Abstract—The rapid growth of network connectivity has increased the frequency and complexity of cyberattacks, placing heavy demands on Intrusion Detection Systems (IDS) to detect evolving threats accurately and efficiently. Traditional IDS models often suffer from high false-positive rates, poor generalization, and limited ability to process high-dimensional and imbalanced network traffic. To address these deficiencies, recent studies have explored machine learning and ensemble-based IDS solutions; however, they commonly lack robust hybrid feature selection, effective handling of class imbalance, and advanced hyperparameter optimization. These gaps limit detection accuracy and scalability in real-world environments. This paper proposes Boruta-Logistic-Extra Trees-Xensemble (BLEX), a modular IDS framework integrating a hybrid feature selection method Boruta + Mutual Information (BOMI) and a Bayesian-optimized stacking ensemble (BATO-tuned ExtraTrees, CatBoost, and Logistic Regression). The framework incorporates Interquartile Range (IQR)-based outlier removal, Borderline-SMOTE for class balancing, and probabilistic hyperparameter tuning using Tree-structured Parzen Estimators. Simulation results on a large Kaggle network-traffic dataset demonstrate the effectiveness of the proposed framework, achieving 97.28% accuracy, 93.49% precision, 91.88% recall, 94.23% F1-Score, and a low Root Mean Square Error (RMSE) of 0.2592. These results highlight the improved generalization, scalability, and robustness of the proposed BLEX model for modern, high-dimensional intrusion detection.
 
Keywords—Intrusion Detection System (IDS), Boruta-mutual information, Bayesian optimization with Tree-structured Parzen Estimator (TPE), social protection, vulnerable, process innovation
 
Copyright © 2026 by the authors. This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited (CC BY 4.0).