Analysis of Defence Mechanisms for Security Federated Learning

This review systematically examines security mechanisms in Federated Learning (FL), addressing the critical privacy-utility trade-offs in sensitive sectors such as healthcare, finance, and the Internet of Things (IoT). FL enables collaborative model training without raw data sharing, but its susceptibility to attacks like data poisoning and gradient leakage threatens its adoption in privacy-sensitive domains. The study evaluates four primary defence strategies: Secure Aggregation (SA), Differential Privacy (DP), Byzantine-robust aggregation, and Hybrid Frameworks, assessing them from theoretical and practical perspectives. SA utilises cryptographic techniques, such as Advanced Encryption Standard (AES)-256, to ensure privacy, while DP introduces noise to balance privacy protection with model utility. Hybrid methods combine these approaches, though they face challenges related to scalability. The review also identifies current research gaps, including the need for adaptive defences against evolving adversarial tactics and lightweight protocols suitable for resource-constrained IoT environments. By synthesising the strengths and limitations of these defence mechanisms, this study provides a roadmap for developing secure, scalable FL systems, contributing to advancing privacy-preserving technologies in critical sectors.