Skip to content

Network Analysis for Fraud

Definition

Network/graph analysis detects fraud by identifying connections between entities (people, devices, addresses, phone numbers) that reveal organized fraud patterns invisible when examining individual accounts.

Entity Graph

graph TD
    A[Person A] -->|same device| B[Person B]
    A -->|same address| C[Person C]
    B -->|same phone| C
    C -->|same email domain| D[Person D]
    D -->|same IP| E[Person E]

    style A fill:#e53935,color:#fff
    style B fill:#e53935,color:#fff
    style C fill:#e53935,color:#fff
    style D fill:#e53935,color:#fff
    style E fill:#e53935,color:#fff

Individually: Each person passes eKYC verification. As a network: Five "unrelated" people sharing devices, addresses, and phones = fraud ring.

Graph Features

Feature What It Reveals
Degree centrality How many connections an entity has (high = hub)
Community detection Clusters of connected entities (fraud rings)
Temporal patterns When connections were formed (bulk creation = suspicious)
Cross-institution links Same entity across multiple financial institutions

Technologies

Tool Type Use
Neo4j Graph database Store and query entity relationships
TigerGraph Graph analytics platform Real-time fraud graph analysis
Amazon Neptune Managed graph DB Cloud-native graph
Graph Neural Networks ML on graphs Learn fraud patterns from graph structure

Key Takeaways

Summary

  • Network analysis catches fraud that individual-level analysis misses — fraud rings, organized crime
  • Entity resolution (linking same person across records) is the foundation
  • Graph databases (Neo4j, TigerGraph) enable real-time relationship querying
  • Cross-institution data sharing dramatically increases graph power