<![CDATA[Network intrusion detection remains a critical challenge in cybersecurity due to evolving attack patterns, class imbalance, and high-dimensional network traffic data. This study investigates the effectiveness of ensemble learning techniques on the CIC-IDS2017 public dataset, integrating decision trees, random forests, and gradient boosting models through stacking, voting, and hybrid Boost-Bag strategies. Data preprocessing involved normalization, handling missing values, and feature selection based on correlation and mutual information to reduce dimensionality while preserving predictive relevance. Empirical evaluation employed stratified 10-fold cross-validation and performance metrics including accuracy, recall, F1-score, and AUC-ROC, with additional analyses of confusion matrices and temporal stability to assess operational reliability. Results indicate that hybrid ensembles achieve superior detection performance, particularly for low-frequency attacks, while maintaining moderate computational overhead compared to individual classifiers. Comparative insights reveal trade-offs between accuracy, minority-class sensitivity, and inference latency, guiding practical deployment considerations. The findings substantiate the theoretical benefits of ensemble diversity and optimized feature selection, offering a robust framework for scalable, interpretable, and resilient network intrusion detection systems.]]>
Copyrights © 2026
