Learning Objectives

• Apply link analysis to fraud detection
• Identify fraud rings and collusion networks
• Use graph algorithms for suspicious pattern discovery
• Graph theory basics
• Week 7: Data science for fraud

1. Link Analysis for Fraud

2. Suspicious Patterns

• Circular Transactions: Money flows in a circle (laundering)
• Star Patterns: One entity connected to many (money mule)
• Bipartite Dense: Many entities share same contact info (shell companies)
• Temporal Patterns: Fast transactions before customer profile updates

3. Graph Algorithms

• Community Detection: Louvain algorithm finds fraud rings
• Shortest Path: Find connections between known fraudsters and new entities
• PageRank: Identify important/influential entities in network
• Centrality Measures: Degree (connections), Betweenness (bridges), Closeness (reachability)
• SimRank: Find similar entities based on network structure

Q1: Why is network analysis useful for fraud detection?

Fraud rarely involves single entity. Networks reveal relationships between fraudsters, shared contact info, circular money flows, and organizing entities (central figures). Q2: What does a star pattern in a transaction network suggest?

One entity receives transactions from many different sources (or sends to many). May indicate a money mule (aggregating funds) or a fraudulent vendor receiving from multiple shell companies. Q3: What is community detection in fraud analysis?

Algorithm (Louvain, Infomap) identifies groups of entities with dense internal connections. Each community represents a potential fraud ring or organized crime group. Q4: How does SimRank help detect synthetic identity fraud?

Identifies accounts with similar attributes (same device, same email pattern, same address) - likely created by same fraudster using synthetic identities. Q5: What is the difference between degree centrality and betweenness centrality?

Degree: number of direct connections (popular entity). Betweenness: how often entity lies on shortest paths between others (gatekeeper/bridge). Betweenness can identify key orchestrators even with few direct connections. Join Discord PreviousData Science for Fraud DetectionNextText Analytics for Forensics