Scholarlink

<p>The exponential growth of the Internet of Things (IoT) ecosystem has drastically amplified exposure to cyberattacks, highlighting the critical need for intelligent, adaptive, and scalable security solutions. To address these challenges, this research introduces a Hybrid Stacked Deep Learning (HSDL) framework tailored for high-precision intrusion detection in IoT networks. The proposed architecture synergizes Vectorized Convolutional Neural Networks (VCNN) for advanced feature extraction with Stacked Long Short-Term Memory (SLSTM) networks to capture long-term temporal dependencies in sequential traffic patterns. To support real-time adaptability in large-scale environments, a Spark-based preprocessing pipeline is employed, accelerating data handling while maintaining efficiency. Furthermore, detection performance is optimized using Deep LSTM with Harris Hawks Optimization (DLSTM-HHO), a bio-inspired metaheuristic that fine-tunes hyperparameters for superior accuracy and generalization. Comprehensive evaluations were conducted on widely recognized benchmark datasets—N-BaIoT, UNSW-NB15, and UNSW-IoT-Botnet. Results demonstrate the remarkable effectiveness of the proposed model, with the HSDL framework achieving a peak accuracy of 99.89%, outperforming conventional baselines such as CNN and LSTM. Similarly, the DLSTM-HHO variant achieved 99% accuracy, significantly exceeding LSTM (95%), LSTM-RNN (83%), DNN (79%), and Naïve Bayes (78%). These findings underscore the capability of the HSDL framework to deliver robust and resource-efficient intrusion detection, effectively minimizing false positives while enabling real-time, scalable, and edge-deployable IoT threat monitoring.</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

Intelligent Deep Learning-Based NetFlow Botnet Detection and AI-Powered Malware Classification for IoT Edge Security

  • 期刊论文

  • 发表日期：

    2025年12月26日

查看全文
  • Oscar León-Granizo,
  • Kerly Palacios-Zamora,
  • Omar Yagual-Muñoz,
  • Denis Mendoza-Cabrera

摘要

The exponential growth of the Internet of Things (IoT) ecosystem has drastically amplified exposure to cyberattacks, highlighting the critical need for intelligent, adaptive, and scalable security solutions. To address these challenges, this research introduces a Hybrid Stacked Deep Learning (HSDL) framework tailored for high-precision intrusion detection in IoT networks. The proposed architecture synergizes Vectorized Convolutional Neural Networks (VCNN) for advanced feature extraction with Stacked Long Short-Term Memory (SLSTM) networks to capture long-term temporal dependencies in sequential traffic patterns. To support real-time adaptability in large-scale environments, a Spark-based preprocessing pipeline is employed, accelerating data handling while maintaining efficiency. Furthermore, detection performance is optimized using Deep LSTM with Harris Hawks Optimization (DLSTM-HHO), a bio-inspired metaheuristic that fine-tunes hyperparameters for superior accuracy and generalization. Comprehensive evaluations were conducted on widely recognized benchmark datasets—N-BaIoT, UNSW-NB15, and UNSW-IoT-Botnet. Results demonstrate the remarkable effectiveness of the proposed model, with the HSDL framework achieving a peak accuracy of 99.89%, outperforming conventional baselines such as CNN and LSTM. Similarly, the DLSTM-HHO variant achieved 99% accuracy, significantly exceeding LSTM (95%), LSTM-RNN (83%), DNN (79%), and Naïve Bayes (78%). These findings underscore the capability of the HSDL framework to deliver robust and resource-efficient intrusion detection, effectively minimizing false positives while enabling real-time, scalable, and edge-deployable IoT threat monitoring.