HybridIoMT: A Dual-Phase Machine Learning Framework for Robust Cybersecurity in Internet of Medical Things

The Internet of Medical Things (IoMT) changes health care by connecting devices and sharing information in ways no one has seen before.Nonetheless, this connectivity exposes massive cybersecurity vulnerabilities that jeopardize patient safety and integrity of information.With new versions of cyber-attacks being developed to breach IoMT environments, current Intrusion Detection Systems (IDSs) are unable recognize and differentiate between these latest threats, thus demanding a different approach.In this work, we present a novel hybrid machine learning model for better supervised IoMT cybersecurity supporting the detection and classification of related anomalies within data streams that are real-time and heterogeneous.We use the WUSTL-EHMS-2020 dataset to provide input to our method which uses network flow measurements and patient biometric data combining unsupervised and supervised learning approaches.We employ a state-of-the-art method with an autoencoder followed by XGBoosting to learn the general patterns and conceptual dependencies of semantic impact.The data with anomalies are now detected based on the autoencoder and the low dimensional, rich representation of the samples is passed to XGBoost so it could classify between classes of attacks.This symbiosis between unsupervised feature learning and supervised classification has the potential to potentially ameliorate various constraints of current IDS paradigms, hereby proposing healthy ground for identifying both common and exotic threats in a shifting plane of IoMT cybersecurity.The accuracy has increased from 0.9378 to 0.9456, and the F1-score improved from 0.9416 to 0.9478.These findings suggest improved detection and classification of cyber threats in IoMT landscapes.