Scholarlink

<p>Ensuring security in the Internet of Things (IoT) remains a critical challenge due to the dynamic nature of attacks and the resource constraints of connected devices. Existing approaches, including conventional machine learning and deep learning models, often struggle to balance high detection accuracy with the computational efficiency required for real-time IoT deployment. To address these challenges, this study introduces the Lightweight Forest-Dynamic Proactive Detection and Tracking (LF-DPDT) framework, which integrates a novel Lightweight Random Forest (LRF) classifier within a layered filtering–detection architecture. The proposed LRF incorporates feature-aware subspacing, imbalance-aware bootstrapping, weighted voting, pruning, and early-exit inference, enabling compact yet highly accurate tree ensembles. Experimental evaluation on the N-BaIoT dataset shows that LF-DPDT achieves 100% accuracy, 99.9% recall, and an F1-score of 1.0, while reducing inference time to 0.26&#xa0;s per instance, which is a  &gt; 60% improvement in latency compared to standard RF and over 70% faster than deep learning baselines (CNN, LSTM). These results confirm that LF-DPDT provides a lightweight, real-time, and scalable solution for securing resource-constrained IoT networks against evolving threats such as Mirai and Gafgyt botnets.</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

LF-DPDT: A Learning Framework for Identifying Potential and Dynamic Threats in the Internet of Things

  • 期刊论文

  • 发表日期：

    2026年3月3日

查看全文
  • Sowmya Somanath,
  • Usha Banavikal Ajay

摘要

Ensuring security in the Internet of Things (IoT) remains a critical challenge due to the dynamic nature of attacks and the resource constraints of connected devices. Existing approaches, including conventional machine learning and deep learning models, often struggle to balance high detection accuracy with the computational efficiency required for real-time IoT deployment. To address these challenges, this study introduces the Lightweight Forest-Dynamic Proactive Detection and Tracking (LF-DPDT) framework, which integrates a novel Lightweight Random Forest (LRF) classifier within a layered filtering–detection architecture. The proposed LRF incorporates feature-aware subspacing, imbalance-aware bootstrapping, weighted voting, pruning, and early-exit inference, enabling compact yet highly accurate tree ensembles. Experimental evaluation on the N-BaIoT dataset shows that LF-DPDT achieves 100% accuracy, 99.9% recall, and an F1-score of 1.0, while reducing inference time to 0.26 s per instance, which is a  > 60% improvement in latency compared to standard RF and over 70% faster than deep learning baselines (CNN, LSTM). These results confirm that LF-DPDT provides a lightweight, real-time, and scalable solution for securing resource-constrained IoT networks against evolving threats such as Mirai and Gafgyt botnets.