Scholarlink

<p>The rapid proliferation of IoT devices has led to an unprecedented growth in network traffic, increasing the risk and complexity of cyberattacks in large-scale environments. This study presents a machine learning-based assessment of IoT attack classification using the BoT-IoT dataset under both weighted and non-weighted class scenarios. Multiple classical algorithms, including Random Forest, Artificial Neural Networks, Support Vector Machines, and Logistic Regression, are systematically and comparatively evaluated using precision, recall, accuracy, F1-score, ROC-AUC, and log loss within a consistent experimental setup. The study emphasizes a rigorous and reproducible benchmarking protocol that incorporates standardized preprocessing, feature engineering, and imbalance-aware evaluation. Experimental results demonstrate that Random Forest and ANN achieve superior performance in detecting diverse IoT attack patterns under varying class distributions. The findings provide practical insights into classifier selection for IoT intrusion detection and establish a foundation for future integration into scalable and real-time security systems.</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

A rigorous comparative evaluation of artificial intelligence techniques for imbalance-aware attack classification in IoT networks

  • 期刊论文

  • 发表日期：

    2026年4月13日

查看全文
  • Tariq Ahamed Ahanger,
  • Usman Tariq,
  • Imdad Ullah

摘要

The rapid proliferation of IoT devices has led to an unprecedented growth in network traffic, increasing the risk and complexity of cyberattacks in large-scale environments. This study presents a machine learning-based assessment of IoT attack classification using the BoT-IoT dataset under both weighted and non-weighted class scenarios. Multiple classical algorithms, including Random Forest, Artificial Neural Networks, Support Vector Machines, and Logistic Regression, are systematically and comparatively evaluated using precision, recall, accuracy, F1-score, ROC-AUC, and log loss within a consistent experimental setup. The study emphasizes a rigorous and reproducible benchmarking protocol that incorporates standardized preprocessing, feature engineering, and imbalance-aware evaluation. Experimental results demonstrate that Random Forest and ANN achieve superior performance in detecting diverse IoT attack patterns under varying class distributions. The findings provide practical insights into classifier selection for IoT intrusion detection and establish a foundation for future integration into scalable and real-time security systems.