发帖
雷达卡
This edition first published 2015
© 2015 by John Wiley & Sons, Ltd
Registered office:JohnWiley&Sons, Ltd, TheAtrium, SouthernGate, Chichester,West Sussex, PO19 8SQ, UnitedKingdom
For details of our global editorial offices, for customer services and for information about how to apply for permission to
reuse the copyright material in this book please see our website at www.wiley.com.
The right of the author to be identified as the author of this work has been asserted in accordance with the Copyright, Designs
======================
Contents
Acknowledgements xix
Preface xxi
References xxxi
Part I Preliminaries 1
1 Tasks 3
1.1 Introduction 3
1.1.1 Knowledge 4
1.1.2 Inference 4
1.2 Inductive learning tasks 5
1.2.1 Domain 5
1.2.2 Instances 5
1.2.3 Attributes 5
1.2.4 Target attribute 6
1.2.5 Input attributes 6
1.2.6 Training set 6
1.2.7 Model 7
1.2.8 Performance 7
1.2.9 Generalization 8
1.2.10 Overfitting 8
1.2.11 Algorithms 8
1.2.12 Inductive learning as search 9
1.3 Classification 9
1.3.1 Concept 10
1.3.2 Training set 10
1.3.3 Model 11
1.3.4 Performance 12
1.3.5 Generalization 13
1.3.6 Overfitting 13
1.3.7 Algorithms 13
1.4 Regression 14
1.4.1 Target function 14
1.4.2 Training set 14
1.4.3 Model 15
1.4.4 Performance 15
1.4.5 Generalization 15
1.4.6 Overfitting 15
1.4.7 Algorithms 16
1.5 Clustering 16
1.5.1 Motivation 16
1.5.2 Training set 17
1.5.3 Model 18
1.5.4 Crisp vs. soft clustering 18
1.5.5 Hierarchical clustering 18
1.5.6 Performance 18
1.5.7 Generalization 19
1.5.8 Algorithms 19
1.5.9 Descriptive vs. predictive clustering 19
1.6 Practical issues 19
1.6.1 Incomplete data 20
1.6.2 Noisy data 20
1.7 Conclusion 20
1.8 Further readings 21
References 22
2 Basic statistics 23
2.1 Introduction 23
2.2 Notational conventions 24
2.3 Basic statistics as modeling 24
2.4 Distribution description 25
2.4.1 Continuous attributes 25
2.4.2 Discrete attributes 36
2.4.3 Confidence intervals 40
2.4.4 m-Estimation 43
2.5 Relationship detection 47
2.5.1 Significance tests 48
2.5.2 Continuous attributes 50
2.5.3 Discrete attributes 52
2.5.4 Mixed attributes 56
2.5.5 Relationship detection caveats 61
2.6 Visualization 62
2.6.1 Boxplot 62
2.6.2 Histogram 63
2.6.3 Barplot 64
2.7 Conclusion 65
2.8 Further readings 66
References 67
Part II Classification 69
3 Decision trees 71
3.1 Introduction 71
3.2 Decision tree model 72
3.2.1 Nodes and branches 72
3.2.2 Leaves 74
3.2.3 Split types 74
3.3 Growing 76
3.3.1 Algorithm outline 76
3.3.2 Class distribution calculation 78
3.3.3 Class label assignment 79
3.3.4 Stop criteria 80
3.3.5 Split selection 82
3.3.6 Split application 86
3.3.7 Complete process 86
3.4 Pruning 90
3.4.1 Pruning operators 91
3.4.2 Pruning criterion 91
3.4.3 Pruning control strategy 100
3.4.4 Conversion to rule sets 101
3.5 Prediction 103
3.5.1 Class label prediction 104
3.5.2 Class probability prediction 104
3.6 Weighted instances 105
3.7 Missing value handling 106
3.7.1 Fractional instances 106
3.7.2 Surrogate splits 113
3.8 Conclusion 114
3.9 Further readings 114
References 116
4 Naïve Bayes classifier 118
4.1 Introduction 118
4.2 Bayes rule 118
4.3 Classification by Bayesian inference 120
4.3.1 Conditional class probability 120
4.3.2 Prior class probability 121
4.3.3 Independence assumption 122
4.3.4 Conditional attribute value probabilities 122
4.3.5 Model construction 123
4.3.6 Prediction 124
4.4 Practical issues 125
4.4.1 Zero and small probabilities 125
