Details

<p>Preface xiii</p> <p>About the Companion Website xvii</p> <p><b>1 Unsupervised Machine Learning (ML) Techniques 1</b></p> <p>Introduction 1</p> <p>Selection of the Right Algorithm in ML 2</p> <p>Classical Multidimensional Scaling of Predictors Data 2</p> <p>Principal Component Analysis (PCA) 6</p> <p><i>k</i>-Means Clustering 13</p> <p>Distance Metrics: Locations of Cluster Centroids 13</p> <p>Replications 14</p> <p>Gaussian Mixture Model (GMM) Clustering 15</p> <p>Optimum Number of GMM Clusters 17</p> <p>Observations and Clusters Visualization 18</p> <p>Evaluating Cluster Quality 21</p> <p>Silhouette Plots 22</p> <p>Hierarchical Clustering 23</p> <p>Step 1 -- Determine Hierarchical Structure 23</p> <p>Step 2 -- Divide Hierarchical Tree into Clusters 25</p> <p>PCA and Clustering: Wine Quality 27</p> <p>Feature Selection Using Laplacian (fsulaplacian) for Unsupervised Learning 35</p> <p>CHW 1.1 The Iris Flower Features Data 37</p> <p>CHW 1.2 The Ionosphere Data Features 38</p> <p>CHW 1.3 The Small Car Data 39</p> <p>CHW 1.4 Seeds Features Data 40</p> <p><b>2 ML Supervised Learning: Classification Models 42</b></p> <p>Fitting Data Using Different Classification Models 42</p> <p>Customizing a Model 43</p> <p>Creating Training and Test Datasets 43</p> <p>Predicting the Response 45</p> <p>Evaluating the Classification Model 45</p> <p>KNN Model for All Categorical or All Numeric Data Type 47</p> <p>KNN Model: Heart Disease Numeric Data 48</p> <p>Viewing the Fitting Model Properties 50</p> <p>The Fitting Model: Number of Neighbors and Weighting Factor 51</p> <p>The Cost Penalty of the Fitting Model 52</p> <p>KNN Model: Red Wine Data 55</p> <p>Using MATLAB Classification Learner 57</p> <p>Binary Decision Tree Model for Multiclass Classification of All Data Types 68</p> <p>Classification Tree Model: Heart Disease Numeric Data Types 70</p> <p>Classification Tree Model: Heart Disease All Predictor Data Types 72</p> <p>Naive Bayes Classification Model for All Data Types 74</p> <p>Fitting Heart Disease Numeric Data to Naive Bayes Model 75</p> <p>Fitting Heart Disease All Data Types to Naive Bayes Model 77</p> <p>Discriminant Analysis (DA) Classifier for Numeric Predictors Only 79</p> <p>Discriminant Analysis (DA): Heart Disease Numeric Predictors 82</p> <p>Support Vector Machine (SVM) Classification Model for All Data Types 84</p> <p>Properties of SVM Model 85</p> <p>SVM Classification Model: Heart Disease Numeric Data Types 87</p> <p>SVM Classification Model: Heart Disease All Data Types 90</p> <p>Multiclass Support Vector Machine (fitcecoc) Model 92</p> <p>Multiclass Support Vector Machines Model: Red Wine Data 95</p> <p>Binary Linear Classifier (fitclinear) to High-Dimensional Data 98</p> <p>CHW 2.1 Mushroom Edibility Data 100</p> <p>CHW 2.2 1994 Adult Census Income Data 100</p> <p>CHW 2.3 White Wine Classification 101</p> <p>CHW 2.4 Cardiac Arrhythmia Data 102</p> <p>CHW 2.5 Breast Cancer Diagnosis 102</p> <p><b>3 Methods of Improving ML Predictive Models 103</b></p> <p>Accuracy and Robustness of Predictive Models 103</p> <p>Evaluating a Model: Cross-Validation 104</p> <p>Cross-Validation Tune-up Parameters 105</p> <p>Partition with K-Fold: Heart Disease Data Classification 106</p> <p>Reducing Predictors: Feature Transformation and Selection 108</p> <p>Factor Analysis 110</p> <p>Feature Transformation and Factor Analysis: Heart Disease Data 113</p> <p>Feature Selection 115</p> <p>Feature Selection Using predictorImportance Function: Health Disease Data 116</p> <p>Sequential Feature Selection (SFS): sequentialfs Function with Model Error Handler 118</p> <p>Accommodating Categorical Data: Creating Dummy Variables 121</p> <p>Feature Selection with Categorical Heart Disease Data 122</p> <p>Ensemble Learning 126</p> <p>Creating Ensembles: Heart Disease Data 130</p> <p>Ensemble Learning: Wine Quality Classification 131</p> <p>Improving fitcensemble Predictive Model: Abalone Age Prediction 132</p> <p>Improving fitctree Predictive Model with Feature Selection (FS): Credit Ratings Data 134</p> <p>Improving fitctree Predictive Model with Feature Transformation (FT): Credit Ratings Data 135</p> <p>Using MATLAB Regression Learner 136</p> <p>Feature Selection and Feature Transformation Using Regression Learner App 145</p> <p>Feature Selection Using Neighborhood Component Analysis (NCA) for Regression: Big Car Data 146</p> <p>CHW 3.1 The Ionosphere Data 148</p> <p>CHW 3.2 Sonar Dataset 149</p> <p>CHW 3.3 White Wine Classification 150</p> <p>CHW 3.4 Small Car Data (Regression Case) 152</p> <p><b>4 Methods of ML Linear Regression 153</b></p> <p>Introduction 153</p> <p>Linear Regression Models 154</p> <p>Fitting Linear Regression Models Using fitlm Function 155</p> <p>How to Organize the Data? 155</p> <p>Results Visualization: Big Car Data 162</p> <p>Fitting Linear Regression Models Using fitglm Function 164</p> <p>Nonparametric Regression Models 166</p> <p>fitrtree Nonparametric Regression Model: Big Car Data 167</p> <p>Support Vector Machine, fitrsvm, Nonparametric Regression Model: Big Car Data 170</p> <p>Nonparametric Regression Model: Gaussian Process Regression (GPR) 172</p> <p>Regularized Parametric Linear Regression 176</p> <p>Ridge Linear Regression: The Penalty Term 176</p> <p>Fitting Ridge Regression Models 177</p> <p>Predicting Response Using Ridge Regression Models 178</p> <p>Determining Ridge Regression Parameter, λ 179</p> <p>The Ridge Regression Model: Big Car Data 179</p> <p>The Ridge Regression Model with Optimum λ: Big Car Data 181</p> <p>Regularized Parametric Linear Regression Model: Lasso 183</p> <p>Stepwise Parametric Linear Regression 186</p> <p>Fitting Stepwise Linear Regression 187</p> <p>How to Specify stepwiselm Model? 187</p> <p>Stepwise Linear Regression Model: Big Car Data 188</p> <p>CHW 4.1 Boston House Price 192</p> <p>CHW 4.2 The Forest Fires Data 193</p> <p>CHW 4.3 The Parkinson’s Disease Telemonitoring Data 194</p> <p>CHW 4.4 The Car Fuel Economy Data 195</p> <p><b>5 Neural Networks 197</b></p> <p>Introduction 197</p> <p>Feed-Forward Neural Networks 198</p> <p>Feed-Forward Neural Network Classification 199</p> <p>Feed-Forward Neural Network Regression 200</p> <p>Numeric Data: Dummy Variables 200</p> <p>Neural Network Pattern Recognition (nprtool) Application 201</p> <p>Command-Based Feed-Forward Neural Network Classification: Heart Data 210</p> <p>Neural Network Regression (nftool) 214</p> <p>Command-Based Feed-Forward Neural Network Regression: Big Car Data 223</p> <p>Training the Neural Network Regression Model Using fitrnet Function: Big Car Data 226</p> <p>Finding the Optimum Regularization Strength for Neural Network Using Cross-Validation: Big Car Data 229</p> <p>Custom Hyperparameter Optimization in Neural Network Regression: Big Car Data 231</p> <p>CHW 5.1 Mushroom Edibility Data 233</p> <p>CHW 5.2 1994 Adult Census Income Data 233</p> <p>CHW 5.3 Breast Cancer Diagnosis 234</p> <p>CHW 5.4 Small Car Data (Regression Case) 234</p> <p>CHW 5.5 Boston House Price 235</p> <p><b>6 Pretrained Neural Networks: Transfer Learning 237</b></p> <p>Deep Learning: Image Networks 