Scholarlink

<p>The rapid growth of the Internet of Things (IoT) has significantly impacted modern industries by allowing seamless connection and automation. However, this has also contributed to a growing number of sophisticated cyber threats that are taking advantage of IoT’s distributed structure, limited computational power, and varied communication protocols. Current Intrusion Detection Systems (IDS) perform poorly with regard to detecting complex, zero-day, and adversarial attacks mainly due to its generalization deficiency and high false alarm rate. Hence, there is an urgent need for an intelligent, adaptive, and lightweight IDS which is capable of ensuring real-time protection in a dynamic IoT environment. To overcome this challenge, his research introduces a hybrid attention-based deep learning architecture called DAIV-ABID (DenseNet–AlexNet Integrated Vision with Attention-Based Intrusion Detection). The proposed framework consists of Attention-Enhanced AlexNet–BERT for log-based anomaly detection in textual format, and Self-Attention DenseNet–ViT for analyzing traffic patterns in the network spatial and contextual properties. The Dense-Level Feature Fusion process combines multimodal representations. At the same time, the Self-Adaptive Model-Agnostic Meta-Learning (SA-MAML) optimizer enables the model to adapt to changing and previously unseen adversarial threats. The experimental results using benchmark datasets, including the UNSW-NB15 dataset, indicate that the DAIV-ABID algorithm achieves approximately 99% accuracy and very high sensitivity and specificity, while reducing false-positive and false-negative rates compared with RNN, LSTM, GRU, and CNN baselines. These results demonstrate that the proposed model can provide reliable, scalable, and low-latency intrusion detection, which is well-suited to real-time security for IoT applications. Future work will aim to enhance adversarial robustness and to study federated learning for decentralized IoT deployments.</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

Advanced IoT intrusion detection using hybrid attention-based deep learning with DenseNet-ViT and AlexNet-BERT

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月21日

查看全文
  • Sirajudeen Ameer John,
  • Senthilnathan Palaniappan,
  • Gopichand G,
  • Ganesan G

摘要

The rapid growth of the Internet of Things (IoT) has significantly impacted modern industries by allowing seamless connection and automation. However, this has also contributed to a growing number of sophisticated cyber threats that are taking advantage of IoT’s distributed structure, limited computational power, and varied communication protocols. Current Intrusion Detection Systems (IDS) perform poorly with regard to detecting complex, zero-day, and adversarial attacks mainly due to its generalization deficiency and high false alarm rate. Hence, there is an urgent need for an intelligent, adaptive, and lightweight IDS which is capable of ensuring real-time protection in a dynamic IoT environment. To overcome this challenge, his research introduces a hybrid attention-based deep learning architecture called DAIV-ABID (DenseNet–AlexNet Integrated Vision with Attention-Based Intrusion Detection). The proposed framework consists of Attention-Enhanced AlexNet–BERT for log-based anomaly detection in textual format, and Self-Attention DenseNet–ViT for analyzing traffic patterns in the network spatial and contextual properties. The Dense-Level Feature Fusion process combines multimodal representations. At the same time, the Self-Adaptive Model-Agnostic Meta-Learning (SA-MAML) optimizer enables the model to adapt to changing and previously unseen adversarial threats. The experimental results using benchmark datasets, including the UNSW-NB15 dataset, indicate that the DAIV-ABID algorithm achieves approximately 99% accuracy and very high sensitivity and specificity, while reducing false-positive and false-negative rates compared with RNN, LSTM, GRU, and CNN baselines. These results demonstrate that the proposed model can provide reliable, scalable, and low-latency intrusion detection, which is well-suited to real-time security for IoT applications. Future work will aim to enhance adversarial robustness and to study federated learning for decentralized IoT deployments.