Federated Learning-Driven Cyberattack Detection Framework for Scalable Smart Healthcare IoT Systems

Abstract

Smart Healthcare Internet of Things (IoT) technologies have revolutionized the healthcare sector by providing facilities to monitor patients intelligently, diagnose remotely, wearable medical systems, telemedicine platforms, and communication of healthcare data in real time. All these developments notwithstanding, the smart healthcare IoT environments are still highly susceptible to cyberattacks because of resource constrained medical devices, different communication protocols, distributed healthcare environments and the sharing of highly sensitive patient data. Traditional, centralized cybersecurity solutions are frequently not scalable, don't offer adaptive protection against changing healthcare cyber threats, nor are they privacy-focused or compliant with regulations, while at the same time maintaining the confidentiality of patient data. Besides, the centralisation of training mechanisms involves transmitting private healthcare information, presenting serious privacy, communication and operational issues. In this paper, the authors present a new Federated Learning-Driven Cyberattack Detection Framework (FL-CDF) for scalable smart healthcare IoT systems. The proposed framework combines federated learning-based decentralized training, intelligent healthcare traffic monitoring, adaptive cyberattack classification, the edge's security analysis, and Explainable Artificial Intelligence (XAI) mechanisms for privacy-preserving healthcare cybersecurity management. Distributed healthcare IoT nodes that include wearable devices, medical sensors, smart monitoring systems, and hospital gateways jointly train a global model of intrusion detection by not sharing raw patient information. A hybrid of the Temporal Convolutional Network (TCN) and Attention-Based Bidirectional Long Short-Term Memory (ABiLSTM) network is designed for sequential healthcare traffic analysis and adaptive cyberattack detection. A proposed hybrid Temporal Convolutional Network (TCN) and Attention-Based Bidirectional Long Short-Term Memory (ABiLSTM) network for sequential healthcare traffic analysis and adaptive cyberattack detection. Explainability modules present a clear interpretation of intrusion decisions, determining which features are most important in healthcare communication, and which factors contribute to attacks. Experimental evaluation shows that the proposed framework can outperform the conventional centralized cybersecurity systems in terms of intrusion detection accuracy, false alarm count, preservation of privacy, capability of scalability and communication overhead. This proposed framework helps create safe and sustainable smart healthcare infrastructures, facilitating decentralized healthcare cybersecurity intelligence, communication protection via privacy awareness, resilient medical IoT ecosystems, and trustworthy digital healthcare environments.