4.4.2 Linear classification 126
4.4.3 Continuous attributes 127
4.4.4 Missing attribute values 128
4.4.5 Reducing naïvety 129
4.5 Conclusion 131
4.6 Further readings 131
References 132
5 Linear classification 134
5.1 Introduction 134
5.2 Linear representation 136
5.2.1 Inner representation function 137
5.2.2 Outer representation function 138
5.2.3 Threshold representation 139
5.2.4 Logit representation 142
5.3 Parameter estimation 145
5.3.1 Delta rule 145
5.3.2 Gradient descent 149
5.3.3 Distance to decision boundary 152
5.3.4 Least squares 153
5.4 Discrete attributes 154
5.5 Conclusion 155
5.6 Further readings 156
References 157
6 Misclassification costs 159
6.1 Introduction 159
6.2 Cost representation 161
6.2.1 Cost matrix 161
6.2.2 Per-class cost vector 162
6.2.3 Instance-specific costs 163
6.3 Incorporating misclassification costs 164
6.3.1 Instance weighting 164
6.3.2 Instance resampling 167
6.3.3 Minimum-cost rule 169
6.3.4 Instance relabeling 174
6.4 Effects of cost incorporation 176
6.5 Experimental procedure 180
6.6 Conclusion 184
6.7 Further readings 185
References 187
7 Classification model evaluation 189
7.1 Introduction 189
7.1.1 Dataset performance 189
7.1.2 Training performance 189
7.1.3 True performance 189
7.2 Performance measures 190
7.2.1 Misclassification error 191
7.2.2 Weighted misclassification error 191
7.2.3 Mean misclassification cost 192
7.2.4 Confusion matrix 194
7.2.5 ROC analysis 200
7.2.6 Probabilistic performance measures 210
7.3 Evaluation procedures 213
7.3.1 Model evaluation vs. modeling procedure evaluation 213
7.3.2 Evaluation caveats 214
7.3.3 Hold-out 217
7.3.4 Cross-validation 219
7.3.5 Leave-one-out 221
7.3.6 Bootstrapping 223
7.3.7 Choosing the right procedure 227
7.3.8 Evaluation procedures for temporal data 230
7.4 Conclusion 231
7.5 Further readings 232
References 233
Part III Regression 235
8 Linear regression 237
8.1 Introduction 237
8.2 Linear representation 238
8.2.1 Parametric representation 239
8.2.2 Linear representation function 240
8.2.3 Nonlinear representation functions 241
8.3 Parameter estimation 242
8.3.1 Mean square error minimization 242
8.3.2 Delta rule 243
8.3.3 Gradient descent 245
8.3.4 Least squares 248
8.4 Discrete attributes 250
8.5 Advantages of linear models 251
8.6 Beyond linearity 252
8.6.1 Generalized linear representation 252
8.6.2 Enhanced representation 255
8.6.3 Polynomial regression 256
8.6.4 Piecewise-linear regression 257
8.7 Conclusion 258
8.8 Further readings 258
References 259
9 Regression trees 261
9.1 Introduction 261
9.2 Regression tree model 262
9.2.1 Nodes and branches 262
9.2.2 Leaves 262
9.2.3 Split types 262
9.2.4 Piecewise-constant regression 262
9.3 Growing 263
9.3.1 Algorithm outline 264
9.3.2 Target function summary statistics 265
9.3.3 Target value assignment 266
9.3.4 Stop criteria 267
9.3.5 Split selection 268
9.3.6 Split application 271
9.3.7 Complete process 272
9.4 Pruning 274
9.4.1 Pruning operators 275
9.4.2 Pruning criterion 275
9.4.3 Pruning control strategy 277
9.5 Prediction 277
9.6 Weighted instances 278
9.7 Missing value handling 279
9.7.1 Fractional instances 279
9.7.2 Surrogate splits 284
9.8 Piecewise linear regression 284
9.8.1 Growing 285
9.8.2 Pruning 289
9.8.3 Prediction 290
9.9 Conclusion 292
9.10 Further readings 292
References 293
10 Regression model evaluation 295
10.1 Introduction 295
10.1.1 Dataset performance 295
10.1.2 Training performance 295
