Scholarlink

<p>This research focuses on enhancing the security of Wireless Body Area Networks (WBANs) in intensive care settings by leveraging advanced deep learning techniques applied to the Medical Information Mart for Intensive Care (MIMIC) dataset. Specific focus is given to two primary classes of medical data: Laboratory Measurements and Imaging Reports. A novel multi-vibrate time series analysis technique is introduced to extract distinctive features unique to each class, enabling a comprehensive understanding of their underlying patterns. The analysis also determines threshold points for distinguishing average data from potential adversarial attacks. These thresholds are seamlessly integrated into a sparse auto-encoder model. Through comprehensive training, the pre-trained sparse auto-encoder demonstrates the ability to accurately predict adversarial attacks in quick time intervals on medical data. Performance evaluation validates the effectiveness of the proposed approach. It achieved an accuracy and macro recall of 96 percentages each, significantly improving over prior methodologies. Likewise, its Mean Absolute Error (MAE) was only 0.007. So, the suggested method contributes to advancing security measures in WBANs, ensuring enhanced patient privacy and data integrity in critical healthcare environments.</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

Enhancing security in WBANs: novel multi-vibrate time series analysis for adversarial attack prediction in intensive care settings

  • 期刊论文

  • 发表日期：

    2025年1月13日

查看全文
  • R.N.L.S. Kalpana,
  • Ajit Kumar Patro,
  • D. Nageshwar Rao

摘要

This research focuses on enhancing the security of Wireless Body Area Networks (WBANs) in intensive care settings by leveraging advanced deep learning techniques applied to the Medical Information Mart for Intensive Care (MIMIC) dataset. Specific focus is given to two primary classes of medical data: Laboratory Measurements and Imaging Reports. A novel multi-vibrate time series analysis technique is introduced to extract distinctive features unique to each class, enabling a comprehensive understanding of their underlying patterns. The analysis also determines threshold points for distinguishing average data from potential adversarial attacks. These thresholds are seamlessly integrated into a sparse auto-encoder model. Through comprehensive training, the pre-trained sparse auto-encoder demonstrates the ability to accurately predict adversarial attacks in quick time intervals on medical data. Performance evaluation validates the effectiveness of the proposed approach. It achieved an accuracy and macro recall of 96 percentages each, significantly improving over prior methodologies. Likewise, its Mean Absolute Error (MAE) was only 0.007. So, the suggested method contributes to advancing security measures in WBANs, ensuring enhanced patient privacy and data integrity in critical healthcare environments.