A Comparative Analysis of Machine Learning Models developed from Homomorphic Encryption based RSA and Paillier algorithm

Abstract

Nowadays, the use of Auto ML for development of machine learning (ML) models is increasing day by day. In which users need to upload their dataset to develop machine learning models. Where security of user's data is a very big problem. Which can be solved using Homomorphic encryption. In this work, first the spiderman correlation of the machine learning models developed using encrypted dataset is compared with each other. In which the algorithms used to encrypt dataset are Homomorphic Encryption (HE) based Rivest Shamir Adleman (RSA) and Paillier algorithm. Where the homomorphic encryption based Rivest Shamir Adleman (RSA) is a multiplicative operation and the HE-based Pailler supports additive operations. These algorithms are used to encrypt data and store it on cloud servers. An in-house key generator is developed to generate keys for HE-based RSA algorithm and used this algorithm to encrypt data. Then a comparison between the time taken by these algorithms to encrypt the data and develop the machine learning model using Azure auto Machine Learning is done where the HE-based RSA algorithm is winner.