Machine Learning Applications
Description
Practical resource on the importance of Machine Learning and Deep Learning applications in various technologies and real-world situations
Machine Learning Applications discusses methodological advancements of machine learning and deep learning, presents applications in image processing, including face and vehicle detection, image classification, object detection, image segmentation, and delivers real-world applications in healthcare to identify diseases and diagnosis, such as creating smart health records and medical imaging diagnosis, and provides real-world examples, case studies, use cases, and techniques to enable the reader's active learning.
Composed of 13 chapters, this book also introduces real-world applications of machine and deep learning in blockchain technology, cyber security, and climate change. An explanation of AI and robotic applications in mechanical design is also discussed, including robot-assisted surgeries, security, and space exploration. The book describes the importance of each subject area and detail why they are so important to us from a societal and human perspective.
Edited by two highly qualified academics and contributed to by established thought leaders in their respective fields, Machine Learning Applications includes information on:
* Content based medical image retrieval (CBMIR), covering face and vehicle detection, multi-resolution and multisource analysis, manifold and image processing, and morphological processing
* Smart medicine, including machine learning and artificial intelligence in medicine, risk identification, tailored interventions, and association rules
* AI and robotics application for transportation and infrastructure (e.g., autonomous cars and smart cities), along with global warming and climate change
* Identifying diseases and diagnosis, drug discovery and manufacturing, medical imaging diagnosis, personalized medicine, and smart health records
With its practical approach to the subject, Machine Learning Applications is an ideal resource for professionals working with smart technologies such as machine and deep learning, AI, IoT, and other wireless communications; it is also highly suitable for professionals working in robotics, computer vision, cyber security and more.
More details
Other editions
Additional editions
Persons
Indranath Chatterjee is a Professor in the Department of Computer Engineering, at Tongmyong University, South Korea. He received his PhD from University of Delhi, India and has authored several books and numerous, research papers. His areas of research are AI, computer vision, computation neuroscience and medical imaging.
Sheetal Zalte is an Assistant Professor in the Department of Computer Science at Shivaji University, India. She earned her PhD from Shivaji University, India, and has published many research papers. Her research area is mobile adhoc networks.
Content
About the Authors xiii
Preface xv
1 Statistical Similarity in Machine Learning 1
Dmitriy Klyushin
1.1 Introduction 1
1.2 Featureless Machine Learning 2
1.3 Two-Sample Homogeneity Measure 3
1.4 The Klyushin-Petunin Test 3
1.5 Experiments and Applications 4
1.6 Summary 6
References 6
2 Development of ML-Based Methodologies for Adaptive Intelligent E-Learning Systems and Time Series Analysis Techniques 11
Indra Kumari, Indranath Chatterjee, and Minho Lee
2.1 Introduction 11
2.1.1 Machine Learning 12
2.1.2 Types of Machine Learning 12
2.1.3 Learning Methods 13
2.1.4 E-Learning with Machine Learning 14
2.1.5 Need for Machine Learning 15
2.2 Methodological Advancement of Machine Learning 16
2.2.1 Automatic Learner Profiling Agent 16
2.2.2 Learning Materials' Content Indexing Agent 17
2.2.3 Adaptive Learning 17
2.2.4 Proposed Research 18
2.2.5 Multi-Perspective Learning 18
2.2.6 Machine Learning Recommender Agent for Customization 19