237</p> <p>Data Stores in MATLAB 241</p> <p>Image and Augmented Image Datastores 243</p> <p>Accessing an Image File 246</p> <p>Retraining: Transfer Learning for Image Recognition 247</p> <p>Convolutional Neural Network (CNN) Layers: Channels and Activations 256</p> <p>Convolution 2-D Layer Features via Activations 258</p> <p>Extraction and Visualization of Activations 261</p> <p>A 2-D (or 2-D Grouped) Convolutional Layer 264</p> <p>Features Extraction for Machine Learning 267</p> <p>Image Features in Pretrained Convolutional Neural Networks (CNNs) 268</p> <p>Classification with Machine Learning 268</p> <p>Feature Extraction for Machine Learning: Flowers 269</p> <p>Pattern Recognition Network Generation 271</p> <p>Machine Learning Feature Extraction: Spectrograms 275</p> <p>Network Object Prediction Explainers 278</p> <p>Occlusion Sensitivity 278</p> <p>imageLIME Features Explainer 282</p> <p>gradCAM Features Explainer 284</p> <p>HCW 6.1 CNN Retraining for Round Worms Alive or Dead Prediction 286</p> <p>HCW 6.2 CNN Retraining for Food Images Prediction 286</p> <p>HCW 6.3 CNN Retraining for Merchandise Data Prediction 287</p> <p>HCW 6.4 CNN Retraining for Musical Instrument Spectrograms Prediction 288</p> <p>HCW 6.5 CNN Retraining for Fruit/Vegetable Varieties Prediction 289</p> <p><b>7 A Convolutional Neural Network (CNN) Architecture and Training 290</b></p> <p>A Simple CNN Architecture: The Land Satellite Images 291</p> <p>Displaying Satellite Images 291</p> <p>Training Options 294</p> <p>Mini Batches 295</p> <p>Learning Rates 296</p> <p>Gradient Clipping 297</p> <p>Algorithms 298</p> <p>Training a CNN for Landcover Dataset 299</p> <p>Layers and Filters 302</p> <p>Filters in Convolution Layers 307</p> <p>Viewing Filters: AlexNet Filters 308</p> <p>Validation Data 311</p> <p>Using shuffle Function 316</p> <p>Improving Network Performance 319</p> <p>Training Algorithm Options 319</p> <p>Training Data 319</p> <p>Architecture 320</p> <p>Image Augmentation: The Flowers Dataset 322</p> <p>Directed Acyclic Graphs Networks 329</p> <p>Deep Network Designer (DND) 333</p> <p>Semantic Segmentation 342</p> <p>Analyze Training Data for Semantic Segmentation 343</p> <p>Create a Semantic Segmentation Network 345</p> <p>Train and Test the Semantic Segmentation Network 350</p> <p>HCW 7.1 CNN Creation for Round Worms Alive or Dead Prediction 356</p> <p>HCW 7.2 CNN Creation for Food Images Prediction 357</p> <p>HCW 7.3 CNN Creation for Merchandise Data Prediction 358</p> <p>HCW 7.4 CNN Creation for Musical Instrument Spectrograms Prediction 358</p> <p>HCW 7.5 CNN Creation for Chest X-ray Prediction 359</p> <p>HCW 7.6 Semantic Segmentation Network for CamVid Dataset 359</p> <p><b>8 Regression Classification: Object Detection 361</b></p> <p>Preparing Data for Regression 361</p> <p>Modification of CNN Architecture from Classification to Regression 361</p> <p>Root-Mean-Square Error 364</p> <p>AlexNet-Like CNN for Regression: Hand-Written Synthetic Digit Images 364</p> <p>A New CNN for Regression: Hand-Written Synthetic Digit Images 370</p> <p>Deep Network Designer (DND) for Regression 374</p> <p>Loading Image Data 375</p> <p>Generating Training Data 375</p> <p>Creating a Network Architecture 376</p> <p>Importing Data 378</p> <p>Training the Network 378</p> <p>Test Network 383</p> <p>YOLO Object Detectors 384</p> <p>Object Detection Using YOLO v4 386</p> <p>COCO-Based Creation of a Pretrained YOLO v4 Object Detector 387</p> <p>Fine-Tuning of a Pretrained YOLO v4 Object Detector 389</p> <p>Evaluating an Object Detector 394</p> <p>Object