An Optimized Ensemble Intrusion Detection Framework with Hybrid Feature Selection and Multi-Dataset Validation
Year: 2026
Type: article
Abstract: Network Intrusion Detection Systems (NIDS) play a vital part in protecting modern communication networks against many classy cyberattacks. But the existing intrusion detection methods suffer from problems including data imbalance, redundant features, and poor performance in various network environments. Hence, this work presents a optimized ensemble Intrusion Detection System (IDS) based on comprehensive data preprocessing, hybrid feature selection, and ensemble learning techniques to improve intrusion detection performance. To enhance the data quality and class balance, the proposed framework includes missing value handling, outlier removal, noise filtering and Synthetic Minority Oversampling Technique (SMOTE). A hybrid feature selection method, which combines the Mutual Information (MI) and Recursive Feature Elimination (RFE), is used to identify the most useful network traffic features. Then, an attack classification ensemble model is constructed using Extreme Gradient Boosting (XGBoost) and Light Gradient Boosting Machine (LightGBM). This framework is tested on the CICIDS2017 and UNSW-NB15 datasets via 5-fold cross-validation, and its performance is measured by standard classification metrics. Additionally, explainability using SHapley Additive exPlanations (SHAP) is integrated to enhance the interpretability of the model prediction. The experimental results show that the proposed framework provides a strong and stable intrusion detection performance in both of the benchmark datasets.
Authors: Pooja Rani, Parvinder Singh
Language: English
Subfield: Computer Networks and Communications
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7225889165
