 31. Broch, J., Maltz, D.A., Johnson, D.B., Hu, Y.-C., Jetcheva, J.: A performance comparison of multi-hop wireless ad hoc network routing protocols, pp. 85–97. ACM (1998) 32. Aschenbruck, N., Ernst, R., Gerhards-Padilla, E., Schwamborn, M.: BonnMotion: a mobility scenario generation and analysis tool, p. 51. ICST (Institute for Computer Sciences, Social-Informatics and Telecommunications Engineering) (2010) 33. Henderson, T.R., Lacage, M., Riley, G.F., Dowell, C., Kopena, J.: Network simulations with the ns-3 simulator, vol. 14, no. 14, p. 527 (2008)
A Review on 3D Reconstruction Techniques from 2D Images M. Aharchi(&) and M. Ait Kbir(&) LIST Laboratory, Faculty of Sciences and Technologies, Tangier, Morocco [email protected], [email protected]
Abstract. In recent years, 3D model visualization techniques have made enormous progress. This evolution has not only touched the technical side but also the hardware side. It is no longer necessary to have expensive machines to see a world in 3D; a simple computer can do the trick. Historically, research has focused on the development of 3D information and acquisition techniques from scenes and objects. These acquisition methods require expertise and complex calibration procedures whenever the acquisition system was used. All this creates an important demand for flexibility in these methods of acquisition because of these different factors, many techniques have emerged. Many of them only need a camera and a computer to create a 3D world from a scene. Keywords: 3D reconstruction Camera Multiple views 2D images Depth perception
1 Introduction Creating a 3D model from 2D images that is realistic as possible is one of the fundamental issues of image-based modeling and computer vision. The best choice is an automatic reconstruction of the scene with little or no user intervention. Currently, there are several methods of 3D reconstruction from 2D images; each algorithm has its own conditions of execution, its strengths as well as its weak points. In this paper, we give a definition and the domains of uses of 3D reconstruction from 2D images. We propose a set of 3D reconstruction algorithms from these 2D images as well as a comparison between them. Afterwards, we conclude this paper with summarize the related to this study.
2 Overview on 3D Reconstruction from Images 2.1
What is 3D Reconstruction from Images
3D reconstruction from multiple images is the creation of three-dimensional models from a set of images. It is the reverse process of obtaining 2D images from 3D scenes. An image does not give us enough information to reconstruct a 3D scene. This is due to the nature of image forming process that consists of projecting a three-dimensional
© Springer Nature Switzerland AG 2020 M. Ben Ahmed et al. (Eds.): SCA 2019, LNITI, pp. 510–522, 2020. https://doi.org/10.1007/978-3-030-37629-1_37
A Review on 3D Reconstruction Techniques from 2D Images
511
scene onto a two-dimensional image. During this process, the depth is lost. The points visible on the images are the projections of the real points on the image. 2.2
Fields of Application of 3D Reconstruction from Images
3D reconstruction has applications in many fields. They are: Medicine, Free-viewpoint video reconstruction, robotic mapping, city planning, Gaming, virtual environments and virtual tourism, landslide inventory mapping, robot navigation, archaeology, augmented reality, reverse engineering, motion capture, Gesture recognition and hand tracking. In medicine, 3D reconstruction from 2D images can be used for both therapeutic and diagnostic purposes by using a camera to take multiple images at multiple angles. Even if there are medical imaging techniques like MRI and CT scanning, they are still expensive and can induce high radiation doses, which is a risk for patients with certain diseases, and they are not suitable for patients with ferromagnetic metallic implants. Both the methods can be done only when the patient is in lying position where the global structure of the bone changes. There are some techniques of 3D reconstruction like Stereo Corresponding Point Based Technique or Non-Stereo corresponding contour method (NCSS) which can be performed while standing and require low radiation dose by using X-ray images [1]. In the world of robotic navigation, an autonomous robot can use the images taken by its camera to create a 3D map of its environment and use it to perform real-time processing in order to find its way or to avoid obstacles that can arise at any moment in its path, as well as making measurements on the space where it is located [2]. The estimation of landslides dimension is a significant challenge while preparing the landslide inventory map, for which satellite aerial/data photography is required, which is very expensive. An alternative is the use of drones for such mapping. The result of 3D reconstruction from 2D images is very accurate and gives the possibility of measurements up to cm level and even small objects could be easily identified. By using images taking by a drone in combination with 3D scene reconstruction algorithms we can provide effective and flexible tools to monitor and map landslide [3]. 2.3
3D Reconstruction from Images Requirements
To obtain, as desired, the coordinates of the points of the scene, it is necessary to solve a certain number of problems: Calibration Problem Calibration problem or how the points of the scene are projected on the image. For this, the pinhole model is used and it is then necessary to know so-called intrinsic parameters of the camera (focal length, center of the image…). Then, it is necessary to know the relative position of the cameras to be able to determine the coordinates of the points of the space in a reference of the scene not linked to the camera. These parameters, called extrinsic, are the position and orientation of the camera in space. Matching Problem Is the ability to recognize and associate the points that appear on several pictures.
