Dynamic Trust Management for Secure Federated Learning in Critical Industrial and IoT Networks

Federated Learning (FL) has evolved privacy-preserving machine learning by enabling decentralized devices, such as Multi-access Edge Computing (MEC) nodes, to collaboratively train models without sharing raw data. This integration leverages edge computation and storage resources for real-time decision-making, reducing latency and enhancing scalability in critical industrial networks, including domains like healthcare, finance, and IoT. However, FL's decentralized architecture makes it vulnerable to adversarial attacks, such as label flipping, which undermine its sustainability and resilience. These vulnerabilities emphasize the need for adaptive trust management mechanisms.To address these challenges, this paper proposes sensitivity and adaptive mechanisms for trust thresholds and smoothing parameters, enabling real-time adjustments based on client performance, behaviour, and variability. Comparative analyses demonstrate that these adaptive methods significantly enhance robustness, fairness, and scalability, ensuring reliable model aggregation and mitigating the impact of malicious clients. This contribution transitions FL from static to more adaptive frameworks, establishing a benchmark for secure, sustainable, and efficient FL in real-world adversarial environments.