A Federated Learning Framework for Privacy-Aware Intrusion Detection (PAID) for Internet of Medical Things

Abstract

The Internet of Medical Things (IoMT) has revolutionised healthcare services, enabling real-time monitoring, remote diagnosis, and intelligent patient care. With the growing connectivity of medical devices, however, the healthcare network is vulnerable to several cyber-attacks, such as Distributed Denial of Service (DDoS), spoofing, botnet attacks, and unauthorised access attacks. So far, traditional intrusion detection systems (IDSs) have been based on centralised learning methods that involve the transfer of sensitive healthcare data to a central server, creating privacy and security issues. This paper presents a Federated Learning (FL) based Privacy-Aware Intrusion Detection (PAID) framework for secure IoMT environments to overcome these challenges. The proposed framework facilitates the sharing of a distributed IDS model that is trained collaboratively by distributed IoMT devices and healthcare gateways without sharing raw network traffic and patient information. To collect the locally trained model parameters and build a powerful global intrusion detection model, a Federated Averaging (FedAvg) mechanism is used. The framework is tested with the CIC IoT-DIAD 2024 dataset with binary and multi-class intrusion detection models. Results of the experiments show that the proposed PAID model achieves better detection results, with 97.9% accuracy for binary classification and 95.6% accuracy for multi-class classification, achieving higher accuracy than the conventional machine learning and centralised deep learning approaches, and maintaining data privacy and scalability in distributed healthcare networks.