New Graph Neural Network Enhances Credit Card Fraud Detection

Credit card fraud detection faces significant challenges due to the highly imbalanced nature of fraud data, constantly evolving fraud schemes, and the complex relationships between transaction entities like customers, merchants, and devices. Traditional methods often struggle to adapt to these dynamic conditions. A new framework, the Time-aware Multi-Relational Guided Graph Neural Network (TMR-GGNN), has been developed to address these issues. This model extends the encoder-decoder Graph Neural Network architecture by incorporating heterogeneous interactions across various entities over time. It constructs a dynamic, multi-relational graph and employs a time-aware relational attention mechanism to weigh transaction relevance based on temporal proximity and semantic context. The TMR-GGNN also features a contrastive learning module in its decoder to differentiate between genuine and synthesized transaction patterns, improving its ability to generalize to rare fraud cases. To further combat class imbalance, it uses a composite loss function combining InfoNCE-based contrastive loss with Focal Loss, enhancing fraud identification and reducing false negatives.