TravelFraudBench: Configurable GNN Fraud Detection Benchmark for Travel Networks
TravelFraudBench (TFG) is a unique benchmark created to evaluate graph neural networks (GNNs) specifically for spotting fraud rings in travel-related data. Unlike other benchmarks like YelpChi or Amazon-Fraud, which tend to focus on single types of nodes or broad patterns, TFG offers tools to analyze different types of fraud ring structures. It simulates three travel fraud scenarios: ticketing fraud, ghost hotels, and account takeovers. The benchmark includes a complex graph with 9 types of nodes and 12 types of edges, allowing full customization of parameters like ring size and fraud rate. Six methods are tested, with GraphSAGE achieving the best performance based on AU metrics.
Key facts
- TravelFraudBench (TFG) is a configurable benchmark for GNN fraud ring detection in travel networks.
- Existing benchmarks include YelpChi, Amazon-Fraud, Elliptic, and PaySim.
- TFG simulates three travel-specific fraud ring types: ticketing fraud, ghost hotel schemes, and account takeover rings.
- The graph has 9 node types and 12 edge types.
- Ring size, count, fraud rate, scale (500 to 200,000 nodes), and composition are configurable.
- Six methods evaluated: MLP, GraphSAGE, RGCN-proj, HAN, RGCN, and PC-GNN.
- Ring-based split eliminates transductive label leakage.
- GraphSAGE achieves the best AU performance.
Entities
—