Scholarlink

<p>The ever-rising deployment of Internet of Things (IoT) applications has thrown new security challenges primarily due to the complexity of network and resource constraints on an edge platform. Conventionally, intrusion detection systems (IDS) face challenges in giving adequate protection to IoT-edge platforms due to constraints on processing powers and the requirement for real-time responses. This research proposes an IoT-edge-tailored advanced intrusion detection system based on deep learning-enhanced ensemble stacking implemented on an NVIDIA Jetson Nano GPU to overcome the above problems. The stacking multi-layer architecture of enhanced IDS incorporates CNN, RNN, DNN, and LSTM models to improve accuracy and enhance resilience against cyber threats. The implementation of the model on an edge GPU demonstrates a system where real-time processing is maintained but with a detection accuracy level that is still high. The proposed study used four diverse datasets involving UNSW-NB15, ToN_IoT, CICIDS2017, and SWaT. It achieved a wide range of high metrics in accuracy (97–99%), precision (94–98%), recall (97–99%), and F1 score (93–99%). The AUC scores are also between 0.93 and 0.99, showing high reliability over different datasets. Experimental results show that the ensemble stacking model outperforms conventional standalone models regarding detection rate and response time, making it appropriate for IoT-edge security applications. This research contributes a scalable, efficient, and resource-aware intrusion-detection approach to extending the resilience of IoT-edge systems in distributed computing environments.</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

Deep learning-based ensemble stacking for enhanced intrusion detection in IoT-edge platforms

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年8月13日

查看全文
  • P. R. Chithra Rani,
  • K. Baalaji

摘要

The ever-rising deployment of Internet of Things (IoT) applications has thrown new security challenges primarily due to the complexity of network and resource constraints on an edge platform. Conventionally, intrusion detection systems (IDS) face challenges in giving adequate protection to IoT-edge platforms due to constraints on processing powers and the requirement for real-time responses. This research proposes an IoT-edge-tailored advanced intrusion detection system based on deep learning-enhanced ensemble stacking implemented on an NVIDIA Jetson Nano GPU to overcome the above problems. The stacking multi-layer architecture of enhanced IDS incorporates CNN, RNN, DNN, and LSTM models to improve accuracy and enhance resilience against cyber threats. The implementation of the model on an edge GPU demonstrates a system where real-time processing is maintained but with a detection accuracy level that is still high. The proposed study used four diverse datasets involving UNSW-NB15, ToN_IoT, CICIDS2017, and SWaT. It achieved a wide range of high metrics in accuracy (97–99%), precision (94–98%), recall (97–99%), and F1 score (93–99%). The AUC scores are also between 0.93 and 0.99, showing high reliability over different datasets. Experimental results show that the ensemble stacking model outperforms conventional standalone models regarding detection rate and response time, making it appropriate for IoT-edge security applications. This research contributes a scalable, efficient, and resource-aware intrusion-detection approach to extending the resilience of IoT-edge systems in distributed computing environments.