Federated Machine Learning for Cross-Bank Credit Card Fraud Detection: A Privacy-Preserving Framework

Credit card fraud continues to be a serious global problem. It takes advantage of the insufficient collaboration between the financial institutions that are bound by strict privacy regulations. Fraud detection systems that operate on their own are not very good at identifying the cross-bank fraud patterns that need to be seen to understand and stop the fraud in question. In this article, we present the theory and practice of a new way to detect the fraud that does not require sharing any of the sensitive data that’s now necessary for such systems to work well. We use federated machine learning (FML) to allow two or more banks to co-train a fraud detection system that works better than what either bank by itself could be able to achieve. We integrate differential privacy, secure multi-party computation, and homomorphic encryption into a single framework that ensures that regulatory compliance is as strong as the system itself.