Anti-Money Laundering (AML) & Compliance: Enhancing Transaction Monitoring and Fraud Detection Using AI Algorithms
Keywords:
Compliance, Financial Fraud Detection, Graph Neural Networks, LSTM, XGBoost, Murex, SAP TRM, Machine Learning, Anti-Money LaunderingAbstract
Traditional rule-based anti-money laundering (AML) systems struggle to detect modern financial fraud exploiting transaction sequences, structures, and behavioural patterns. This study proposes a hybrid framework combining XGBoost, Long Short-Term Memory (LSTM) networks, and Graph Neural Networks (GNNs) for AML detection, evaluated using the IBM AML Transactions dataset with SMOTE oversampling, feature engineering, and graph construction. Results show XGBoost achieves high precision for everyday transactions but detects few fraudulent cases (0.06), while LSTM fails to identify laundering schemes. GNNs demonstrate potential to model transaction structures linked to washing rings and detect patterns like structuring, layering, and circular flows beyond static threshold-based systems. A two-stage method integrating XGBoost filtering with GNN-based clustering of high-risk entities is proposed, providing a unified model assessment and practical implementation guidelines for explainable ensemble-based AML frameworks to enhance enterprise detection workflows.
