Scholarlink

The experimental growth of Internet of Things (IoT) devices introduces security challenges, notably in terms of malware targeting resource-constrained environments. Traditional malware detection techniques are increasingly inadequate in confronting IoT malware. This limitation necessitates the development of adaptive and interpretable malware detection framework. This study proposes a deep learning (DL)-based hybrid feature extraction enabled malware detection architecture for an optimal IoT malware detection and classification. It leverages temporal convolutional neural networks (TCNNs)-Bi-directional long-short term memory (Bi-LSTM) and MobileNet V2-Linformer driven feature extraction, an attention-based feature fusion, and a fine-tuned and interpretable extremely randomized tree (ERT) IoT malware classifier. The experiment validation is conduced on a curated dataset comprises 4000 raw IoT malware binaries. The proposed model outperforms the state-of-the-art malware detection models by achieving a generalization accuracy of 99.17% with computational footprints of 10.8 million parameters and 1.5 Giga floating point operations. The study findings underscore the potential of proposed model in offering reliable, interpretable, and scalable solution to contemporary malware threats. It provides a valuable platform for future security research in edge computing environments.

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

Explainable Deep Learning-Based Internet of Things Malware Detection Model

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Abdul Rahaman Wahab Sait

摘要

The experimental growth of Internet of Things (IoT) devices introduces security challenges, notably in terms of malware targeting resource-constrained environments. Traditional malware detection techniques are increasingly inadequate in confronting IoT malware. This limitation necessitates the development of adaptive and interpretable malware detection framework. This study proposes a deep learning (DL)-based hybrid feature extraction enabled malware detection architecture for an optimal IoT malware detection and classification. It leverages temporal convolutional neural networks (TCNNs)-Bi-directional long-short term memory (Bi-LSTM) and MobileNet V2-Linformer driven feature extraction, an attention-based feature fusion, and a fine-tuned and interpretable extremely randomized tree (ERT) IoT malware classifier. The experiment validation is conduced on a curated dataset comprises 4000 raw IoT malware binaries. The proposed model outperforms the state-of-the-art malware detection models by achieving a generalization accuracy of 99.17% with computational footprints of 10.8 million parameters and 1.5 Giga floating point operations. The study findings underscore the potential of proposed model in offering reliable, interpretable, and scalable solution to contemporary malware threats. It provides a valuable platform for future security research in edge computing environments.