Enhancing Anti-Money Laundering Systems Using Knowledge Graphs and Graph Neural Networks

In today's increasingly complex financial landscape, traditional anti-money laundering (AML) systems are often inadequate in combating sophisticated financial crimes. This research aims to bridge that gap by integrating knowledge graphs with graph neural networks (GNNs) to enhance AML detection capabilities. The study leverages financial transactional data to construct a knowledge graph, employing GNN architectures, particularly Graph Attention Networks (GAT), to predict and detect potential money laundering activities. Empirical results demonstrate that GNNs are highly effective at uncovering intricate transaction patterns that conventional methods frequently miss. However, the GAT model encounters issues with generalization and overfitting, especially on larger test datasets. Sensitivity analyses highlight the critical influence of features such as transaction timestamps and payment formats on model performance. This research provides a data-driven, Artificial Intelligence (AI)-enhanced approach to advancing AML systems, offering practical insights for optimizing models and improving detection accuracy. Additionally, the findings present valuable recommendations for financial institutions and regulatory bodies, aiming to enhance compliance and fortify the security of financial markets. Future research will focus on further optimizing these models to address existing challenges and improve generalization.