Pinar Merve, Aktas Abdulsamet, Ulku Eyup Emre
Computer Engineering Department, Technology Faculty, Marmara University, Maltepe, Istanbul, Turkey.
Comput Biol Med. 2025 Mar;186:109603. doi: 10.1016/j.compbiomed.2024.109603. Epub 2025 Jan 1.
To address critical security challenges in the Internet of Medical Things (IoMT), this study develops a feature selection framework to improve detection accuracy and computational efficiency in IoMT cybersecurity. By optimizing feature selection, the framework aims to enhance the security and operational integrity of real-time healthcare systems.
This study integrates Random Subset Feature Selection (RSFS) with Correlation Feature Selection (CFS) to create a novel feature selection framework tailored to IoMT datasets. The framework reduces data complexity and filters irrelevant features to improve model performance. It was tested on four IoMT datasets (WUSTL-EHMS-2020, TON-IoT, ICU-Dataset, ECU-IoHT) using machine learning models, including Random Forest, K-Nearest Neighbors, Support Vector Machine, Extreme Gradient Boosting, and a Deep Neural Network.
The proposed framework achieved exceptional results: 99.82% accuracy on the TON-IoT dataset, 99.99% on the ICU-Dataset, 96.37% on the WUSTL-EHMS-2020 dataset, and 99.99% on the ECU-IoHT dataset. These results surpassed existing methods while utilizing a reduced feature set. The framework demonstrated significant improvements in detection accuracy and processing efficiency, addressing high-dimensional data challenges typical of IoMT environments.
This study introduces a robust, scalable feature selection framework for IoMT cybersecurity, providing a practical solution to prevailing security gaps. By ensuring enhanced patient data protection and operational resilience, the framework holds potential for broad implementation in safeguarding critical IoMT infrastructures, advancing the field of secure healthcare systems.
为应对医疗物联网（IoMT）中的关键安全挑战，本研究开发了一个特征选择框架，以提高IoMT网络安全中的检测准确性和计算效率。通过优化特征选择，该框架旨在增强实时医疗系统的安全性和运行完整性。
本研究将随机子集特征选择（RSFS）与相关特征选择（CFS）相结合，创建了一个针对IoMT数据集量身定制的新型特征选择框架。该框架降低了数据复杂性，过滤掉无关特征以提高模型性能。使用包括随机森林、K近邻、支持向量机、极端梯度提升和深度神经网络在内的机器学习模型，在四个IoMT数据集（WUSTL-EHMS-2020、TON-IoT、ICU-Dataset、ECU-IoHT）上对其进行了测试。
所提出的框架取得了优异的成果：在TON-IoT数据集上的准确率为99.82%，在ICU-Dataset上为99.99%，在WUSTL-EHMS-2020数据集上为96.37%，在ECU-IoHT数据集上为99.99%。这些结果在使用减少的特征集的同时超过了现有方法。该框架在检测准确性和处理效率方面有显著提高，解决了IoMT环境中典型的高维数据挑战。
本研究为IoMT网络安全引入了一个强大、可扩展的特征选择框架，为当前的安全漏洞提供了一个切实可行的解决方案。通过确保加强患者数据保护和运营弹性，该框架在保护关键IoMT基础设施、推动安全医疗系统领域发展方面具有广泛实施的潜力。