512
M. Aharchi and M. Ait Kbir
Reconstruction Problem It is a question of determining the 3D coordinates of the points from the associations made and the parameters of calibration. The Density of the Reconstruction Once the coordinates of a certain number of points in space have been obtained, it is necessary to find the surface to which these points belong to obtain a mesh, a dense model. Otherwise, in some cases, when we obtain a large number of points, the cloud of points formed is enough to visually define the shape of the object but the reconstruction is then sparse.
3 Active and Passive Reconstruction Methods The depths restoration process of visible points on the image can be achieved by active or passive methods. 3.1
Active Methods
In order to acquire a depth map, active methods actively interfere with the object to be reconstructed through radiometric or mechanical techniques (laser rangefinder, structured light and other active detection techniques). For example, a depth map can be reconstructed using a depth gauge to measure the depth relative to an object placed on a rotating plate or using radiometric methods through moving light sources, colored visible light, time-of-flight lasers, microwaves or ultrasounds that emit radiance towards the object and then measure its reflected part. 3.2
Passive Methods
Passive methods of 3D reconstruction do not interfere with objects to be rebuilt; they only use a sensor to measure the luminance reflected or emitted by the surface of the object in order to deduce its 3D structure through image processing. The sensor used in the camera is an image sensor in sensitive to visible light. The input elements for this process are a set of digital images (one, two or more) or video. For this case, we are talking about image-based reconstruction and the output element is a 3D model [5].
4 2D to 3D Conversion Algorithms Depending on the number of input images, we can categorize the existing conversion algorithms into two groups: algorithms based on two or more images and algorithms based on a single still image. In the first case, the two or more input images could be taken by multiple fixed cameras located either at different viewing angles or by a single camera with moving objects in the scenes. We call the depth cues used by the first group the multi-ocular depth cues. The second group of depth cues operates on a single still image, and they are referred to as the monocular depth cues. Table 1 summarizes the depth cues used in 2D to 3D conversion algorithms.
A Review on 3D Reconstruction Techniques from 2D Images
513
Table 1. Depth cue used in 2D to 3D conversion algorithms. Number of input images Depth cues Two or more images Binocular disparity Motion parallax Image blur Silhouette Structure from motion One single image Linear perspective Atmosphere scattering Shape from shading
4.1
Binocular Disparity
By using two images of the same scene captured from slightly different points of view, we can manage to recover the depth of a point present on the two images. First, a corresponding set of points in both images are found. Then, using the method of triangulation we can get to determine the depth of a point on the images [6] (Fig. 1).
Fig. 1. Binocular disparity [15].
We assume that Pl and Pr are the two projections of the points P on the two images and O1 and Or are the origins of the coordinate systems of the two cameras. Based on the relationship between the triangles (P P1 pr) and (P O1 Or) the depth Z of the point P can be obtained where D = xr − xl. Z¼f
T D
514
4.2
M. Aharchi and M. Ait Kbir
Motion Parallax
The relative movement between the camera and the scene provides important clues in the perception of depth. Objects that are close to the camera move faster than the objects that are near. The extraction of 3D structures and the camera is termed as structure from motion. The motion can be seen as a form of disparity over time, represented by the concept of motion field. The motion field is the 2D velocity vectors of the image points and the observed scene. The basic assumptions for structure-frommotion are that do not deform objects and their movements are linear. This fact has been exploited in several applications, such as wiggle stereoscopy [7] where motion parallax is used as a metaphor for stereoscopic images, or parallax scrolling [8] used in games where, by moving foreground and background at different speeds, a depth sensation is evoked. The strength of this cue is relatively high when compared to other monocular cues and also when compared to binocular disparity (Fig. 2).
Fig. 2. Motion parallax mechanics.
