Fraud Analytics Using Descriptive, Predictive, and Social Network Techniques

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.

• List of Figures xvForeword xxiiiPreface xxvAcknowledgments xxixChapter 1 Fraud: Detection, Prevention, and Analytics! 1Introduction 2Fraud! 2Fraud Detection and Prevention 10Big Data for Fraud Detection 15Data-Driven Fraud Detection 17Fraud-Detection Techniques 19Fraud Cycle 22The Fraud Analytics Process Model 26Fraud Data Scientists 30A Fraud Data Scientist Should Have Solid Quantitative Skills 30A Fraud Data Scientist Should Be a Good Programmer 31A Fraud Data Scientist Should Excel in Communication and Visualization Skills 31A Fraud Data Scientist Should Have a Solid Business Understanding 32A Fraud Data Scientist Should Be Creative 32A Scientific Perspective on Fraud 33References 35Chapter 2 Data Collection, Sampling, and Preprocessing 37Introduction 38Types of Data Sources 38Merging Data Sources 43Sampling 45Types of Data Elements 46Visual Data Exploration and Exploratory Statistical Analysis 47Benford’s Law 48Descriptive Statistics 51Missing Values 52Outlier Detection and Treatment 53Red Flags 57Standardizing Data 59Categorization 60Weights of Evidence Coding 63Variable Selection 65Principal Components Analysis 68RIDITs 72PRIDIT Analysis 73Segmentation 74References 75Chapter 3 Descriptive Analytics for Fraud Detection 77Introduction 78Graphical Outlier Detection Procedures 79Statistical Outlier Detection Procedures 83Break-Point Analysis 84Peer-Group Analysis 85Association Rule Analysis 87Clustering 89Introduction 89Distance Metrics 90Hierarchical Clustering 94Example of Hierarchical Clustering Procedures 97k-Means Clustering 104Self-Organizing Maps 109Clustering with Constraints 111Evaluating and Interpreting Clustering Solutions 114One-Class SVMs 117References 118Chapter 4 Predictive Analytics for Fraud Detection 121Introduction 122Target Definition 123Linear Regression 125Logistic Regression 127Basic Concepts 127Logistic Regression Properties 129Building a Logistic Regression Scorecard 131Variable Selection for Linear and Logistic Regression 133Decision Trees 136Basic Concepts 136Splitting Decision 137Stopping Decision 140Decision Tree Properties 141Regression Trees 142Using Decision Trees in Fraud Analytics 143Neural Networks 144Basic Concepts 144Weight Learning 147Opening the Neural Network Black Box 150Support Vector Machines 155Linear Programming 155The Linear Separable Case 156The Linear Nonseparable Case 159The Nonlinear SVM Classifier 160SVMs for Regression 161Opening the SVM Black Box 163Ensemble Methods 164Bagging 164Boosting 165Random Forests 166Evaluating Ensemble Methods 167Multiclass Classification Techniques 168Multiclass Logistic Regression 168Multiclass Decision Trees 170Multiclass Neural Networks 170Multiclass Support Vector Machines 171Evaluating Predictive Models 172Splitting Up the Data Set 172Performance Measures for Classification Models 176Performance Measures for Regression Models 185Other Performance Measures for Predictive Analytical Models 188Developing Predictive Models for Skewed Data Sets 189Varying the Sample Window 190Undersampling and Oversampling 190Synthetic Minority Oversampling Technique (SMOTE) 192Likelihood Approach 194Adjusting Posterior Probabilities 197Cost-sensitive Learning 198Fraud Performance Benchmarks 200References 201Chapter 5 Social Network Analysis for Fraud Detection 207Networks: Form, Components, Characteristics, and Their Applications 209Social Networks 211Network Components 214Network Representation 219Is Fraud a Social Phenomenon? An Introduction to Homophily 222Impact of the Neighborhood: Metrics 227Neighborhood Metrics 228Centrality Metrics 238Collective Inference Algorithms 246Featurization: Summary Overview 254Community Mining: Finding Groups of Fraudsters 254Extending the Graph: Toward a Bipartite Representation 266Multipartite Graphs 269Case Study: Gotcha! 270References 277Chapter 6 Fraud Analytics: Post-Processing 279Introduction 280The Analytical Fraud Model Life Cycle 280Model Representation 281Traffic Light Indicator Approach 282Decision Tables 283Selecting the Sample to Investigate 286Fraud Alert and Case Management 290Visual Analytics 296Backtesting Analytical Fraud Models 302Introduction 302Backtesting Data Stability 302Backtesting Model Stability 305Backtesting Model Calibration 308Model Design and Documentation 311References 312Chapter 7 Fraud Analytics: A Broader Perspective 313Introduction 314Data Quality 314Data-Quality Issues 314Data-Quality Programs and Management 315Privacy 317The RACI Matrix 318Accessing Internal Data 319Label-Based Access Control (LBAC) 324Accessing External Data 325Capital Calculation for Fraud Loss 326Expected and Unexpected Losses 327Aggregate Loss Distribution 329Capital Calculation for Fraud Loss Using Monte Carlo Simulation 331An Economic Perspective on Fraud Analytics 334Total Cost of Ownership 334Return on Investment 335In Versus Outsourcing 337Modeling Extensions 338Forecasting 338Text Analytics 340The Internet of Things 342Corporate Fraud Governance 344References 346About the Authors 347Index 349