Analysis of Encrypted Machine Learning Model Using Fully Homomorphic Encryption and CKKS Scheme

Abstract

As encryption transmission was becoming a common phenomenon, it is important for everyone to ensure their own privacy and data protection. The capacity of Fully Homomorphic Encryption (FHE) to carry out calculations throughout the encoded domain has drawn more attention. Training a machine learning model can be properly outsourced by utilizing FHE. The primary goal of FHE is to make sure performing computation on encrypted files without decoding anything besides the result. CKKS scheme is used due to work with polynomials and it provides a good trade-off between security and efficiency as compared to standard computations on vectors. Network risk and hacking into system a major risk to credentials information because they make it possible for an unauthorized user to retrieve sensitive and important data after analyzing computation results performed on plain data. Thus, this study provides a solution and comparison to the issue of privacy protection incorporating with machine learning in data-driven applications. Our proposed takes a normal dataset as input and provides an encrypted form of the dataset and after that we can perform computation on encrypted form. The used machine learning algorithm is Logistic Regression (LR) to predict the encrypted data. The methodology may maintain accuracy and security of the earlier methods to produce the conclusion, independently of the distributed situation.