Scholarlink

<p>The expedited growth of Internet of Things (IoT) and Industrial IoT (IIoT) ecosystems, fueled by the widespread deployment of interconnected devices, has significantly increased security vulnerabilities. As cyber threats continue to evolve, it is imperative to develop robust, adaptive, and intelligent intrusion detection systems to safeguard these environments. Most of the existing intrusion detection methods feature either an isolated feature selection method or a single-model classifier, usually leading to very poor performance for complex and evolving attack surfaces within IoT. In response, this research proposes a novel framework for classifying network attacks, validated across three benchmark datasets: CICIDS2017, UNSW-NB15, and NSL-KDD. The framework operates in two phases: Phase 1 employs an ensemble feature selection strategy, integrating Information Gain, Random Forest, and XGBoost, to optimize feature representation and reduce dimensionality. In Phase 2, different Machine learning classifiers use the feature kernel that is created by intersecting the top-ranked features. The main findings demonstrate that the proposed ensemble feature selection significantly improves the detection accuracy while reducing the model complexity, achieves consistent performance across heterogeneous datasets. The framework attains detection accuracies of 86.47%, 89.22%, and 99.90% on NSL-KDD, UNSW-NB15, and CICIDS2017, respectively, indicating its effectiveness and practical suitability for IoT-based intrusion detection.</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

Secure-IoT: an optimized ensemble feature selection framework for enhanced network security in IoT systems

  • 期刊论文

  • 发表日期：

    2026年3月7日

查看全文
  • Faraj Chishti,
  • Geetanjali Rathee

摘要

The expedited growth of Internet of Things (IoT) and Industrial IoT (IIoT) ecosystems, fueled by the widespread deployment of interconnected devices, has significantly increased security vulnerabilities. As cyber threats continue to evolve, it is imperative to develop robust, adaptive, and intelligent intrusion detection systems to safeguard these environments. Most of the existing intrusion detection methods feature either an isolated feature selection method or a single-model classifier, usually leading to very poor performance for complex and evolving attack surfaces within IoT. In response, this research proposes a novel framework for classifying network attacks, validated across three benchmark datasets: CICIDS2017, UNSW-NB15, and NSL-KDD. The framework operates in two phases: Phase 1 employs an ensemble feature selection strategy, integrating Information Gain, Random Forest, and XGBoost, to optimize feature representation and reduce dimensionality. In Phase 2, different Machine learning classifiers use the feature kernel that is created by intersecting the top-ranked features. The main findings demonstrate that the proposed ensemble feature selection significantly improves the detection accuracy while reducing the model complexity, achieves consistent performance across heterogeneous datasets. The framework attains detection accuracies of 86.47%, 89.22%, and 99.90% on NSL-KDD, UNSW-NB15, and CICIDS2017, respectively, indicating its effectiveness and practical suitability for IoT-based intrusion detection.