Privacy-Preserving Logistic Regression as a Cloud Service Based on Residue Number System

Abstract

The security of data storage, transmission, and processing is emerging as an important consideration in many data analytics techniques and technologies. For instance, in machine learning, the datasets could contain sensitive information that cannot be protected by traditional encryption approaches. Homomorphic encryption schemes and secure multi-party computation are considered as a solution for privacy protection. In this paper, we propose a homomorphic Logistic Regression based on Residue Number System (LR-RNS) that provides security, parallel processing, scalability, error detection, and correction. We verify it using six known datasets from medicine (diabetes, cancer, drugs, etc.) and genomics. We provide experimental analysis with 30 configurations for each dataset to compare the performance and quality of our solution with the state of the art algorithms. For a fair comparison, we use the same 5-fold cross-validation technique. The results show that LR-RNS demonstrates similar accuracy and performance of the classification algorithm at various thresholds settings but with the reduction of training time from 85.9% to 96.1%.