Robust Fraud Transaction Detection: A Two-Player Game Approach

Qi Tan

Network and Distributed System Security (NDSS) Symposium 2026 · Day 1 · Applied Cryptography

This talk presents **Gamer**, a novel fraud detection system that models the adversarial interaction between fraudsters and detection systems as a **two-player game**. The core insight is that fraudsters exploit insufficient training data through **feature falsification** -- strategically modifying transaction features to evade AI-based detection models. Rather than using conventional adversarial training or data augmentation, which only locally enhance robustness, Gamer uses **game-theoretic equilibrium** to determine optimal feature selection strategies that account for the cost-profit dynamics of both attackers and defenders.

AI review

A game-theoretic approach to fraud detection robustness that models attacker-defender dynamics through Nash equilibrium feature selection. The causal analysis of why feature falsification works is genuinely insightful, and the 67.5% F1 improvement on real Ant Group data during an active attack campaign gives it credibility. Not an offensive research talk, but the attacker modeling is realistic enough to be interesting.

Watch on YouTube