Scholarlink

<p>The Internet of Things (IoT) has significantly raised vulnerability to cyber threats, highlighting the need for effective and scalable Intrusion Detection Systems (IDS). This research presents a hybrid framework that combines Principal Component Analysis (PCA) with an optimized ensemble Random Forest classifier to improve intrusion detection accuracy in high-dimensional and imbalanced IDS datasets. To assess the performance of the framework, a systematic evaluation was made using three benchmark datasets—NSL-KDD, UNSW-NB15, and CICIDS2017. Comprehensive experiments indicate that PCA improves feature selection and model generalization, whereas Random Forest offers significant interpretability and robustness. The proposed method attains a peak accuracy of 99.88% on the NSL-KDD dataset and demonstrates competitive performance across other datasets. A comparative analysis with recent state-of-the-art models indicates that the framework consistently outperforms or matches deep learning-based methods. Although deep learning models like Transformers and attention-based architectures yield promising results in IDS research, they are computationally expensive and less appropriate for real-time IoT implementations. Our methodology offers a lightweight but efficient alternative, with future research directions aimed at incorporating advanced deep learning models with adaptive learning mechanisms.</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

Securing the Internet of Things Through Intrusion Detection System Utilizing Machine Ensemble Learning and Feature Extraction Techniques

  • 期刊论文

  • 发表日期：

    2026年1月7日

查看全文
  • Chandrakant Mallick,
  • Suryakanta Nayak,
  • Kamakhya Narain Singh,
  • Manas Ranjan Senapati

摘要

The Internet of Things (IoT) has significantly raised vulnerability to cyber threats, highlighting the need for effective and scalable Intrusion Detection Systems (IDS). This research presents a hybrid framework that combines Principal Component Analysis (PCA) with an optimized ensemble Random Forest classifier to improve intrusion detection accuracy in high-dimensional and imbalanced IDS datasets. To assess the performance of the framework, a systematic evaluation was made using three benchmark datasets—NSL-KDD, UNSW-NB15, and CICIDS2017. Comprehensive experiments indicate that PCA improves feature selection and model generalization, whereas Random Forest offers significant interpretability and robustness. The proposed method attains a peak accuracy of 99.88% on the NSL-KDD dataset and demonstrates competitive performance across other datasets. A comparative analysis with recent state-of-the-art models indicates that the framework consistently outperforms or matches deep learning-based methods. Although deep learning models like Transformers and attention-based architectures yield promising results in IDS research, they are computationally expensive and less appropriate for real-time IoT implementations. Our methodology offers a lightweight but efficient alternative, with future research directions aimed at incorporating advanced deep learning models with adaptive learning mechanisms.