Transaction Flow Anomaly Detection in Real-Time Payment Networks Through Spatiotemporal Graph Representation Learning

The proliferation of digital payment systems has introduced unprecedented challenges in detecting anomalous transaction patterns within real-time payment networks. Traditional rule-based detection methods struggle to capture the complex spatiotemporal dependencies inherent in modern financial transaction flows. This study proposes a novel framework leveraging Spatiotemporal Graph Representation Learning (STGRL) for transaction flow anomaly detection in real-time payment networks. By modeling payment networks as dynamic graphs where nodes represent accounts and edges encode transaction relationships, our approach integrates Graph Neural Networks with temporal learning mechanisms to capture both spatial topological patterns and temporal evolution dynamics. The proposed methodology employs a multi-layered graph convolutional architecture enhanced with attention mechanisms to learn rich node embeddings that encode transaction behavioral patterns. Experimental evaluation on real-world payment transaction datasets demonstrates that our spatiotemporal graph-based approach achieves superior detection accuracy of 96.8% with significantly reduced false positive rates compared to conventional machine learning baselines. The framework proves particularly effective in identifying coordinated fraud rings and sophisticated money laundering schemes that exhibit complex spatiotemporal patterns across the network.