Detection Using R-CNN Algorithms 396</p> <p>R-CNN 397</p> <p>Fast R-CNN 397</p> <p>Faster R-CNN 398</p> <p>Transfer Learning (Re-Training) 399</p> <p>R-CNN Creation and Training 399</p> <p>Fast R-CNN Creation and Training 403</p> <p>Faster R-CNN Creation and Training 408</p> <p>evaluateDetectionPrecision Function for Precision Metric 413</p> <p>evaluateDetectionMissRate for Miss Rate Metric 417</p> <p>HCW 8.1 Testing yolov4ObjectDetector and fasterRCNN Object Detector 424</p> <p>HCW 8.2 Creation of Two CNN-based yolov4ObjectDetectors 424</p> <p>HCW 8.3 Creation of GoogleNet-Based Fast R-CNN Object Detector 425</p> <p>HCW 8.4 Creation of a GoogleNet-Based Faster R-CNN Object Detector 426</p> <p>HCW 8.5 Calculation of Average Precision and Miss Rate Using GoogleNet-Based Faster R-CNN Object Detector 427</p> <p>HCW 8.6 Calculation of Average Precision and Miss Rate Using GoogleNet-Based yolov4</p> <p>Object Detector 427</p> <p>HCW 8.7 Faster RCNN-based Car Objects Prediction and Calculation of Average Precision for Training and Test Data 427</p> <p><b>9 Recurrent Neural Network (RNN) 430</b></p> <p>Long Short-Term Memory (LSTM) and BiLSTM Network 430</p> <p>Train LSTM RNN Network for Sequence Classification 437</p> <p>Improving LSTM RNN Performance 441</p> <p>Sequence Length 441</p> <p>Classifying Categorical Sequences 445</p> <p>Sequence-to-Sequence Regression Using Deep Learning: Turbo Fan Data 446</p> <p>Classify Text Data Using Deep Learning: Factory Equipment Failure Text Analysis -- 1 453</p> <p>Classify Text Data Using Deep Learning: Factory Equipment Failure Text Analysis -- 2 462</p> <p>Word-by-Word Text Generation Using Deep Learning -- 1 465</p> <p>Word-by-Word Text Generation Using Deep Learning -- 2 473</p> <p>Train Network for Time Series Forecasting Using Deep Network Designer (DND) 475</p> <p>Train Network with Numeric Features 486</p> <p>HCW 9.1 Text Classification: Factory Equipment Failure Text Analysis 491</p> <p>HCW 9.2 Text Classification: Sentiment Labeled Sentences Data Set 492</p> <p>HCW 9.3 Text Classification: Netflix Titles Data Set 492</p> <p>HCW 9.4 Text Regression: Video Game Titles Data Set 492</p> <p>HCW 9.5 Multivariate Classification: Mill Data Set 493</p> <p>HCW 9.6 Word-by-Word Text Generation Using Deep Learning 494</p> <p><b>10 Image/Video-Based Apps 495</b></p> <p>Image Labeler (IL) App 495</p> <p>Creating ROI Labels 498</p> <p>Creating Scene Labels 499</p> <p>Label Ground Truth 500</p> <p>Export Labeled Ground Truth 501</p> <p>Video Labeler (VL) App: Ground Truth Data Creation, Training, and Prediction 502</p> <p>Ground Truth Labeler (GTL) App 513</p> <p>Running/Walking Classification with Video Clips using LSTM 520</p> <p>Experiment Manager (EM) App 526</p> <p>Image Batch Processor (IBP) App 533</p> <p>HCW 10.1 Cat Dog Video Labeling, Training, and Prediction -- 1 537</p> <p>HCW 10.2 Cat Dog Video Labeling, Training, and Prediction -- 2 537</p> <p>HCW 10.3 EM Hyperparameters of CNN Retraining for Merchandise Data Prediction 538</p> <p>HCW 10.4 EM Hyperparameters of CNN Retraining for Round Worms Alive or Dead Prediction 539</p> <p>HCW 10.5 EM Hyperparameters of CNN Retraining for Food Images Prediction 540</p> <p><b>Appendix A Useful MATLAB Functions 543</b></p> <p>A.1 Data Transfer from an External Source into MATLAB 543</p> <p>A.2 Data Import Wizard 543</p> <p>A.3 Table Operations 544</p> <p>A.4 Table Statistical Analysis 547</p> <p>A.5 Access to Table Variables (Column Titles) 547</p> <p>A.6 Merging Tables with Mixed Columns and Rows 547</p> <p>A.7 