Securing Machine Learning Models: Homomorphic Encryption and its impact on Classifiers

Homomorphic encryption (HME) enables encrypted computations and provides secure data analysis while addressing key concerns around data privacy and regulatory compliance. It has significant implications for machine learning (ML) models, particularly in enhancing the privacy, security, and usability of ML models. Training ML models on homomorphically encrypted data is a growing area of research. While HME offers several security-related benefits to ML models, there are concerns about the ML models performance, when applied over homomorphically encrypted data. Research efforts are needed to analyse the potential impact of HME on the performance of ML models. This paper deals with our work in this direction, presenting the analysis of various ML classifiers on homomorphically encrypted data. We discuss the performance of classifiers such as logisticregression, Support Vector Machines, Neural networks and Decision trees on data encrypted using HME. We follow a systematic approach to analyse the performance in terms of accuracy and efficiency. Our results indicate that the performance of classifiers is nearly identical to the results obtained when they applied on unencrypted data.