A resource-efficient federated learning framework for intrusion detection in IoMT networks

Abstract

The emergence of medical sensors in Smart Healthcare Systems (SHS) has enabled the development of sophisticated Internet of Medical Things (IoMT) networks which is crucial for tracking vital physiological parameters in consumer electronics. These networks are progressively incorporated into consumer devices allowing users to monitor their health parameters effortlessly. Nonetheless, they encounter considerable security and privacy issues stemming from weaknesses in data transfer. Despite the potential of Intrusion Detection Systems (IDS), there is a critical need for a real-time, highly precise attack detection system optimized for the edge-centric Internet of Medical Things (IoMT) environment in consumer sensor devices. Therefore, we introduce the federated kolmogorov-arnold network (FedIoMT), an advanced federated learning framework that utilizes meta-learning along with an advanced clustering method to achieve robust model aggregation. Our FedIoMT acts as both an adaptive personalized updating mechanism and a shared global classifier. In addition, we propose a novel intrusion detection model that uses the kolmogorov-arnold convolutional network (KANConvNet) as its local classifier, which improves scalability and interpretability in the FL framework updates. Our FedIoMT model demonstrates superior performance compared to other existing FL-based methods, achieving validation accuracies of 99.38%, 99.23%, 99.26%, and 99.2% across four benchmark IoMT datasets under the IID setting. FedIoMT utilizes significantly more floating point operations per second (FLOP) counts up to 2.8 times more than other models and also excels in memory efficiency, consuming up to 93% less peak memory across devices compared to the alternatives. We thoroughly evaluated FedIoMT on embedded consumer devices, demonstrating its resource-efficient characteristics.