Scholarlink

<p>The Internet of Things (IoT) faces critical security challenges as its rapid expansion exposes networks to escalating cyber threats. Individual IoT devices require labor-intensive management of security updates, but machine learning can effectively leverage the vast data these devices generate. This study introduces a hybrid intrusion detection system (HIDS) to enhance IoT network security by combining deep learning techniques. The proposed HIDS utilizes a CNN-LSTM ensemble with an iterative approach to build a robust classifier. Adaptive synthetic sampling (ADASYN) helps fix the problem of uneven data, while recursive feature elimination improves the choice of features, making the model more effective and accurate. The tests using the KDDCup99 dataset show that the proposed system can accurately detect network attacks, reaching an accuracy of up to 99.21%. The results show that simpler CNN-LSTM models work just as well as more complicated ones, providing good performance while being more efficient. This work illustrates the importance of integrating deep learning-based intrusion detection within cybersecurity frameworks to bolster resilience against sophisticated network threats. The findings also point out the importance of high-quality, balanced data in achieving optimal IDS performance.</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 IoT-fog environments with a hybrid deep learning intrusion detection system

  • 期刊论文

  • 发表日期：

    2025年10月16日

查看全文
  • Suhua Chen,
  • Xu Fang

摘要

The Internet of Things (IoT) faces critical security challenges as its rapid expansion exposes networks to escalating cyber threats. Individual IoT devices require labor-intensive management of security updates, but machine learning can effectively leverage the vast data these devices generate. This study introduces a hybrid intrusion detection system (HIDS) to enhance IoT network security by combining deep learning techniques. The proposed HIDS utilizes a CNN-LSTM ensemble with an iterative approach to build a robust classifier. Adaptive synthetic sampling (ADASYN) helps fix the problem of uneven data, while recursive feature elimination improves the choice of features, making the model more effective and accurate. The tests using the KDDCup99 dataset show that the proposed system can accurately detect network attacks, reaching an accuracy of up to 99.21%. The results show that simpler CNN-LSTM models work just as well as more complicated ones, providing good performance while being more efficient. This work illustrates the importance of integrating deep learning-based intrusion detection within cybersecurity frameworks to bolster resilience against sophisticated network threats. The findings also point out the importance of high-quality, balanced data in achieving optimal IDS performance.