Scholarlink

<p>As Internet of Things (IoT) networks continue to grow quickly, the need for powerful security mechanisms to identify and reduce potential cyber threats has become increasingly critical. Although IoT devices and services offer significant convenience and connectivity, their inherent vulnerabilities expose them to serious security breaches. In this context, network intrusion detection systems (IDS) play a vital role in safeguarding IoT environments. To overcome these challenges, this study proposes a Distributed and Decentralized Learning framework (DISTRUD) for intrusion detection in large-scale IoT networks. The framework introduces an extended loss function for high-dimensional data with significant variability and complex network traffic flows. In addition, the framework includes formal proofs of the non-negativity, differentiability, and boundedness of the distributed and decentralized learning loss function which ensures mathematical consistency and stability during training. The method is implemented in two phases. Initial training minimizing the distributed and decentralized learning loss function to accurately capture the underlying structure of network traffic while remaining resilient to noise and outliers are focused on Phase 1. The model undergoes refinement using reweighted instances with emphasis on difficult-to-classify samples to enhance detection accuracy in Phase 2. The experimental outcomes demonstrate that the DISTRUD model achieves detection accuracy of 96.65% in Phase 1 and 98.82% in Phase 2, while reducing the false positive rate (FPR) to 4.76% and 2.55%, respectively. The results suggest that the DISTRUD framework provides effective, scalable, and reliable intrusion detection in large-scale IoT network 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

Enhancing IoT Intrusion Detection with Distributed Learning: A Novel Framework for Large-Scale Networks

  • 期刊论文

  • 发表日期：

    2026年4月21日

查看全文
  • Raghu Raman Duraiswamy,
  • Palani Uthirapathy

摘要

As Internet of Things (IoT) networks continue to grow quickly, the need for powerful security mechanisms to identify and reduce potential cyber threats has become increasingly critical. Although IoT devices and services offer significant convenience and connectivity, their inherent vulnerabilities expose them to serious security breaches. In this context, network intrusion detection systems (IDS) play a vital role in safeguarding IoT environments. To overcome these challenges, this study proposes a Distributed and Decentralized Learning framework (DISTRUD) for intrusion detection in large-scale IoT networks. The framework introduces an extended loss function for high-dimensional data with significant variability and complex network traffic flows. In addition, the framework includes formal proofs of the non-negativity, differentiability, and boundedness of the distributed and decentralized learning loss function which ensures mathematical consistency and stability during training. The method is implemented in two phases. Initial training minimizing the distributed and decentralized learning loss function to accurately capture the underlying structure of network traffic while remaining resilient to noise and outliers are focused on Phase 1. The model undergoes refinement using reweighted instances with emphasis on difficult-to-classify samples to enhance detection accuracy in Phase 2. The experimental outcomes demonstrate that the DISTRUD model achieves detection accuracy of 96.65% in Phase 1 and 98.82% in Phase 2, while reducing the false positive rate (FPR) to 4.76% and 2.55%, respectively. The results suggest that the DISTRUD framework provides effective, scalable, and reliable intrusion detection in large-scale IoT network environments.