Federated transfer learning for attack detection for Internet of Medical Things

In the healthcare sector, cyberattack detection systems are crucial for ensuring the privacy of patient data and building trust in the increasingly connected world of medical devices and patient monitoring systems. In light of the increasing prevalence of Internet of Medical Things (IoMT) technologies, it is essential to establish an efficient intrusion detection system (IDS). IDSs are crucial for protecting patient data and ensuring medical device reliability. Federated learning (FL) has emerged as an effective technique for enhancing distributed cyberattack detection systems. By distributing the learning process across multiple IoMT gateways, FL-based IDS offers several benefits, such as improved detection accuracy, reduced network latency, and minimized data leakage. However, as client data may not exhibit a uniform independent and identically distributed (IID) pattern, the heterogeneity of data distribution poses a significant challenge in implementing FL-based IDS for IoMT applications. In this paper, we propose a collaborative learning framework for IDS in IoMT applications. Specifically, we introduce a Federated Transfer Learning (FTL) IDS that enables clients to obtain their personalized FL model while benefiting from the knowledge of other clients. Our methodology enables clients to obtain a personalized model that addresses the challenges posed by the heterogeneity of data distribution. The experimental results show that the proposed model achieves superior detection performance with 95-99% accuracy. Moreover, our model exhibits strong performance in identifying zero-day attacks.