FraudTransformer: Time-Aware GPT for Transaction Fraud Detection
This work addresses fraud detection for banking systems, but it is incremental as it builds on existing transformer architectures with specific modifications.
The paper tackled the problem of detecting payment fraud in banking streams by introducing FraudTransformer, a GPT-style model enhanced with time and positional encoders, which achieved the highest AUROC and PRAUC on a large industrial dataset compared to classical baselines and transformer ablations.
Detecting payment fraud in real-world banking streams requires models that can exploit both the order of events and the irregular time gaps between them. We introduce FraudTransformer, a sequence model that augments a vanilla GPT-style architecture with (i) a dedicated time encoder that embeds either absolute timestamps or inter-event values, and (ii) a learned positional encoder that preserves relative order. Experiments on a large industrial dataset -- tens of millions of transactions and auxiliary events -- show that FraudTransformer surpasses four strong classical baselines (Logistic Regression, XGBoost and LightGBM) as well as transformer ablations that omit either the time or positional component. On the held-out test set it delivers the highest AUROC and PRAUC.