Abstract—While this connectivity is transforming clinical care, it also presents a massive attack surface, one that is still largely unprotected, which can be exploited with Denial-of-Service (DoS) and Distributed DoS attacks and via reconnaissance, Message Queuing Telemetry (MQTT) protocol abuse, Address Resolution Protocol (ARP) spoofing, and malformed-data injection. The original test and evaluation of the Cloud-Integrated Cloud-of-Things Medical Things 2024 dataset focused only on Central Processing Unit based classifiers with two coarse taxonomies and omitted the fine-grained taxonomy of Wi-Fi/MQTT and the use of Graphics Processing Unit (GPU) based classifiers, which were not evaluated. This paper builds on these gaps by comparing six machine learning classifiers implemented using NVIDIA Rapid Analytics on Platforms in Data Science (RAPIDS) to others that are machine learning—XGBoost with Compute Unified Device Architecture (CUDA) support, and CatBoost with GPU acceleration and evaluating them against 7,160,831 training samples in all fifty-one attack classes. The accuracy, precision, recall, F1-Score, per-sample inference latency, and novel composite metric latency-adjusted accuracy index are used to evaluate performance, which combines the predictive performance with the suitability for real-time deployment in Internet of Medical Thing (IoMT) gateway scenarios. The highest accuracy (83.63%) and per-sample latency (0.0018 ms) are achieved by GPU RF, and CatBoost achieves the best detection per class with 81.24% accuracy and a weighted F1-Score of 0.807. The label-encoding vocabulary mismatch between training and test is a practically relevant failure mode that is shown to collapse XGBoost to near-random performance. The models that are linear and distance-based in nature do not perform well on the 51-class problem, indicating that non-linear ensemble models are required to discriminate between attacks in clinical IoMT environments. Overall, the findings provide a sound basis for the scalable and latency-efficient use of GPU-accelerated ensemble learning for real-time ID in IoMT systems.
 
Keywords—Internet of Medical Thing (IoMT), intrusion detection system, Cloud-Integrated Cloud-of-Things Medical Things (CICIoMT) 2024, Rapid Analytics on Platforms in Data Science (RAPIDS) cuML, CatBoost, random forest, healthcare cybersecurity, latency-aware accuracy index
 
Copyright © 2026 by the authors. This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited (CC BY 4.0).