Scholarlink

<p>Intrusion detection is crucial for modern cyber-security, aiming to detect and mitigate unauthorized access and malicious activities within networks. Traditional systems often rely on linear models, which may fail to capture complex patterns in advanced threats. This study proposes a heuristic approach to select the best deep features through an ensemble framework consisting of Random Forest (RF) and Support Vector Machine (SVM) classifiers using soft voting. The novelty of this work is the deep feature extraction through Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), and Autoencoder for intrusion detection. These extracted features were fed separately for classification. In order to balance the explainability and accuracy, the ensemble classifier consisting of RF and SVM, uses feature importance ranking (RF) and unambiguous decision bounds (SVM) to provide partial interpretability. Evaluating the NSL-KDD and CICDDoS2019 datasets, our ensemble model with CNN features achieves near-perfect accuracy of 99.99% and 99.91% respectively. RNN and Autoencoder models also show high performance, each reaching 99.3% accuracy on the CICDDoS2019 dataset with Soft Voting classifiers. This work highlights the importance of the ensemble method with better feature selection in intrusion detection. The proposed system outperformed the standard methods by providing a robust, scalable solution for organizational security.</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

Enhanced Intrusion Detection with Advanced Deep Features and Ensemble Classifier Techniques

  • 期刊论文

  • 发表日期：

    2025年4月15日

查看全文
  • Pawan Toralkar,
  • Kavita Mainalli,
  • Shridhar Allagi,
  • Sanjoy Kumar Debnath,
  • Susama Bagchi,
  • Wai Yie Leong,
  • Muhammad Numan Ali Khan

摘要

Intrusion detection is crucial for modern cyber-security, aiming to detect and mitigate unauthorized access and malicious activities within networks. Traditional systems often rely on linear models, which may fail to capture complex patterns in advanced threats. This study proposes a heuristic approach to select the best deep features through an ensemble framework consisting of Random Forest (RF) and Support Vector Machine (SVM) classifiers using soft voting. The novelty of this work is the deep feature extraction through Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), and Autoencoder for intrusion detection. These extracted features were fed separately for classification. In order to balance the explainability and accuracy, the ensemble classifier consisting of RF and SVM, uses feature importance ranking (RF) and unambiguous decision bounds (SVM) to provide partial interpretability. Evaluating the NSL-KDD and CICDDoS2019 datasets, our ensemble model with CNN features achieves near-perfect accuracy of 99.99% and 99.91% respectively. RNN and Autoencoder models also show high performance, each reaching 99.3% accuracy on the CICDDoS2019 dataset with Soft Voting classifiers. This work highlights the importance of the ensemble method with better feature selection in intrusion detection. The proposed system outperformed the standard methods by providing a robust, scalable solution for organizational security.