2.2.6.1 E-Learning 19
2.2.7 Data Creation 19
2.2.8 Naïve Bayes model 19
2.2.9 K-Means Model 20
2.3 Machine Learning on Time Series Analysis 21
2.3.1 Time Series Representation 22
2.3.2 Time Series Classification 24
2.3.3 Time Series Forecasting 25
2.4 Conclusion 26
Acknowledgment 28
Conflict of Interest 28
References 28
3 Time-Series Forecasting for Stock Market Using Convolutional Neural Network 31
Partha Pratim Deb, Diptendu Bhattacharya, Indranath Chatterjee, and Sheetal Zalte
3.1 Introduction 31
3.2 Materials 33
3.3 Methodology 33
3.3.1 The Convolutional Neural Network 34
3.4 Accuracy Measurement 35
3.5 Result and Discussion 35
3.6 Conclusion 47
Acknowledgement 47
References 48
4 Comparative Study for Applicability of Color Histograms for CBIR Used for Crop Leaf Disease Detection 49
Jayamala Kumar Patil, Sampada Abhijit Dhole, Vinay Sampatrao Mandlik, and Sachin B. Jadhav
4.1 Introduction 49
4.2 Literature Review 50
4.3 Methodology 51
4.3.1 Color Features 52
4.3.1.1 RGB Color Model/Space 53
4.3.1.2 HSV Color Space 53
4.3.1.3 YCbCr Color Space 54
4.3.1.4 Color Histogram 54
4.3.2 Database 54
4.3.3 Parameters for Performance Analysis 57
4.3.4 Experimental Procedure for CBIR Using Color Histogram for Detection of Disease 58
4.4 Results and Discussions 60
4.4.1 Results of CBIR Using Color Histogram for Detection of Soybean Alfalfa Mosaic Virus Disease 60
4.4.2 Results of CBIR Using Color Histogram for Detection of Soybean Septoria Brown Spot (SBS) Disease 62
4.4.3 Results of CBIR Using Color Histogram for Detection of Soybean Healthy Leaf 63
4.5 Conclusion 63
References 65
Biographies of Authors 67
5 Stock Index Forecasting Using RNN-Long Short-Term Memory 69
Partha Pratim Deb, Diptendu Bhattacharya, and Sheetal Zalte
5.1 Introduction 69
5.2 Materials 71
5.3 Methodology 71
5.3.1 RNN 71
5.3.2 LSTM 72
5.4 Result and Discussion 73
5.4.1 Comparison Table for the Method TAIEX 80
5.4.2 Comparison Table for Method BSE-SENSEX 80
5.4.3 Comparison Table for Method KOSPI 80
5.5 Conclusion 81
Acknowledgement 83
References 84
6 Study and Analysis of Machine Learning Models for Detection of Phishing URLs 85
Shreyas Desai, Sahil Salunkhe, Rashmi Deshmukh, and Sheetal Zalte
6.1 Introduction 85
6.2 Literature Review 86
6.3 Methodology 87
6.3.1 Proposed Work 87
6.3.2 Traditional Methods 87
6.3.2.1 Blacklist Method 88
6.3.2.2 Heuristic-Based Model 88
6.3.2.3 Visual Similarity 89
6.3.2.4 Machine Learning-Based Approach 89
6.4 Results and Experimentation 89
6.4.1 Dataset Creation 89
6.4.2 Feature Extraction 90
6.4.3 Training Data and Comparison 90
6.4.3.1 XGB (eXtreme Gradient Boosting) 90
6.4.3.2 Logistic Regression (LR) 90
6.4.3.3 RFC (Random Forest Classifier) 91
6.4.3.4 Decision Tree 91
6.4.3.5 SVM (Support Vector Machines) 91
6.4.3.6 KNN (K-Nearest Neighbors) 91
6.5 Model-Metric Analysis 91
6.6 Conclusion 94
References 94
7 Real-World Applications of BC Technology in Internet of Things 97
Pardeep Singh, Ajay Kumar, and Mayank Chopra
7.1 Introduction 97
7.1.1 Relevance and Benefits of Blockchain Technology Applications 98
7.2 Review of Existing Study 100
7.3 Background of Blockchain 101
7.3.1 Blockchain Stakeholders 101
7.3.2 What is Bitcoin? 102
7.3.3 Emergence of Bitcoin 102
7.3.4 Working of Bitcoin 102
7.3.5 Risk in Bitcoin 103
7.3.6 Legal Issues in Bitcoin 103
7.4 Blockchain Technology in Internet of Things 104
7.4.1 Need of Integrating Blockchain with IoT 104
7.4.1.1 IoT Data Traceability and Reliability 105
7.4.1.2 Superior Interoperability 105
7.4.1.3 Increased Security 105
7.4.1.4 IoT System Autonomous Interactions 106
7.4.2 Hyperledger 106
7.4.3 Ethereum 107
7.4.4 Iota 107
7.5 Challenges and Concerns in Integrating Blockchain with the IoT 108
7.5.1 Blockchain Challenges and Concern 108
7.5.1.1 Scalability 108
7.5.1.2 Privacy Infringement 109
7.5.2 Privacy and Security issues with Internet of Things 109
