Benchmark RadarFinFraudBench
FinFraudBench is a heterogeneous graph benchmark for financial fraud detection. It contains two datasets (CreditCard-Fraud and BankTrans-Fraud) with up to 8.99M nodes and 89.23M directed typed edges, preserving six financial entity types and fourteen edge types. The evaluation protocol covers ranking and imbalance-sensitive classification metrics.
- Released
- 2026-08-15
- Readiness
- Inspectable
- Primary field
- Finance & Economics
Why it matters
Existing graph-based fraud detection benchmarks often oversimplify financial systems and lack realistic conditions. FinFraudBench provides large-scale heterogeneous graphs with natural fraud rates, enabling more realistic evaluation of fraud detection methods and comparison across models.
Motivation
The increasing complexity of digital financial systems has reshaped financial fraud detection from isolated transaction classification into relational risk reasoning over interconnected financial entities.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.