BitSentry: a graph-based approach for tracking illicit transactions in bitcoin

Abstract

Detecting illicit transactions in financial networks is crucial to modern anti-money laundering (AML) e↵orts, particularly within decentralized systems like Bitcoin. Inherent anonymity and obfuscation strategies make tracing unlawful activities significantly more challenging. Traditional rule-based detection mechanisms and standard machine learning techniques often fail to generalize well against evolving patterns of money laundering, which are deliberately engineered to circumvent known safeguards. Recent advances in Graph Neural Networks (GNNs) have shown substantial promise in fraud detection, given their ability to capture structural and relational intricacies in transaction networks. Through graph restructuring, frameworks such as FraudLens [ACSAC 23] have addressed issues like label imbalance and topological bias. However, these approaches often overlook deeper graph-theoretic insights that can further refine the quality of input graphs before GNN processing.