A Deep Feature Ensemble Framework for Intrusion Detection in Internet of Medical Things
Chitty Avula, Sathyanarayana Bachala · Engineering Technology & Applied Science Research · 2025
The Internet of Medical Things (IoMT) has transformed healthcare through real-time monitoring, remote diagnostics, and intelligent analytics. However, the rapid proliferation of connected medical devices introduces significant cybersecurity threats that endanger patient safety and data privacy. To address these challenges, this study proposes SIDHELNet (Semantically Improved Deep feature ensemble-driven Heterogeneous Ensemble Learning Network), a novel intrusion detection framework tailored for the IoMT environment. SIDHELNet introduces a novel fusion of semantic Word2Vec embeddings with parallel Bi-LSTM and Bi-GRU layers to extract enriched temporal-spatial features, followed by a nine-classifier heterogeneous ensemble. This design distinguishes SIDHELNet from prior hybrid IDS by enabling high adaptability and robustness across heterogeneous IoMT traffic. These features are classified using a Heterogeneous Ensemble Learning (HEL) model that fuses predictions from multiple learners to improve accuracy and robustness. Experimental evaluations on the WUSDL-EHMS-2020 and UNSW-NB15 datasets show that SIDHELNet achieves a detection accuracy of 99.61%, precision of 99.58%, recall of 99.62%, F1-score of 99.59%, and an AUC of 0.998, outperforming existing methods and validating SIDHELNet's high reliability, generalizability, and real-time capability.