10.1.3 True performance 295
10.2 Performance measures 296
10.2.1 Residuals 296
10.2.2 Mean absolute error 297
10.2.3 Mean square error 297
10.2.4 Root mean square error 299
10.2.5 Relative absolute error 299
10.2.6 Coefficient of determination 300
10.2.7 Correlation 301
10.2.8 Weighted performance measures 301
10.2.9 Loss functions 302
10.3 Evaluation procedures 303
10.3.1 Hold-out 304
10.3.2 Cross-validation 304
10.3.3 Leave-one-out 305
10.3.4 Bootstrapping 305
10.3.5 Choosing the right procedure 307
10.4 Conclusion 309
10.5 Further readings 309
References 310
Part IV Clustering 311
11 (Dis)similarity measures 313
11.1 Introduction 313
11.2 Measuring dissimilarity and similarity 313
11.3 Difference-based dissimilarity 314
11.3.1 Euclidean distance 314
11.3.2 Minkowski distance 315
11.3.3 Manhattan distance 316
11.3.4 Canberra distance 316
11.3.5 Chebyshev distance 317
11.3.6 Hamming distance 318
11.3.7 Gower’s coefficient 318
11.3.8 Attribute weighting 320
11.3.9 Attribute transformation 320
11.4 Correlation-based similarity 321
11.4.1 Discrete attributes 322
11.4.2 Pearson’s correlation similarity 322
11.4.3 Spearman’s correlation similarity 323
11.4.4 Cosine similarity 323
11.5 Missing attribute values 324
11.6 Conclusion 325
11.7 Further readings 325
References 326
12 k-Centers clustering 328
12.1 Introduction 328
12.1.1 Basic principle 328
12.1.2 (Dis)similarity measures 329
12.2 Algorithm scheme 330
12.2.1 Initialization 331
12.2.2 Stop criteria 331
12.2.3 Cluster formation 331
12.2.4 Implicit cluster modeling 332
12.2.5 Instantiations 332
12.3 k-Means 334
12.3.1 Center adjustment 335
12.3.2 Minimizing dissimilarity to centers 336
12.4 Beyond means 338
12.4.1 k-Medians 338
12.4.2 k-Medoids 339
12.5 Beyond (fixed)k 342
12.5.1 Multiple runs 343
12.5.2 Adaptivek-centers 343
12.6 Explicit cluster modeling 343
12.7 Conclusion 345
12.8 Further readings 345
References 347
13 Hierarchical clustering 349
13.1 Introduction 349
13.1.1 Basic approaches 349
13.1.2 (Dis)similarity measures 349
13.2 Cluster hierarchies 351
13.2.1 Motivation 351
13.2.2 Model representation 352
13.3 Agglomerative clustering 353
13.3.1 Algorithm scheme 353
13.3.2 Cluster linkage 356
13.4 Divisive clustering 361
13.4.1 Algorithm scheme 361
13.4.2 Wrapping a flat clustering algorithm 361
13.4.3 Stop criteria 362
13.5 Hierarchical clustering visualization 364
13.6 Hierarchical clustering prediction 366
13.6.1 Cutting cluster hierarchies 366
13.6.2 Cluster membership assignment 368
13.7 Conclusion 369
13.8 Further readings 370
References 371
14 Clustering model evaluation 373
14.1 Introduction 373
14.1.1 Dataset performance 373
14.1.2 Training performance 374
14.1.3 True performance 374
14.2 Per-cluster quality measures 376
14.2.1 Diameter 376
14.2.2 Separation 377
14.2.3 Isolation 378
14.2.4 Silhouette width 379
14.2.5 Davies–Bouldin index 382
14.3 Overall quality measures 385
14.3.1 Dunn index 386
14.3.2 Average Davies–Bouldin index 387
14.3.3 C index 388
14.3.4 Average silhouette width 389
14.3.5 Loglikelihood 390
14.4 External quality measures 393
14.4.1 Misclassification error 393
14.4.2 Rand index 394
14.4.3 General relationship detection measures 396
14.5 Using quality measures 397
14.6 Conclusion 398
14.7 Further readings 398
References 399
Part V Getting Better Models 401
15 Model ensembles 403
15.1 Introduction 403
15.2 Model committees 404
15.3 Base models 406
