Generative Pretraining at Scale: Transformer-Based Encoding of Transactional Behavior for Fraud Detection

TL;DR

This paper presents a scalable GPT-based autoregressive model for fraud detection in payment systems, effectively capturing transactional behavior and anomalies without labeled data, enhancing security in large online payment platforms.

Contribution

Introduces a novel GPT-based autoregressive model with differential convolution for improved fraud detection in transactional data, addressing token explosion and sequence reconstruction.

Findings

Model achieves high accuracy in fraud detection

Effective in large-scale online payment systems

Unsupervised pretraining enhances feature representation

Abstract

In this work, we introduce an innovative autoregressive model leveraging Generative Pretrained Transformer (GPT) architectures, tailored for fraud detection in payment systems. Our approach innovatively confronts token explosion and reconstructs behavioral sequences, providing a nuanced understanding of transactional behavior through temporal and contextual analysis. Utilizing unsupervised pretraining, our model excels in feature representation without the need for labeled data. Additionally, we integrate a differential convolutional approach to enhance anomaly detection, bolstering the security and efficacy of one of the largest online payment merchants in China. The scalability and adaptability of our model promise broad applicability in various transactional contexts.