4.3
Image Blur
Evidence for the use of image blur in depth perception has been reported by Mather [9] and by Marshall et al. [4]. Their papers described experiments on ambiguous figureground stimuli, containing two regions of texture separated by a wavy boundary. Objects that are in-focus are clearly pictured whilst objects at other distances are defocused. The following general expression relates the distance d of a point from a lens to the radius s of its blurred image [25]: d ¼ Frv=ðrv Fðr þ sÞÞ
ð1Þ
Where F is focal length, r is lens aperture radius, and v is the distance of the image plane from the lens. If we know the values of F, r, and v and a measure of image blur s
A Review on 3D Reconstruction Techniques from 2D Images
515
is available, then absolute distance can be calculated. Equation (1) can be used to predict retinal blur as a function of distance, on assuming typical values for the optical parameters of the human eye (r = 1.5 mm, v = 16 mm) (Fig. 3).
Fig. 3. Blur as a depth cue in random patterns.
The upper rectangle is more far than the lower black rectangle because the upper rectangle has sharply defined edges. 4.4
Silhouette
A silhouette of an object in an image refers to the contour separating the object from the background. Shape-from-silhouette methods require multiple views of the scene taken by cameras from different viewpoints. Such a process together with correct texturing generates a full 3D model of the objects in the scene, allowing viewers to observe a live scene from an arbitrary viewpoint. Shape-from-silhouette requires accurate camera calibration. For each image, the silhouette of the target objects is segmented using background subtraction. The retrieved silhouettes are back projected to a common 3D space with projection centers equal to the camera locations. Back-projecting a silhouette produces a cone-like volume. The intersection of all the cones forms the visual hull of the target 3D object, which is often processed in the voxel representation. This 3D reconstruction procedure is referred to as shape-from-silhouette [10] (Fig. 4).
Fig. 4. Silhouette volume intersection.
516
M. Aharchi and M. Ait Kbir
C denotes a cube, which is an example of a 3D object, S denotes a 2D screen, PA denotes a viewpoint in 3D space, DA denotes a 2D polygon on the screen, which is the silhouette of the cube, and VA denotes the visual cone backprojected from the viewpoint PA. PB, DB, and VB denote the corresponding meanings to PA, DA, and VA [18]. 4.5
Linear Perspective
Linear perspective refers to parallel lines such as roads or pathways that converge with distance. The points of line of these lines are less visible than those of the nearest ones. The approach proposed by Battiato, Curti et al. [12] works for images containing surfaces with rigid geometry. The intersection with the most intersection points in the neighborhood is considered to be the vanishing point. The major lines close to the vanishing point are marked as the vanishing lines. Between each pair of neighboring vanishing lines, a set of gradient planes is assigned, each corresponding to a single depth level. The pixels closer to the vanishing points are assigned a larger depth value and the density of the gradient planes is higher [11] (Fig. 5).
Fig. 5. Blur as a depth cue in random patterns [16].
4.6
Atmosphere Scattering
Atmosphere scattering approach is based on the fact that the power and direction of light are changed when the light passes through the atmosphere because of small particles present in it. The objects that are close to the camera appear clearer while those near are more blurred [12]. In 1997, Krotkov and Cozman and [13] presented an analysis of this conversion. It was based on Lord Rayleigh’s 1871 physical scattering model. Their algorithm is suitable for estimating the depth of outdoor images containing a portion of sky (Fig. 6).
A Review on 3D Reconstruction Techniques from 2D Images
517
Fig. 6. Depth map from atmosphere scattering [18].
4.7
Shape from Shading
Shape from shading is a technique that allows knowing the surface normal of an object by observing the reflectance of the light on this object. The amount of light reflected by the surface of the object depends on the orientation of the object. Woodham introduced this technique in 1980. When the data is a single image, we call it shape from shading, and it was analyzed by B. K. Horn P. In 1989 Photometric stereo has since been generalized to many other situations, like non-Lambertian surface finishes and extended light sources. Multiple images of an object under different lighting are analyzed to produce an estimated normal direction at each pixel [14] (Fig. 7).
Fig. 7. Explanatory figure of shape from shading [17].
4.8
Structure from Motion
Structure from Motion (SfM) is a technique that uses a series of two-dimensional images of a scene or object to reconstruct its three-dimensional structure. SfM can produce 3D models based on high-resolution point clouds. SfM is based on the same principles as stereoscopic photogrammetry. In stereophogrammetry triangulation is used to calculate the relative 3-D positions (x, y, z,) of objects from stereo pairs. Traditionally these techniques require expensive specialized equipment and software.