15.3.1 Different training sets 406
15.3.2 Different algorithms 412
15.3.3 Different parameter setups 412
15.3.4 Algorithm randomization 412
15.3.5 Base model diversity 418
15.4 Model aggregation 420
15.4.1 Voting/Averaging 420
15.4.2 Probability averaging 422
15.4.3 Weighted voting/averaging 424
15.4.4 Using as attributes 427
15.5 Specific ensemble modeling algorithms 431
15.5.1 Bagging 431
15.5.2 Stacking 433
15.5.3 Boosting 433
15.5.4 Random forest 443
15.5.5 Random Naïve Bayes 446
15.6 Quality of ensemble predictions 448
15.7 Conclusion 449
15.8 Further readings 450
References 451
16 Kernel methods 454
16.1 Introduction 454
16.2 Support vector machines 457
16.2.1 Classification margin 457
16.2.2 Maximum-margin hyperplane 460
16.2.3 Primal form 460
16.2.4 Dual form 464
16.2.5 Soft margin 468
16.3 Support vector regression 473
16.3.1 Regression tube 474
16.3.2 Primal form 475
16.3.3 Dual form 475
16.4 Kernel trick 482
16.5 Kernel functions 484
16.5.1 Linear kernel 485
16.5.2 Polynomial kernel 485
16.5.3 Radial kernel 485
16.5.4 Sigmoid kernel 486
16.6 Kernel prediction 487
16.7 Kernel-based algorithms 489
16.7.1 Kernel-based SVM 489
16.7.2 Kernel-based SVR 492
16.8 Conclusion 494
16.9 Further readings 495
References 496
17 Attribute transformation 498
17.1 Introduction 498
17.2 Attribute transformation task 499
17.2.1 Target task 499
17.2.2 Target attribute 500
17.2.3 Transformed attribute 500
17.2.4 Training set 500
17.2.5 Modeling transformations 500
17.2.6 Nonmodeling transformations 503
17.3 Simple transformations 504
17.3.1 Standardization 504
17.3.2 Normalization 505
17.3.3 Aggregation 506
17.3.4 Imputation 507
17.3.5 Binary encoding 508
17.4 Multiclass encoding 510
17.4.1 Encoding and decoding functions 511
17.4.2 1-ok-kencoding 514
17.4.3 Error-correcting encoding 515
17.4.4 Effects of multiclass encoding 519
17.5 Conclusion 521
17.6 Further readings 521
References 522
18 Discretization 524
18.1 Introduction 524
18.2 Discretization task 525
18.2.1 Motivation 525
18.2.2 Task definition 526
18.2.3 Discretization as modeling 527
18.2.4 Discretization quality 529
18.3 Unsupervised discretization 530
18.3.1 Equal-width intervals 530
18.3.2 Equal-frequency intervals 531
18.3.3 Nonmodeling discretization 532
18.4 Supervised discretization 533
18.4.1 Pure-class discretization 533
18.4.2 Bottom-up discretization 535
18.4.3 Top-down discretization 546
18.5 Effects of discretization 551
18.6 Conclusion 553
18.7 Further readings 553
References 556
19 Attribute selection 558
19.1 Introduction 558
19.2 Attribute selection task 559
19.2.1 Motivation 559
19.2.2 Task definition 560
19.2.3 Algorithms 561
19.3 Attribute subset search 562
19.3.1 Search task 562
19.3.2 Initial state 563
19.3.3 Search operators 564
19.3.4 State selection 564
19.3.5 Stop criteria 565
19.4 Attribute selection filters 568
19.4.1 Simple statistical filters 568
19.4.2 Correlation-based filters 571
19.4.3 Consistency-based filters 575
19.4.4 RELIEF 577
19.4.5 Random forest 584
19.4.6 Cutoff criteria 585
19.4.7 Filter-driven search 586
19.5 Attribute selection wrappers 588
19.5.1 Subset evaluation 588
19.5.2 Wrapper attribute selection 591
19.6 Effects of attribute selection 593
19.7 Conclusion 598
19.8 Further readings 599
References 600
20 Case studies 602
20.1 Introduction 602
20.1.1 Datasets 603
................................................
提升卡
置顶卡
沉默卡
变色卡
抢沙发
千斤顶
显身卡
京公网安备 11010802022788号
