Transaction Fraud Detection Method Based on Event Temporal Graph Attention Network

In recent years, financial crimes such as telecom network fraud have shown highly dynamic characteristics. Criminals implement money laundering through complex fund transfer paths, posing severe challenges to anti-money laundering supervision. Traditional account-based graph construction methods are limited in detecting such abnormal transactions, as they fail to capture the temporal order of transactions and the topological anomalies of fund transfers. To address this, a transaction fraud detection model based on event temporal graph attention network (ETGAT) is proposed. This model constructs graphs with transaction events as nodes, enhances node features through time encoding, and captures the global information of the graph using a Graph Attention Network (GAT) layer. Experimental results show that ETGAT exhibits good performance and interpretability, providing a new solution for transaction fraud detection.