Scholarlink

<p>The increasing sophistication of cyber threats necessitates advanced intrusion detection systems (IDS) capable of adaptive and precise threat mitigation. This study presents an optimized deep learning (DL)-based IDS leveraging a deep neural network (DNN) with rectified linear unit (ReLU) activations and a tabular model utilizing the fastai framework. Both models were trained and evaluated on the NSL-KDD dataset following extensive preprocessing, including feature scaling, outlier handling, and class balancing. The fastai model achieved an accuracy of 84.19%, precision of 85.37%, recall of 83.92%, and F1-score of 84.64%, outperforming the DNN, which attained 79.14% accuracy, 81.25% precision, 78.60% recall, and 79.90% F1-score. Automated feature engineering and tenfold cross-validation were applied to enhance generalization and stability. The results demonstrate that deep learning provides a scalable, high-accuracy IDS framework capable of addressing the dynamic and evolving nature of cyber threats.</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

Optimized intrusion detection with deep learning classification models

  • 期刊论文

  • 发表日期：

    2025年7月22日

查看全文
  • Nabil M. Eldakhly

摘要

The increasing sophistication of cyber threats necessitates advanced intrusion detection systems (IDS) capable of adaptive and precise threat mitigation. This study presents an optimized deep learning (DL)-based IDS leveraging a deep neural network (DNN) with rectified linear unit (ReLU) activations and a tabular model utilizing the fastai framework. Both models were trained and evaluated on the NSL-KDD dataset following extensive preprocessing, including feature scaling, outlier handling, and class balancing. The fastai model achieved an accuracy of 84.19%, precision of 85.37%, recall of 83.92%, and F1-score of 84.64%, outperforming the DNN, which attained 79.14% accuracy, 81.25% precision, 78.60% recall, and 79.90% F1-score. Automated feature engineering and tenfold cross-validation were applied to enhance generalization and stability. The results demonstrate that deep learning provides a scalable, high-accuracy IDS framework capable of addressing the dynamic and evolving nature of cyber threats.