7.6 Blockchain Applications for the Internet of Things (BIoT Applications) 110
7.6.1 BIoT Applications for Smart Agriculture 111
7.6.2 Blockchain for Smart Agriculture 111
7.6.3 Intelligent Irrigation Driven by IoT 111
7.7 Application of BIoT in Healthcare 112
7.7.1 Interoperability 113
7.7.2 Improved Analytics and Data Storage 113
7.7.3 Increased Security 113
7.7.4 Immutability 114
7.7.5 Quicker Services 114
7.7.5.1 Transparency 114
7.8 Application of BIoT in Voting 115
7.9 Application of BIoT in Supply Chain 116
7.10 Summary 116
References 117
8 Advanced Persistent Threat: Korean Cyber Security Knack Model Impost and Applicability 123
Indra Kumari and Minho Lee
8.1 Introduction 123
8.2 Background Study 124
8.3 Literature Review 126
8.4 Research Questions 131
8.5 Research Objectives 131
8.6 Research Hypothesis 131
8.7 Phases of APT Outbreak 131
8.7.1 Gain Access 132
8.7.2 Establish Foothold 132
8.7.3 Deepen Access 133
8.7.4 Move Laterally 133
8.7.5 Look, Learn, and Remain 133
8.8 Research Methodology 134
8.8.1 South Korea Cyber Security Initiatives and Applicability 135
8.8.2 Korea's Cyber-Security Program Proposals 137
8.8.2.1 Modernized Multi-Negotiator Retreat Arrangement 137
8.8.2.2 Headway of the Realms Exemplary 137
8.8.2.3 Scrutiny of Over apt in Cyber Retreat 137
8.8.2.4 Indiscriminate Inconsistency Revealing 138
8.9 A Deception Exemplary of Counter-Offensive 138
8.10 Conclusion 141
Acknowledgment 142
Conflict of Interest 142
References 142
9 Integration of Blockchain Technology and Internet of Things: Challenges and Solutions 145
Aman Kumar Dhiman and Ajay Kumar
9.1 Introduction 145
9.2 Overview of Blockchain-IoT Integration 146
9.3 How Blockchain-IoT Work Together 146
9.3.1 Network in IoT Devices 147
9.3.2 Network in IoT with Blockchain Technology 148
9.3.3 Data Flow in IoT Devices 148
9.3.4 Data Flow in IoT with Blockchain 149
9.3.5 The Role of Blockchain in IoT 149
9.3.6 The Role of IoT in Blockchain 150
9.4 Blockchain-IoT Applications 151
9.5 Related Studies on Integration of IoT and Blockchain Applications 153
9.6 Challenges of Blockchain-IoT Integration 155
9.7 Solutions of Blockchain-IoT Integration 155
9.8 Future Directions for Blockchain-IoT Integration 156
9.9 Conclusion 157
References 157
10 Machine Learning Techniques for SWOT Analysis of Online Education System 161
Priyanka P. Shinde, Varsha P. Desai, T. Ganesh Kumar, Kavita S. Oza, and Sheetal Zalte
10.1 Introduction 161
10.2 Motivation 162
10.3 Objectives 163
10.4 Methodology 163
10.5 Dataset Preparation 164
10.6 Data Visualization and Analysis 170
10.6.1 Observations 171
10.7 Machine Learning Techniques Implementation 178
10.7.1 K-Nearest Neighbors 178
10.7.2 Decision Tree 178
10.7.3 Random Forest 178
10.7.4 Support Vector Machine 179
10.7.5 Logistic Regression 179
10.8 Conclusion 179
References 180
11 Crop Yield and Soil Moisture Prediction Using Machine Learning Algorithms 183
Debarghya Acharjee, Nibedita Mallik, Dipa Das, Mousumi Aktar, and Parijata Majumdar
11.1 Introduction 183
11.2 Literature Review 185
11.3 Methodology 187
11.4 Result and Discussion 190
11.5 Conclusion 191
References 193
12 Multirate Signal Processing in WSN for Channel Capacity and Energy Efficiency Using Machine Learning 195
Prashant R. Dike, T. S. Vishwanath, V. M. Rohokale, and D. S. Mantri
12.1 Introduction 195
12.2 Energy Management in WSN 197
12.3 Different Strategies to Increase Energy Efficiency 197
12.4 Algorithm Development 198
12.5 Results 202
12.6 Summary 203
References 203
13 Introduction to Mechanical Design of AI-Based Robotic System 207
Mohammad Zubair
13.1 Introduction 207
13.2 Mechanisms in a Robot 209
13.2.1 Serial Manipulator 209
13.2.2 Parallel Manipulator 209
13.3 Kinematics 212
13.3.1 Degree of Freedom 214
13.3.2 Position and Orientation in a Robotic System 215
13.4 Conclusion 216
Acknowledgment 217
Conflict of Interest 217
