54 Homomorphic SVM Inference for Fraud Detection

Abstract

Nowadays, cloud computing has become a very promising solution for almost all companies, as it offers the possibility of saving costs by outsourcing computation on-demand. However, some companies deal with private information, which must be protected before outsourcing. Banks, whose financial information is highly sensitive, are one remarkable example of this problem. Their typical processes must be run on their systems for security and regulation reasons, which impedes to take advantage of the scalability and flexibility benefits introduced by the cloud. A relevant example on which we focus in this work is the case of fraud detection systems, for which we propose the use of modern lattice-based homomorphic encryption for its secure execution. To this end, we implement and validate the performance of a homomorphic SVM (Support Vector Machine) classifier for secure fraud detection, showing the feasibility of securely outsourcing fraud detection inference.