Dataset: SAML-D which is a synthetic anti-money laundering dataset.

Money laundering involves converting illicit funds from criminal activity into seemingly legitimate assets, making it difficult to distinguish legal from illegal money. It typically unfolds in three stages: placement, layering, and integration. Transactional data is available during the layering stage, enabling pattern recognition and model-based detection. With global laundering estimated at 2–5% of GDP ($800B–$2T), high-recall detection is critical. We built two deep learning models, Transformer model and Temporal Graph Neural Network, to identify suspicious transactions during this stage.