References 217
Index 219
1
Statistical Similarity in Machine Learning
Dmitriy Klyushin
Department of Computer Science and Cybernetics, Kyiv, Ukraine
1.1 Introduction
In machine learning, the accuracy of algorithms depends on how accurately the hypothesis about the proximity of objects in the feature space is fulfilled. It is this property that guarantees the possibility of generalization based on training samples. The hypothesis of proximity (similarity) of objects in the feature space assumes that objects of one class form a compact region with a smooth boundary. A classic demonstration of this conjecture is the famous Fisher iris problem, in which points of three classes form easily separable and dense clouds on a plane. This problem illustrates both the strength and the weakness of the compactness hypothesis. The strength of this hypothesis is that we can easily draw boundaries between sets of points and classify them. The weakness of the compactness hypothesis is that that we cannot generalize it to the case when the object is defined not by one point, but by many points. Such situations often arise in medical research, when we take a lot of cells from a patient and measure different features of these cells. As a result, a patient is represented not by a vector in a feature space, but by a matrix of feature samples (moreover, the order of the numbers in the columns of this matrix is random). Of course, it is possible to reduce this matrix to a vector by averaging values in the columns and considering only a vector of means, but it is obvious that this leads to losing of important information about the distribution of feature values. In fact, what can be said about a distribution, knowing only the estimate of its mathematical expectation?
The hypothesis of compactness ignores the randomness of training data, so we must replace it with an alternative postulate on the proximity between random samples, guided by the laws of mathematical statistics. We propose to use the well-known concept of sample homogeneity in mathematical statistics, i.e. a hypothesis that samples are drawn from a same distribution. Returning to the terminology of machine learning, this means that samples of features of objects have identical distributions. Within this approach, we can use a wide variety of statistical criteria to test the homogeneity hypothesis.
In the chapter, we introduce an alternative concept of proximity in machine learning and propose to use the hypothesis about homogeneity of samples instead of the hypothesis of compactness, as well as provide examples of its effective use.
1.2 Featureless Machine Learning
The pioneers of featureless, or relational machine learning, were scientific schools of Duin (Duin et al. 1997, 1999; Pekalska and Duin 2001; Pekalska and Duin 2005) and Mottl (Mottl et al. 2001, 2002, 2017; Seredin et al. 2012). Their idea was to replace a feature vector of an object by a similarity measure to the training dataset using a metrics. It is obvious that this is not a solution of the problem of classification of objects using a matrix of feature values. The point is that in such cases, it is necessary to use not geometric but statistical tools, for example, two-sample tests of homogeneity, such as the Kolmogorov-Smirnov test and the Mann-Whitney-Wilcoxon test. Using these criteria, we can test the hypothesis that feature samples are homogeneous. However, this is not a complete solution of the posed problem. Testing the homogeneity hypothesis using the tests mentioned above and various other tests, for example, Cramer-von Mises and Anderson-Darling, we cannot obtain a numerical measure of similarity. These tests provide only p-values that denote the probability of the samples being homogeneous. We shall describe a solution allowing measuring the similarity between samples as follows.
