Tab-AML: A Transformer Based Transaction Monitoring Model for Anti-Money Laundering

Money laundering signifies a major challenge and risk in the global economic landscape. We introduce TabAML, a deep-learning model for transaction monitoring that achieves high detection accuracy while significantly reducing false positives. Incorporating a dual-masked Transformer encoder with a shared embedding component and Residual Attention Layers, Tab-AML achieved an ROC-AUC of 93.01, a 98% true positive rate, and a 51% false positive rate on the SAMLD dataset, outperforming models such as TabTransformer and XGBoost. These findings highlight the potential of transformer-based models in advancing anti-money laundering efforts.