Privacy-Preserving Machine Learning Models for Transaction Monitoring in Global Banking Networks

Abstract

Monitoring transaction flows in real time is crucial for global banks to detect and mitigate money laundering, terrorist financing, fraud, and other illicit activities. Although traditional systems rely on local data troves, privacy laws in many jurisdictions limit the use of such data outside their borders. Privacy-by-design, data minimization, explainability, and auditability are some of the expectations from global financial regulators that drive global banks to explore privacy-compliant solutions using advanced technologies, complying not only with the financial but also with the privacy regulators. Recent progress in privacy-preserving machine-learning (ML) models enables banks to preserve the privacy of customers and transactions and help the security and compliance teams reduce risk and improve detection capabilities. These techniques have just begun to be applied in financial transaction monitoring and can be of significant help when embedded in the monitoring models and significant efforts are applied to reduce their limitations. Thus, core privacy-preserving techniques used in the development of the ML monitoring models cover federated learning over banking data and secure computation and homomorphic encryption.