A Scalable Inductive Graph Neural Network for Money Laundering Detection

Detecting money laundering remains a significant challenge due to the complexity and dynamic nature of financial transaction networks. Graph Neural Networks (GNNs) have shown promise in modeling these networks by capturing hidden patterns and relationships between entities. However, existing GNN-based methods often struggle with scalability and computational efficiency, particularly when applied to large anti-money laundering (AML) datasets. To address these limitations, we propose GraphSAGE+, which is based on GraphSAGE and incorporates a pattern-based oversampling module to generate synthetic money laundering transactions, thereby mitigating the severe class imbalance problem in AML data. GraphSAGE+ achieves detection performance comparable to state-of-the-art models such as PNA (Principal Neighborhood Aggregation) while offering significantly faster training and inference times. These advantages make GraphSAGE+ a practical and scalable solution for real-world AML transaction monitoring.