Go to AAAI 2025 Workshop AICT homepageCounterfactual Based Probabilistic Graphs for Explainable Money Laundering Detection
Keywords: Counterfactual, Bayesian network, Anti-money laundering, Explainable Artificial Intelligence
TL;DR: Injecting counterfactual samples into the data to generate a probabilistic graph allows for the explanation of black-box money laundering models and the prediction of criminal intent.
Abstract: Anti-money laundering (AML) is a critical challenge for the global financial sector, and deep neural networks have become an essential tool for AML monitoring. However, existing black-box models often lack explainability and fail to provide in-depth analysis of the intent behind behaviors. The method proposed in this paper constructs a Bayesian network for the AML problem by injecting counterfactual examples into the dataset to explain the black-box model through inference. In addition, the method use backward inference to uncover the intent behind anomalous transaction behaviors. Experiments conducted on various AML models and datasets show that our approach provides model-agnostic explanations and can infer the intrinsic intent of money launderers, providing valuable insights for decision-makers.
Submission Number: 8
Loading