PRIVACY PRESERVING AND SECURE MACHINE LEARNING

Abstract

Privacy-preserving and secure machine learning is increasingly important because applications process sensitive personal, financial, healthcare, and organizational data. Centralized data collection creates risks such as data leakage, unauthorized access, inference attacks, and privacy violations. This work proposes a Privacy-Preserving and Secure Machine Learning framework that combines secure data preprocessing, feature selection, Federated Learning, Differential Privacy, Homomorphic Encryption, Secure Aggregation, authentication, and access control. In the proposed framework, data remains on local client devices, while model updates are shared with a server for collaborative training. Differential Privacy reduces the risk of revealing records, while Homomorphic Encryption protects model parameters during transmission and computation. Secure Aggregation prevents exposure of client updates. A Deep Neural Network is used as the prediction engine. The framework is evaluated using accuracy, precision, recall, F1-score, privacy protection, communication cost, and execution time. The approach aims to provide scalable, accurate, and trustworthy machine learning for sensitive applications. Keywords: Machine Learning, Federated Learning, Deep Neural Network, Data Security, Data Privacy, Authentication, Access Control.