Federated Learning-Based Privacy-Preserving Analytics for Large-Scale Cloud Platforms

Abstract

The rapid growth of cloud-based data analytics has raised significant concerns regarding data privacy, security, and centralized data ownership. To address these challenges, this paper proposes a federated learning-based framework for privacy-preserving analytics in large-scale cloud platforms. The proposed approach enables multiple distributed clients to collaboratively train machine learning models without sharing raw data, thereby ensuring data confidentiality and regulatory compliance. A secure aggregation mechanism is incorporated to protect intermediate model updates, while communication-efficient optimization techniques are employed to reduce overhead in large-scale environments. The framework is evaluated under realistic cloud settings, demonstrating its effectiveness in maintaining high model accuracy while significantly enhancing data privacy. Experimental results show that the proposed method achieves competitive performance compared to centralized approaches, with reduced risk of data leakage and improved scalability. The study highlights the potential of federated learning as a viable solution for secure and distributed analytics in modern cloud ecosystems.