AI-Driven Fraud Detection in Payments Using Transformer-Based Behavioral Modeling

The rapid expansion of digital payments has led to an increase in fraudulent activities, requiring AI-powered fraud detection mechanisms. In this paper, we present a Transformerbased fraud detection framework, leveraging Generative Pretrained Transformers (GPT) for encoding long-term transactional behavior sequences. By pretraining on large-scale payment data, our model reconstructs behavioral patterns and detects anomalies in real-time, improving fraud risk assessment. Our approach integrates unsupervised pretraining, differential convolutional operations, and contrastive learning to enhance anomaly detection without labeled data. The model is evaluated on real-world transactions from one of China’s largest payment platforms (WeChat Pay), demonstrating superior fraud detection accuracy compared to traditional machine learning models. Results show that applying AI-driven behavioral analysis to payment transactions significantly improves the identification of high-risk users and fraudulent activities, reducing false positives while maintaining transaction security. This research highlights the potential of deep learning in financial fraud prevention and presents a scalable solution for real-time fraud detection in payment ecosystems