Scholarlink

<p>The rapid expansion of the internet of things (IoT) has introduced significant security vulnerabilities, as traditional signature-based detection methods struggle to keep pace with evolving cyber threats. This study proposes a lightweight intrusion detection system (IDS) that leverages boosting ensemble learning to identify malicious traffic at the network edge. Using the large-scale Aposemat IoT-23 dataset, comprising over 25&#xa0;million records, we performed extensive data preprocessing and exploratory data analysis to ensure statistical stability and address class imbalance. We evaluated several machine learning architectures, including logistic regression, Naïve Bayes, Random Forest, AdaBoost, XGBoost, and LightGBM. Experimental results from five-fold cross-validation show that LightGBM outperforms other models, achieving 99.9% accuracy, F1 Score, and ROC AUC, with significantly lower training time of 612&#xa0;s. Furthermore, confusion matrix analysis confirms that ensemble methods effectively minimize critical false negatives and false positives. These findings suggest that enhanced ensemble techniques provide a scalable, high-performance solution for real-time malware detection in resource-constrained IoT environments, bridging the gap between detection accuracy and computational efficiency.</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 internet of things intrusion detection through high-performance boosting ensemble learning

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月26日

查看全文
  • Vitaliy Yakovyna,
  • Anton Fadieiev

摘要

The rapid expansion of the internet of things (IoT) has introduced significant security vulnerabilities, as traditional signature-based detection methods struggle to keep pace with evolving cyber threats. This study proposes a lightweight intrusion detection system (IDS) that leverages boosting ensemble learning to identify malicious traffic at the network edge. Using the large-scale Aposemat IoT-23 dataset, comprising over 25 million records, we performed extensive data preprocessing and exploratory data analysis to ensure statistical stability and address class imbalance. We evaluated several machine learning architectures, including logistic regression, Naïve Bayes, Random Forest, AdaBoost, XGBoost, and LightGBM. Experimental results from five-fold cross-validation show that LightGBM outperforms other models, achieving 99.9% accuracy, F1 Score, and ROC AUC, with significantly lower training time of 612 s. Furthermore, confusion matrix analysis confirms that ensemble methods effectively minimize critical false negatives and false positives. These findings suggest that enhanced ensemble techniques provide a scalable, high-performance solution for real-time malware detection in resource-constrained IoT environments, bridging the gap between detection accuracy and computational efficiency.