Scholarlink

<p>Identification of vulnerabilities and threats from a real IoT network that generates a huge amount of unbalanced and high-dimensional data is still a challenging research problem. As of now, network security professionals have developed various Deep Learning-based intrusion detection systems incorporating only single-view approaches, which may yield poor performance on IoT network data. To address this issue, we have developed a novel Multi-view Learning (MVL)-based intrusion detection system, named <InlineEquation ID="IEq3"> <EquationSource Format="TEX">\(MV^2AE\)</EquationSource> </InlineEquation>, incorporating two Autoencoder (AE) models, viz., Vanilla AE and Variational AE. The novelty of the <InlineEquation ID="IEq4"> <EquationSource Format="TEX">\(MV^2AE\)</EquationSource> </InlineEquation> method is the integration of the two AE models, which enables the extraction of compressed spatial features and new probabilistic-based samples. This makes the <InlineEquation ID="IEq5"> <EquationSource Format="TEX">\(MV^2AE\)</EquationSource> </InlineEquation> method robust in handling the imbalance issue. The experimental results show that the proposed method yields more than 99.23% accuracy on six IoT network intrusion datasets, viz., Edge-IIoTset, UNSW-BoTIoT, X-IIoTID, N-BaIoT, CICIoT2023, and WUSTL-IIOT-2021 datasets, which reflects a comparatively better or similar performance with the state-of-the-art methods. The implementation details of the <InlineEquation ID="IEq6"> <EquationSource Format="TEX">\(MV^2AE\)</EquationSource> </InlineEquation> method can be accessed at <a href="https://github.com/3Clinton/MV2AE">https://github.com/3Clinton/MV2AE</a>.</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

\(MV^{2}AE\): multi-view learning with feature extraction using autoencoders to identify IoT network attacks

  • 期刊论文

  • 发表日期：

    2025年10月25日

查看全文
  • Yarin Vashum,
  • Boby Clinton Urikhimbam,
  • Nazrul Hoque,
  • Ibotombi Singh Sarangthem,
  • Dhruba K. Bhattacharyya

摘要

Identification of vulnerabilities and threats from a real IoT network that generates a huge amount of unbalanced and high-dimensional data is still a challenging research problem. As of now, network security professionals have developed various Deep Learning-based intrusion detection systems incorporating only single-view approaches, which may yield poor performance on IoT network data. To address this issue, we have developed a novel Multi-view Learning (MVL)-based intrusion detection system, named \(MV^2AE\) , incorporating two Autoencoder (AE) models, viz., Vanilla AE and Variational AE. The novelty of the \(MV^2AE\) method is the integration of the two AE models, which enables the extraction of compressed spatial features and new probabilistic-based samples. This makes the \(MV^2AE\) method robust in handling the imbalance issue. The experimental results show that the proposed method yields more than 99.23% accuracy on six IoT network intrusion datasets, viz., Edge-IIoTset, UNSW-BoTIoT, X-IIoTID, N-BaIoT, CICIoT2023, and WUSTL-IIOT-2021 datasets, which reflects a comparatively better or similar performance with the state-of-the-art methods. The implementation details of the \(MV^2AE\) method can be accessed at https://github.com/3Clinton/MV2AE.