Homomorphic cipher for protecting privacy in machine learning
K. Venkata Sravani, A. P. Siva Kumar · 2025
Privacy-preserving machine learning (PPML) is a field that focuses on protecting the privacy and security of data utilized in machine learning models. Several methods, such as Secure Multi-Party Computation, Differential Privacy, Homomorphic Encryption (HE), and Federated Learning, are used to guarantee the confidentiality of data and safeguard user identities. Machine learning (ML) models are crucial across sectors, utilizing sensitive data like medical records and financial information. However, concerns over data privacy necessitate the development of privacy-preserving techniques. This paper presents a way for ensuring privacy in machine learning using HE. HE allows computations to be conducted on encrypted data, assuring the preservation of privacy during the analysis procedure. Our methodology integrates fully HE with deep learning and ML models, such as MLP, simple neural network, and logistic regression. Experiments conducted on heart disease datasets showcase superior accuracy and less evaluation time. Experiment results indicate consistent accuracy between encrypted and plain data models, validating the efficiency of our approach in safeguarding user privacy.