518
M. Aharchi and M. Ait Kbir
To create a 3D reconstruction one simply needs many images of an area or an object with a high degree overlap, taken from different angles. The camera doesn’t need to be specialized, standard consumer-grade cameras work well for SfM methods. The images are often be taken from a moving sensor or by a one or multiple people at different locations and angles. SfM involves the three main stages: Step 1: Match corresponding features and measure distances between them on the camera image plane d, d’. Scale Invariant Feature Transform (SIFT) [30] allows corresponding features to be matched even with large variations in scale and viewpoint and under conditions of partial occlusion and changing illumination. Step 2: When we have the matching locations of multiple points on two or more photos, there is usually just one mathematical solution for where the photos were taken. Therefore, we can calculate individual camera positions (x, y, z), (x’, y’, z’), orientations i, i’, focal lengths f, f’, and relative positions of corresponding features b, h, in a single step known as “bundle adjustment”. This is where the term Structure from motion comes from. Scene structure refers to all these parameters; motion refers to movement of the camera. Step 3: Next, a dense point cloud and 3D surface is determined using the camera parameters and using the SfM points. This step is called “multiviewstereo matching” (MVS) (Fig. 8).
Fig. 8. Explanatory figure of structure from motion.
A Review on 3D Reconstruction Techniques from 2D Images
519
5 Comparison To perform an effective evaluation of the performances of the different algorithms of 2D to 3D conversion requires careful design of criteria and sets of data. Many articles do not provide a quantitative performance analysis explicit, which complicates the evaluation process. It is therefore unfair to say that a method represents a lower error rate or a slower execution time than others methods. Another complication is that for each individual algorithms, each having different characteristics and performances. Therefore, when we discuss issues such as accuracy or speed of algorithms of each depth cues, we rely, where appropriate, on the experimental results presented in the articles of these representative algorithms. The comparison is based on 8 qualitative aspects. The results are presented in the Table 2. 5.1
Relative or Absolute Depth
Most algorithms that rely on camera settings can recover the actual depth. Some algorithms provide a real (absolute) distance between the viewing camera and objects, and they are able to estimate the actual size of the object; other algorithms measure relative depth by analyzing shading, edges and junction, etc., providing a relative depth but not the actual values. On the other hand, monocular depth cannot be used to estimate the actual depth. Except in the case where correct the values obtained using machine learning techniques. 5.2
Depth Range
This aspect describes the effective depth range a human can perceive based on each individual depth cue. For example, linear perspective works in all ranges; and atmospheric scattering works only at large distance. 5.3
Real Time Processing
Some of the articles provide explicit run-time and environment parameters, others simply state that the algorithm is suitable for a real-time application or does not mention speed at all. The speed of algorithms is related to their accuracy. Greater accuracy requires more processing time. In order to obtain a real-time execution speed, it is simply necessary to reduce the accuracy to an acceptable limit. Traditional depth-of-focus methods [15] are fast but less accurate. Several researchers have published different techniques to improve accuracy, but they require very high computing costs. One of the examples of these suggestions for improving this algorithm is that proposed by Ahmad and Choi [16]. There the optimization technique of dynamic programming. Shape-fromSilhouette is memory intensive and normally computationally. Thanks to various techniques such as parallel PC processing or 2D intersection, it is possible to meet the real-time criterion. The hardware accelerated visual hulls algorithm developed by Li et al. [17] linked to several consumer PCs following a server-client architecture and makes arbitrary views of the visual hull directly from the input silhouette.
520
5.4
M. Aharchi and M. Ait Kbir
Image Acquisition
This aspect describes whether the method used is active or passive. In other words, parameters of the image acquisition system. Almost all multi-ocular depth cues require a special camera set-up, and most monocular depth cues do not. 5.5
Image Content
Indicates the characteristic that the image must have in order to be processed by algorithms and work properly. 5.6
Motion Presence
This aspect describes whether the points on the image are moving on the different images or not. It is only applicable for multi-ocular depth cues. Since monocular depths operate on a single picture. 5.7
Dense or Sparse Depth Map
This aspect deals with the density level of the depth map. It can be either dense or sparse. Some depth cues can generate dense and sparse depth maps, depending on whether the specific algorithm uses local feature points or global structures. A dense depth map is constructed using the features of the overall image. Each level of depth is assigned to each pixel of the image. A sparse depth map provides only depth values for feature points. It is more suitable for the extraction of 3D shapes. Table 2. Depth cue comparison Depth cues
Relative/Absolute depth
Depth range
Real time processing
Image acquisition
Image content
Motion presence
Dense or sparse depth map
State of depth cue
Binocular disparity
Absolute