Multi-cloud privacy-preserving logistic regression

Abstract

Clouds can significantly reduce the cost and time of business solutions. However, cloud services introduce significant security and privacy challenges when they process sensitive information. For instance, a dataset for machine learning could contain delicate information that traditional encryption approaches cannot protect during data analysis. Homomorphic Encryption (HE) schemes and secure Multi-Party Computation (MPC) are considered solutions for privacy protection in third-party infrastructures. In this paper, we propose a Multi-Cloud Logistic Regression based on Residue Number System (MC-LR-RNS) that provides security, parallel processing, and scalability. To validate the efficiency and practicability of the solution, we provide its analysis with different configurations, datasets, and cloud service providers. We use six available datasets from medicine (diabetes, cancer, drugs, etc.) and genomics. The analysis shows that MC-LR-RNS provides the same levels of quality as non-HE solutions and improved performance due to multi-cloud parallel computations.