Federated Learning-Based Secure and Privacy-Preserving AI Models for Distributed Computing Environments

Abstract

Federated Learning (FL) has become an attractive system that promises to support machine learning collaboration in distributed computing with no data sharing. Nevertheless, FL is susceptible to adversarial attacks like poisoning attacks, model inversion, and communication-based vulnerabilities, even though it is privacy-conscious. In this research, the authors suggest a federated learning system that is secure and privacy-sensitive and combines secure aggregation and differential privacy to increase robustness and confidentiality of the data. The proposed model is examined by the distributed non-IID datasets under the conditions of real-life simulation. The experimental findings prove that the framework attains an accuracy of 93.5 percent, which is very close to the centralized learning performance (94.2 percent) but much better than the standard federated learning in adversarial conditions. Linear scalability is verified by the communication overhead analysis, which has manageable bandwidth needs. The results show that the suggested framework has the potential to balance predictive performance, privacy protection, and resilience to security threats and can be used in critical distributed applications, including healthcare, finance, and IoT ecosystems.