A Threshold Multi-CA Authentication Model for Privacy-Preserving Federated Semi-Supervised Learning in Healthcare IoT

Federated semi-supervised learning (FSSL) has emerged as a transformative approach for privacy-preserving personalized healthcare in wearable IoT systems, enabling collaborative model training across distributed edge devices. While existing FSSL frameworks offer collaborative learning capabilities, the reliance on centralized authentication architectures may face challenges in balancing security and operational efficiency, especially under dynamic workloads typical of real-time health monitoring scenarios. To address the challenges, a multi-CA threshold authentication model, which combines established cryptographic and optimization techniques to improve decentralized authentication, is proposed. First, Shamir’s secret sharing is adapted to distribute trust across multiple certificate authorities, reducing dependency on centralized components. Second, grey wolf optimization is employed to dynamically allocate authentication tasks, leveraging its natural-inspired swarm intelligence for load balancing. Third, Levy flight strategies are integrated to enhance search diversity during task scheduling. This combined approach aims to support resource-constrained healthcare IoT environments while maintaining security requirements. Experimental evaluations indicate that the model demonstrates consistent performance in managing authentication workloads and mitigating potential security risks. The model exhibits improved resilience against node compromises without requiring strict threshold configurations. By integrating a secret sharing scheme and adaptive optimization, we reduce authentication latency by 22% compared to centralized CA architectures, demonstrating critical real-time suitability for emergency healthcare scenarios like sudden cardiac event detection, where low-latency authentication is essential for timely interventions.