FSSL-TCN-Per: A Federated Semi-Supervised Personalized Model for Next-Generation Secure Bio-Medical and Healthcare Solutions

The rapid growth of wearable Internet of Things (IoT) devices for continuous health monitoring necessitates robust, low-latency, and privacy-preserving anomaly detection frameworks. Traditional centralized learning approaches pose significant risks to patient privacy and are infeasible for real-time on-device inference due to computational and communication constraints. This paper proposes FSSL-TCN-Per, a novel, real-time, lightweight Federated Semi-Supervised Learning framework incorporating Temporal Convolutional Networks (TCNs) with personalized fine-tuning layers for accurate and resource-efficient anomaly detection. The model leveraging Federated Semi-Supervised Learning (FSSL), personalized adaptation, and lightweight privacy mechanisms including Secure Aggregation and Differential Privacy for real-time health anomaly detection along with pseudo-labeling and entropy minimization to effectively learn from limited labeled data, while a client-specific personalization layer enhances adaptability to individual user profiles. Evaluated on the Secure Healthcare IoT Monitoring Dataset comprising biometric and network-level signals from 2000 patients, the proposed system achieves 99.1% accuracy, 26 ms inference latency, and low power consumption (210 mW). A real-world deployment using a Raspberry Pi and smartwatch emulator validates its edge readiness and privacy-preserving capabilities. The results demonstrate that FSSL-TCN-Per offers a scalable and secure solution for real-time health anomaly detection in wearable healthcare IoT systems.