Enhancing Anti-Money Laundering Frameworks: An Application of Graph Neural Networks in Cryptocurrency Transaction Classification

Abstract

Cryptocurrency money laundering is a pressing issue, as it not only facilitates and hides criminal activities but also disrupts markets and the overall financial system. To respond this challenge, researchers are trying to develop robust Anti-Money Laundering (AML) frameworks. These efforts play a crucial role in promoting societal welfare by mitigating the impact of criminal activities. This paper explores the application of Graph Neural Networks (GNNs) for classifying Bitcoin transactions. The research specifically employs Graph Convolutional Networks (GCNs), Graph Attention Networks (GATs), the Chebyshev spatial convolutional neural networks, and GraphSAGE networks. Based on the dataset analysis, we experiment with different subsets of features. Our findings suggest that the use of Graph Neural Network convolutions, combined with a final linear layer and skip connections, allow for an improvement in the state-of-the-art results, especially when Chebyshev and GATv2 convolutions are used.