Research Paper on Node Anomaly Detection Using Graph Encoder

Anti-money laundering (AML) represents a long-standing data mining challenge within the financial industry. Money laundering (ML) significantly contributes to the operations of transnational and organized crime, which in turn undermines a nation’s economic stability, governance, and social structure. As key facilitators of financial transactions, financial institutions are mandated by governments to assist in identifying and preventing money laundering—a vital measure in combating criminal activity and maintaining economic integrity. AML systems typically rely on user identity information and financial transaction records to detect suspicious behavior. However, there has been a growing trend of money laundering conducted by organized criminal networks, while most existing detection methods primarily examine the actions of individual accounts, overlooking group dynamics. To bridge this gap, this paper presents a deep graph learning framework that incorporates group-level analysis for detecting organized money laundering activities. We introduce a community-based encoder that models user transaction networks to capture collective behavior patterns indicative of criminal organizations. Furthermore, we propose a localized enhancement mechanism that clusters similar transaction behaviors, enabling the identification of criminal syndicates. Our comprehensive evaluation, using real-world data from a leading global bank card provider, demonstrates that the proposed group-aware deep graph learning method significantly outperforms traditional approaches in both batch and real-time scenarios. These results highlight the potential of leveraging group-level insights for more effective money laundering detection.