Fraud Analytics Using Descriptive, Predictive, and Social Network Techniques A Guide to Data Science for Fraud Detection

List of Figures xv

Foreword xxiii

Preface xxv

Acknowledgments xxix

Chapter 1 Fraud: Detection, Prevention, and Analytics! 1

Introduction 2

Fraud! 2

Fraud Detection and Prevention 10

Big Data for Fraud Detection 15

Data-Driven Fraud Detection 17

Fraud-Detection Techniques 19

Fraud Cycle 22

The Fraud Analytics Process Model 26

Fraud Data Scientists 30

A Fraud Data Scientist Should Have Solid Quantitative Skills 30

A Fraud Data Scientist Should Be a Good Programmer 31

A Fraud Data Scientist Should Excel in

Communication and Visualization Skills 31

A Fraud Data Scientist Should Have a Solid Business Understanding 32

A Fraud Data Scientist Should Be Creative 32

A Scientific Perspective on Fraud 33

References 35

Chapter 2 Data Collection, Sampling, and Preprocessing 37

Introduction 38

Types of Data Sources 38

Merging Data Sources 43

Sampling 45

Types of Data Elements 46

Visual Data Exploration and Exploratory Statistical Analysis 47

Benford’s Law 48

Descriptive Statistics 51

Missing Values 52

Outlier Detection and Treatment 53

Red Flags 57

Standardizing Data 59

Categorization 60

Weights of Evidence Coding 63

Variable Selection 65

Principal Components Analysis 68

RIDITs 72

PRIDIT Analysis 73

Segmentation 74

References 75

Chapter 3 Descriptive Analytics for Fraud Detection 77

Introduction 78

Graphical Outlier Detection Procedures 79

Statistical Outlier Detection Procedures 83

Break-Point Analysis 84

Peer-Group Analysis 85

Association Rule Analysis 87

Clustering 89

Introduction 89

Distance Metrics 90

Hierarchical Clustering 94

Example of Hierarchical Clustering Procedures 97

k-Means Clustering 104

Self-Organizing Maps 109

Clustering with Constraints 111

Evaluating and Interpreting Clustering Solutions 114

One-Class SVMs 117

References 118

Chapter 4 Predictive Analytics for Fraud Detection 121

Introduction 122

Target Definition 123

Linear Regression 125

Logistic Regression 127

Basic Concepts 127

Logistic Regression Properties 129

Building a Logistic Regression Scorecard 131

Variable Selection for Linear and Logistic Regression 133

Decision Trees 136

Basic Concepts 136

Splitting Decision 137

Stopping Decision 140

Decision Tree Properties 141

Regression Trees 142

Using Decision Trees in Fraud Analytics 143

Neural Networks 144

Basic Concepts 144

Weight Learning 147

Opening the Neural Network Black Box 150

Support Vector Machines 155

Linear Programming 155

The Linear Separable Case 156

The Linear Nonseparable Case 159

The Nonlinear SVM Classifier 160

SVMs for Regression 161

Opening the SVM Black Box 163

Ensemble Methods 164

Bagging 164

Boosting 165

Random Forests 166

Evaluating Ensemble Methods 167

Multiclass Classification Techniques 168

Multiclass Logistic Regression 168

Multiclass Decision Trees 170

Multiclass Neural Networks 170

Multiclass Support Vector Machines 171

Evaluating Predictive Models 172

Splitting Up the Data Set 172

Performance Measures for Classification Models 176

Performance Measures for Regression Models 185

Other Performance Measures for Predictive Analytical Models 188

Developing Predictive Models for Skewed Data Sets 189

Varying the Sample Window 190

Undersampling and Oversampling 190

Synthetic Minority Oversampling Technique (SMOTE) 192

Likelihood Approach 194

Adjusting Posterior Probabilities 197

Cost-sensitive Learning 198

Fraud Performance Benchmarks 200

References 201

Chapter 5 Social Network Analysis for Fraud Detection 207

Networks: Form, Components, Characteristics, and Their Applications 209

Social Networks 211

Network Components 214

Network Representation 219

Is Fraud a Social Phenomenon? An Introduction to Homophily 222

Impact of the Neighborhood: Metrics 227

Neighborhood Metrics 228

Centrality Metrics 238

Collective Inference Algorithms 246

Featurization: Summary Overview 254

Community Mining: Finding Groups of Fraudsters 254

Extending the Graph: Toward a Bipartite Representation 266

Multipartite Graphs 269

Case Study: Gotcha! 270

References 277

Chapter 6 Fraud Analytics: Post-Processing 279

Introduction 280

The Analytical Fraud Model Life Cycle 280

Model Representation 281

Traffic Light Indicator Approach 282

Decision Tables 283

Selecting the Sample to Investigate 286

Fraud Alert and Case Management 290

Visual Analytics 296

Backtesting Analytical Fraud Models 302

Introduction 302

Backtesting Data Stability 302

Backtesting Model Stability 305

Backtesting Model Calibration 308

Model Design and Documentation 311

References 312

Chapter 7 Fraud Analytics: A Broader Perspective 313

Introduction 314

Data Quality 314

Data-Quality Issues 314

Data-Quality Programs and Management 315

Privacy 317

The RACI Matrix 318

Accessing Internal Data 319

Label-Based Access Control (LBAC) 324

Accessing External Data 325

Capital Calculation for Fraud Loss 326

Expected and Unexpected Losses 327

Aggregate Loss Distribution 329

Capital Calculation for Fraud Loss Using Monte Carlo Simulation 331

An Economic Perspective on Fraud Analytics 334

Total Cost of Ownership 334

Return on Investment 335

In Versus Outsourcing 337

Modeling Extensions 338

Forecasting 338

Text Analytics 340

The Internet of Things 342

Corporate Fraud Governance 344

References 346

About the Authors 347

Index 349

BART BAESENS is a full professor at KU Leuven, and a lecturer at the University of Southampton. He has done extensive research on analytics, customer relationship management, web analytics, fraud detection, and credit risk management. He regularly advises and provides consulting support to international firms with respect to their analytics and credit risk management strategy.

VÉRONIQUE VAN VLASSELAER is a PhD researcher in the Department of Decision Sciences and Information Management at KU Leuven. Her research focuses on the development of new techniques for fraud detection by combining predictive and network analytics.

WOUTER VERBEKE is an assistant professor at Vrije Universiteit Brussel (Brussels, Belgium). His research is situated in the field of predictive analytics and complex network analysis with applications in fraud, marketing, credit risk, human resources management, and mobility.