Advancing IoMT Security with Privacy-Preserving Federated Learning Techniques

The rapid expansion of the Internet of Medical Things (IoM<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$T$</tex>) has improved the healthcare system by enabling real-time patient monitoring, diagnosis, and treatment through connected devices. However, these advancements introduce secu-rity risks, making Intrusion Detection Systems (IDS) essential for detecting and mitigating cyber threats. Traditional IDS solutions rely on centralized machine learning (ML) models, which require transferring sensitive patient data to a central server. This raises privacy concerns and poses challenges in handling large-scale data efficiently. To address these issues, this paper proposes a novel Federated Learning (FL) framework that enhances IoMT security while preserving data privacy. Our approach integrates Local Differential Privacy (LDP) to protect individual data points and employs a Neural Oblivious Decision Ensembles (NODE) model, optimized for tabular IoMT data. We utilize Federated Averaging (FedAvg) as the primary aggregation algorithm and benchmark it against FedAvgM, FedAdam, and FedAdagrad using the Flower FL framework. The framework was evaluated on state-of-the-art datasets for real-world IoMT attack scenarios: the IoMT-TrafficData dataset was used for training, and the CICIoMT2024 dataset was used for inference, facilitating the multiclass classification of various attack types. Our results indicate that our approach achieves high efficiency and strong generalization capabilities across diverse IoM<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$T$</tex> attack scenarios.