GraphALM: Active Learning for Detecting Money Laundering Transactions on Blockchain Networks

In recent years, the decentralization, anonymity, and cross-border capabilities of cryptocurrencies have significantly increased their use in money laundering activities. In an era rigorously regulated by enhanced global anti-money laundering (AML) measures, designing an efficient approach to detect potential money laundering in blockchain is essential. In this research, we present GraphALM, an active learning model based on reinforcement learning, aiming to improve the detection performance of money laundering activities in blockchain transactions. This model addresses the challenge of efficiently sampling training data batches to identify illicit activities within the vast and complex dataset of Bitcoin transactions. Additionally, we have constructed a new Realistic and Demand (RD) Bitcoin dataset, augmented with feature uncertainty, to better simulate real-world scenarios. The results of our experiments demonstrate the effectiveness, robustness, and explainability of our proposed model, contributing to the application of active learning strategies in the field of financial regulation within blockchain networks.