Business Knowledge Series course

Presented by Bart Baesens, Ph.D. Professor at KU Leuven (Belgium), and lecturer at the University of Southampton (UK); or Christophe Mues, Ph.D., Professor at the School of Management of the University of Southampton (UK); or Cristian Bravo, Ph.D., Associate Professor, Business Analytics, University of Southampton (UK); or Wouter Verbeke, Ph.D., Assistant Professor, Business Informatics, University of Brussels (Belgium); or Stefan Lessmann, Ph.D., Professor, School of Business and Economics, Humboldt University (Germany)

A typical organization loses an estimated 5 of its yearly revenue to fraud. This course shows how learning fraud patterns from historical data can be used to fight fraud. The course discusses the use of supervised learning (using a labeled data set), unsupervised learning (using an unlabeled data set), and social network learning (using a networked data set). The techniques can be applied across a wide variety of fraud applications, such as insurance fraud, credit card fraud, anti-money laundering, healthcare fraud, telecommunications fraud, click fraud, tax evasion, and counterfeiting. The course provides a mix of both theoretical and technical insights, as well as practical implementation details. During the course, the instructor reports extensively on his recent research insights about the topic. Various real-life case studies and examples are presented for further clarification.

Learn how to

• Preprocess data for fraud detection (sampling, missing values, outliers, categorization, and so on).
• Build fraud detection models using supervised analytics (logistic regression, decision trees, neural networks, ensemble models, and so on).
• Build fraud detection models using unsupervised analytics (hierarchical clustering, non-hierarchical clustering, k-means, self organizing maps, and so on).
• Build fraud detection models using social network analytics (homophily, featurization, egonets, PageRank, bigraphs, and so on).

Who should attend

Fraud analysts, data miners, and data scientists; consultants working in fraud detection; validators auditing fraud models; and researchers in financial services companies, banks, insurance companies, government institutions, health-care institutions, and consulting firms

Formats available	Standard duration
Classroom:	2 days
Live Web Classroom:	4 half-day session(s)	System Requirements
e-Learning:	21 hours/180 day license	System Requirements

Before attending this course, you should have a basic knowledge of statistics, including descriptive statistics, confidence intervals, and hypothesis testing.

This course addresses SAS Enterprise Miner software.

Base SAS and SAS Social Network Analytics are also used in this course.

Introduction

Fraud Detection

• The importance of fraud detection.
• Defining fraud.
• Anomalous behavior.
• Fraud cycle.
• Types of fraud.
• Examples of insurance fraud and credit card fraud.
• Key characteristics of successful fraud analytics models.
• Fraud detection challenges.
• Approaches to fraud detection.

Data Preprocessing

• Motivation.
• Types of variables.
• Sampling.
• Visual data exploration.
• Missing values.
• Outlier detection and treatment.
• Standardizing data.
• Transforming data.
• Coarse classification and grouping of attributes.
• Recoding categorical variables.
• Segmentation.
• Variable selection.

Supervised Methods for Fraud Detection

• Target definition.
• Linear regression.
• Logistic regression.
• Decision trees.
• Ensemble methods: bagging, boosting, random forests.
• Neural networks.
• Dealing with skewed class distributions.
• Evaluating fraud detection models.

Unsupervised Methods for Fraud Detection

• Unsupervised learning.
• Clustering approaches: hierarchical clustering, k-means clustering, self-organizing maps.
• Peer group analysis.
• Break point analysis.

Social Networks for Fraud Detection

• Social networks and applications.
• Is fraud a social phenomenon?
• Social network components.
• Visualizing social networks.
• Social network metrics.
• Community mining.
• Social-network-based inference (network classifiers and collective inference).
• From unipartite toward bipartite graphs.
• Featurizing a bigraph.
• Fraud propagation.
• Case study.

Fraud Analytics: Putting It All to Work

• Quantitative monitoring: backtesting, benchmarking.
• Qualitative monitoring: data quality, model design, documentation, corporate governance.