Scholarlink

<p>Machine learning (ML) based algorithms have been extensively utilized to develop intrusion detection systems (IDS) that identify abnormal behaviors within network operations. The exponential growth in network traffic data has intensified the complexity of intrusion analysis, highlighting the limitations of single classifiers in handling massive datasets effectively. Thus, the given paper proposes a novel tree-based stacking ensemble model (SET), integrating five classifiers namely Support Vector Machine (SVM), Decision tree (DT), Random forest (RF), Extreme gradient boosting (XGBoost) and Naive Bayes (NB) classifiers, coupled with rigorous input feature scaling, cross-validation, and hyperparameter optimization to minimize overfitting. The proposed model introduces a feature importance fusion mechanism, aggregating importance scores from RF and XGBoost classifiers to select the most discriminative features, thereby improving dimensionality reduction and overall model interpretability. The model’s performance is evaluated on NSL-KDD and UNSW-NB15 datasets, incorporating robust feature selection to identify critical attributes. Results demonstrate that the proposed model attains the highest accuracy (%) and the lowest false alarm rate (FAR), outperforming existing intrusion detection approaches.</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

Enhancing intrusion detection accuracy with feature fusion and stacked ensemble approach: a dual-level learning framework

  • 期刊论文

  • 发表日期：

    2025年9月4日

查看全文
  • Seema Rani,
  • Sanjeev Kumar

摘要

Machine learning (ML) based algorithms have been extensively utilized to develop intrusion detection systems (IDS) that identify abnormal behaviors within network operations. The exponential growth in network traffic data has intensified the complexity of intrusion analysis, highlighting the limitations of single classifiers in handling massive datasets effectively. Thus, the given paper proposes a novel tree-based stacking ensemble model (SET), integrating five classifiers namely Support Vector Machine (SVM), Decision tree (DT), Random forest (RF), Extreme gradient boosting (XGBoost) and Naive Bayes (NB) classifiers, coupled with rigorous input feature scaling, cross-validation, and hyperparameter optimization to minimize overfitting. The proposed model introduces a feature importance fusion mechanism, aggregating importance scores from RF and XGBoost classifiers to select the most discriminative features, thereby improving dimensionality reduction and overall model interpretability. The model’s performance is evaluated on NSL-KDD and UNSW-NB15 datasets, incorporating robust feature selection to identify critical attributes. Results demonstrate that the proposed model attains the highest accuracy (%) and the lowest false alarm rate (FAR), outperforming existing intrusion detection approaches.