To fill the distance matrix, Euclidean and pseudo-Euclidean distances, as well as kernels, are used. It is quite obvious that such an approach is not acceptable for estimating the similarity between matrices whose columns are random samples of features. The use of metrics in such cases is impossible.
Recently, the minimal learning machine (Kuli 2013) and the extreme minimal learning machine (Souza Junior et al. 2015) were developed. The authors used nonlinear distance regression, estimating dissimilarity between objects. There are numerous metrics and learning techniques in this field (Mesquita et al. 2017; Caldas et al. 2018; Florêncio et al. 2018; Maia et al. 2018; Cao et al. 2019; Kärkkäinen 2019; Bicego 2020; Florêncio et al. 2020; Nanni et al. 2020; Silva et al. 2020 etc.) Details of the surveys of these issues are provided in (Costa et al. 2020; Hämäläinen et al. 2020). All these methods use the Euclidean distance. Therefore, they are unacceptable for solving the problem stated above: to classify objects represented by matrices of independent identically distributed random values.
Our goal is to extend the featureless approach to similarity-based classification using the nonparametric similarity measure and nonparametric two-sample test of homogeneity. Due to the nonparametric nature of these tools, we do not use any assumption about a hypothetical distribution of training sample. Also, as we shall demonstrate below, these tools are universal in the sense that using the proposed test, we can test the homogeneity hypothesis for all possible variants: different location parameters and the same scale parameter, the same location parameter and different scale parameters, and both different location and scale parameters. The proposed similarity measure also is universal because it is applicable to both samples without ties and with ties (duplicates).
1.3 Two-Sample Homogeneity Measure
Consider training samples a = (a1, a2, ., an) ? A and b = (b1, b2, ., bn) ? B from populations A and B obeying distributions F and G that are absolutely continuous. The classification problem for a test sample c = (c1, c2, ., cn) is reduced to testing the homogeneity of c and a and c and b. There are various nonparametric two-sample tests of homogeneity (Derrick et al. 2019). However, every test has own drawbacks. For example, the Kolmogorov-Smirnov test is a universal test in the sense that it tests the general hypothesis F = G, but it is very sensible to outliers and need in large size of samples. The Wilcoxon sign rank test is not universal because it tests only the hypothesis about location shift (i.e. whether E(a) significantly differs from E(c)). In our opinion, the most effective and universal tool was developed in Klyushin and Petunin (2003).
1.4 The Klyushin-Petunin Test
The two-sample test of homogeneity (Klyushin and Petunin 2003) is nonparametric. This test uses the Hill's assumption (Hill 1968): for exchangeable random values a1, a2, ., an ? F with continuous distribution, we have
(1.1)where a(i) and a(j) are order statistics and x obeys F.
Let us findand estimate the deviation of the observable relative frequency hij from the expected probability (1.1) constructing a confidence interval for a probability of success in the Bernoulli scheme (Pires and Amado 2008). Since p-statistics is invariant in respect of the selection of a confidence interval for binomial proportion (Klyushin and Martynenko 2021), we may select the most simple one for computations, the Wilson confidence interval where
(1.2)If g = 3, the confidence level of (1.2) is greater than 0.95 (Klyushin and Petunin 2003). Since the number of all the intervals (a(i), a(j)) where i < j is equal to , the homogeneity measure for samples a and c is
(1.3)Note that h in (1.3) is also a binomial proportion. Therefore, the test for homogeneity may be formulated in the following way: samples are homogeneous if the confidence interval for the binomial h covers 0.95, else it is rejected.
1.5 Experiments and Applications
Consider the results of two numerical experiments in which samples were drawn from the normal distribution Gaussian(a, 1) and Gaussian(0, 1) and Gaussian(0, 1 - a) and Gaussian(0, 1). We considered 100 pairs of samples containing 100 random numbers. The p-statistics and p-value of the Kolmogorov-Smirnov statistics (KS-statistics) were averaged. The null hypothesis is accepted if the p-statistic is greater than 0.95 or the p-value of KS-test is less than 0.05. We tested hypotheses about shift of location and scale parameters. In the first case, the null hypothesis supposes that distributions have the same mathematical expectation. In the latter case, the null hypothesis supposes that...