Data Plotting 548</p> <p>A.8 Data Normalization 549</p> <p>A.9 How to Scale Numeric Data Columns to Vary Between 0 and 1 549</p> <p>A.10 Random Split of a Matrix into a Training and Test Set 550</p> <p>A.11 Removal of NaN Values from a Matrix 550</p> <p>A.12 How to Calculate the Percent of Truly Judged Class Type Cases for a Binary Class Response 550</p> <p>A.13 Error Function m-file 551</p> <p>A.14 Conversion of Categorical into Numeric Dummy Matrix 552</p> <p>A.15 evaluateFit2 Function 553</p> <p>A.16 showActivationsForChannel Function 554</p> <p>A.17 upsampLowRes Function 555</p> <p>A.18A preprocessData function 555</p> <p>A.18B preprocessData2 function 555</p> <p>A.19 processTurboFanDataTrain function 556</p> <p>A.20 processTurboFanDataTest Function 556</p> <p>A.21 preprocessText Function 557</p> <p>A.22 documentGenerationDatastore Function 557</p> <p>A.23 subset Function for an Image Data Store Partition 560</p> <p>Index 561</p>

<p><b>Kamal I. M. Al-Malah</b> received his PhD degree from Oregon State University in 1993. He served as a Professor of Chemical Engineering in Jordan and Gulf countries, as well as Former Chairman of the Chemical Engineering Department at the University of Hail in Saudi Arabia. Professor Al-Malah is an expert in both Aspen Plus^® and MATLAB^® applications. He has created a bundle of Windows-based software for engineering applications.

<p><b>In-depth resource covering machine and deep learning methods using MATLAB tools and algorithms, providing insights and algorithmic decision-making processes</b> <p><i>Machine and Deep Learning Using MATLAB</i> introduces early career professionals to the power of MATLAB to explore machine and deep learning applications by explaining the relevant MATLAB tool or app and how it is used for a given method or a collection of methods. Its properties, in terms of input and output arguments, are explained, the limitations or applicability is indicated via an accompanied text or a table, and a complete running example is shown with all needed MATLAB command prompt code. <p>The text also presents the results, in the form of figures or tables, in parallel with the given MATLAB code, and the MATLAB written code can be later used as a template for trying to solve new cases or datasets. Throughout, the text features worked examples in each chapter for self-study with an accompanying website providing solutions and coding samples. Highlighted notes draw the attention of the user to critical points or issues. <p>Readers will also find information on: <ul><li>Numeric data acquisition and analysis in the form of applying computational algorithms to predict the numeric data patterns (clustering or unsupervised learning)</li> <li>Relationships between predictors and response variable (supervised), categorically sub-divided into classification (discrete response) and regression (continuous response)</li> <li>Image acquisition and analysis in the form of applying one of neural networks, and estimating net accuracy, net loss, and/or RMSE for the successive training, validation, and testing steps</li> <li>Retraining and creation for image labeling, object identification, regression classification, and text recognition</li></ul> <p><i>Machine and Deep Learning Using MATLAB</i> is a useful and highly comprehensive resource on the subject for professionals, advanced students, and researchers who have some familiarity with MATLAB and are situated in engineering and scientific fields, who wish to gain mastery over the software and its numerous applications.

US