Scholarlink

<p>Internet of Things (IoT) devices have made advancements in security vulnerabilities, so a strong intrusion detection system (IDS) should be used that is able to identify malicious behavior in real-time without compromising user privacy. The paper gives a new hybrid deep learning (DL) architecture, Trans-CNN-Bi-GRU that incorporates Transformer layers, Convolutional Neural Networks (CNN), and Bidirectional Gated Recurrent Units (Bi-GRU) to effectively represent spatial, temporary, and long-range dependencies in the IoT network traffic. Compared to conventional CNNGRU or Transformer CNN architectures, the proposed three-level hybridization gives a better account of the nature of IoT traffic patterns. In addition, the Artificial Bee Optimization (ABO) algorithm is an algorithm that combines feature selection and hyperparameter optimization, which hastens convergence and improves the performance of generalization. Different privacy was added to facilitate privacy during training and inference through Differential Privacy and Homomorphic Encryption. It was shown experimentally that the proposed model reached a state-of-the-art accuracy of 96.78 that was significantly higher than the conventional models SVM, Random Forest, CNN, and LSTM. The privacy-saving protocols led to the minimal trade-off and an accuracy of 95.67 in the case of the differential privacy (= 0.5) and homomorphic encryption. Altogether, the combination of the Trans-CNN-Bi-GRU application with Abo optimization and privacy protection creates a safe and efficient IDS system of the IoT network.</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

ABO optimized hybrid Trans-CNN-Bi-GRU approach for intrusion detection in IoT networks: a privacy-preserving solution

  • 期刊论文

  • 发表日期：

    2025年11月21日

查看全文
  • R. Pavithra Guru,
  • Thomas M. Chen,
  • Mithileysh Sathiyanarayanan

摘要

Internet of Things (IoT) devices have made advancements in security vulnerabilities, so a strong intrusion detection system (IDS) should be used that is able to identify malicious behavior in real-time without compromising user privacy. The paper gives a new hybrid deep learning (DL) architecture, Trans-CNN-Bi-GRU that incorporates Transformer layers, Convolutional Neural Networks (CNN), and Bidirectional Gated Recurrent Units (Bi-GRU) to effectively represent spatial, temporary, and long-range dependencies in the IoT network traffic. Compared to conventional CNNGRU or Transformer CNN architectures, the proposed three-level hybridization gives a better account of the nature of IoT traffic patterns. In addition, the Artificial Bee Optimization (ABO) algorithm is an algorithm that combines feature selection and hyperparameter optimization, which hastens convergence and improves the performance of generalization. Different privacy was added to facilitate privacy during training and inference through Differential Privacy and Homomorphic Encryption. It was shown experimentally that the proposed model reached a state-of-the-art accuracy of 96.78 that was significantly higher than the conventional models SVM, Random Forest, CNN, and LSTM. The privacy-saving protocols led to the minimal trade-off and an accuracy of 95.67 in the case of the differential privacy (= 0.5) and homomorphic encryption. Altogether, the combination of the Trans-CNN-Bi-GRU application with Abo optimization and privacy protection creates a safe and efficient IDS system of the IoT network.