Scholarlink

In the context of real-time Network Intrusion Detection (NID), where traditional feature selection methods and Explainable Artificial Intelligence (XAI) techniques frequently prove to be too resource-intensive. The growing complexity of network attacks poses significant challenges for NID. To overcome the existing challenges this work presents a Bidirectional Long Short-Term Memory (BiLSTM) model for NID feature selection that is dynamic and interpretable, based on attention. The BiLSTM framework’s attention mechanism finds the most significant features from the NSL-KDD dataset, which enhances a Convolutional Neural Network’s (CNN) classifier’s performance. The suggested model outperforms several conventional and XAI-based models, achieving a classification accuracy of 95% on both test and validation datasets by training the CNN on these chosen characteristics. The results demonstrate the attention mechanism’s dual advantages of improving detection accuracy.

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

Dynamic Feature Selection with Attention Mechanism: BiLSTM-CNN Hybrid Approach for Network Intrusion Detection

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Arpa Tasnim,
  • Rafid Mehda,
  • Md. Sazzadur Rahman,
  • Nazneen Akhter

摘要

In the context of real-time Network Intrusion Detection (NID), where traditional feature selection methods and Explainable Artificial Intelligence (XAI) techniques frequently prove to be too resource-intensive. The growing complexity of network attacks poses significant challenges for NID. To overcome the existing challenges this work presents a Bidirectional Long Short-Term Memory (BiLSTM) model for NID feature selection that is dynamic and interpretable, based on attention. The BiLSTM framework’s attention mechanism finds the most significant features from the NSL-KDD dataset, which enhances a Convolutional Neural Network’s (CNN) classifier’s performance. The suggested model outperforms several conventional and XAI-based models, achieving a classification accuracy of 95% on both test and validation datasets by training the CNN on these chosen characteristics. The results demonstrate the attention mechanism’s dual advantages of improving detection accuracy.