behavioral patterns

Synthetic Tabular Generators Fail to Preserve Behavioral Fraud Patterns: A Benchmark on Temporal, Velocity, and Multi-Account Signals

This research introduces "behavioral fidelity" as a new evaluation dimension for synthetic tabular data, measuring whether generated data preserves temporal and structural behavioral patterns critical for fraud detection. It proves that dominant row-independent generators are inherently incapable of reproducing complex multi-account fraud graph motifs.