Scholarlink

<p>The Internet of Things (IoT) has experienced exponential growth, with a vast number of interconnected devices infiltrating daily life and industries. As this ecosystem expands, IoT security becomes increasingly vital due to the rising sophistication of cyber threats like malware, DDoS attacks, and data breaches, which endanger the confidentiality, integrity, and availability of IoT-based systems. Challenges in IoT security are rooted in the diverse nature of devices, limited resources at IoT endpoints, and the complex, dynamic network environment. This research aims to address the need for highly accurate and efficient IoT threat detection. We conduct an in-depth comparison between ensemble machine learning classifiers and advanced hybrid neural network models, using data from three key datasets: IoT-23, N-BaIoT, and CICIDS2017. Through rigorous experimentation, our refined techniques yield remarkable results. The proposed models achieved 95% accuracy on the IoT-23 dataset, along with notable area under the curve (AUC) scores. For the N-BaIoT and CICIDS2017 datasets, an impressive 99.99% accuracy was attained, highlighting the models’ adaptability to different threat scenarios. This research contributes to IoT security by comprehensively evaluating state-of-the-art threat detection methods, providing valuable insights for researchers and practitioners, and suggesting directions for future research in optimizing model selection and data processing techniques.</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

Empirical evaluation of ensemble learning and hybrid CNN-LSTM for IoT threat detection on heterogeneous datasets

  • 期刊论文

  • 发表日期：

    2025年4月23日

查看全文
  • Ahsan Nazir,
  • Jingsha He,
  • Nafei Zhu,
  • Ahsan Wajahat,
  • Fahim Ullah,
  • Sirajuddin Qureshi,
  • Muhammad Salman Pathan

摘要

The Internet of Things (IoT) has experienced exponential growth, with a vast number of interconnected devices infiltrating daily life and industries. As this ecosystem expands, IoT security becomes increasingly vital due to the rising sophistication of cyber threats like malware, DDoS attacks, and data breaches, which endanger the confidentiality, integrity, and availability of IoT-based systems. Challenges in IoT security are rooted in the diverse nature of devices, limited resources at IoT endpoints, and the complex, dynamic network environment. This research aims to address the need for highly accurate and efficient IoT threat detection. We conduct an in-depth comparison between ensemble machine learning classifiers and advanced hybrid neural network models, using data from three key datasets: IoT-23, N-BaIoT, and CICIDS2017. Through rigorous experimentation, our refined techniques yield remarkable results. The proposed models achieved 95% accuracy on the IoT-23 dataset, along with notable area under the curve (AUC) scores. For the N-BaIoT and CICIDS2017 datasets, an impressive 99.99% accuracy was attained, highlighting the models’ adaptability to different threat scenarios. This research contributes to IoT security by comprehensively evaluating state-of-the-art threat detection methods, providing valuable insights for researchers and practitioners, and suggesting directions for future research in optimizing model selection and data processing techniques.