Scholarlink

<p>Accurately identifying cyber attacks within the Internet of Medical Things (IoMT) is essential to safeguard patient safety, protect sensitive medical data, ensure regulatory compliance, and preserve trust in healthcare systems. This research presents ensemble learning techniques aimed at detecting a diverse range of cyberattacks in IoMT networks. We implement three ensemble models, StackRF, StackMean, and StackMax, to improve detection accuracy beyond what individual deep learning models can achieve. Notably, StackMean computes the average predicted class probabilities, StackMax focuses on the highest predicted class probabilities for more definitive outcomes, and StackRF aggregates the outputs of base learners using a random forest approach. Theoretical analysis shows that the proposed stacked model can improve detection results compared with standalone deep learning methods. Moreover, these stacked models offer increased robustness against adversarial attacks by diminishing dependence on specific neural network structures. Due to the lack of samples for cyber-attacks, we employ the synthetic minority oversampling technique (SMOTE) algorithm to generate synthetic samples and balance the data. In addition, we utilize the kernel principal component analysis (KPCA) to extract nonlinear factors embedded in the complex IoMT data. These techniques improve the detection accuracy of our proposed methodology. The efficacy of the methods is assessed using the modern IoMT dataset, WUSTL-EHMS-2020, revealing the superior performance of the stacked models over the standalone ones. Furthermore, statistical testing is conducted to compare the accuracy of individual models against that of the stacked models, illustrating the enhanced accuracy of the stacked approach. Moreover, it is demonstrated that incorporating SMOTE into the stacked deep learning methods yields superior sensitivity and AUC compared to the stacked models without SMOTE.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Improving cyber-attack detection in Internet of Medical Things using ensemble deep learning methods

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年9月29日

查看全文
  • Wang Wu,
  • Harrou Fouzi,
  • Senouci Sidi-Mohammed,
  • Sun Ying

摘要

Accurately identifying cyber attacks within the Internet of Medical Things (IoMT) is essential to safeguard patient safety, protect sensitive medical data, ensure regulatory compliance, and preserve trust in healthcare systems. This research presents ensemble learning techniques aimed at detecting a diverse range of cyberattacks in IoMT networks. We implement three ensemble models, StackRF, StackMean, and StackMax, to improve detection accuracy beyond what individual deep learning models can achieve. Notably, StackMean computes the average predicted class probabilities, StackMax focuses on the highest predicted class probabilities for more definitive outcomes, and StackRF aggregates the outputs of base learners using a random forest approach. Theoretical analysis shows that the proposed stacked model can improve detection results compared with standalone deep learning methods. Moreover, these stacked models offer increased robustness against adversarial attacks by diminishing dependence on specific neural network structures. Due to the lack of samples for cyber-attacks, we employ the synthetic minority oversampling technique (SMOTE) algorithm to generate synthetic samples and balance the data. In addition, we utilize the kernel principal component analysis (KPCA) to extract nonlinear factors embedded in the complex IoMT data. These techniques improve the detection accuracy of our proposed methodology. The efficacy of the methods is assessed using the modern IoMT dataset, WUSTL-EHMS-2020, revealing the superior performance of the stacked models over the standalone ones. Furthermore, statistical testing is conducted to compare the accuracy of individual models against that of the stacked models, illustrating the enhanced accuracy of the stacked approach. Moreover, it is demonstrated that incorporating SMOTE into the stacked deep learning methods yields superior sensitivity and AUC compared to the stacked models without SMOTE.