Powerful Graph Neural Networks for Money Laundering Pattern Detection

Abstract

This research enhances graph neural networks (GNN) for identifying suspicious accounts involved in money laundering. Extending the work of Egressy et al. (2024), we modify and propose a set of techniques that enable GNNs to detect suspicious subgraph patterns in the weighted temporal networks underlying financial data. These techniques include a novel message passing scheme, which allows for the indication of edge directionality within a single aggregator function, element-wise edge weight multiplication, and an LSTM aggregator that can learn from the sequential order of edges imposed by timestamps. Our LAS-GNN model is based on an inductive learning framework and can generalize across different networks. Experimental results on synthetic networks show that LAS-GNN is robust and can identify basic money laundering patterns, such as smurfing motifs and simple cycles up to length 6, to near perfection, outperforming a graph isomorphism network benchmark with edge features. While the primary focus is on the financial crime domain, the findings have broader applications in dynamic network settings.