A MACHINE LEARNING-BASED HYBRID MODEL FOR ROBUST INTRUSION DETECTION

The increasing frequency and complexity of cyberattacks have intensified the need for reliable and intelligent intrusion detection systems (IDS). Traditional methods, such as signature-based and anomaly-based detection, often suffer from high false positives and limited adaptability to novel threats. To address these limitations, this study develops a hybrid machine learning framework based on stacked ensemble learning for more robust and accurate intrusion detection. The framework integrates classifiers, including Decision Tree (DT), Support Vector Machine (SVM), and Random Forest (RF), optimized with feature selection and preprocessing techniques, and evaluated on the UNSW-NB15 dataset. Results show that all models achieved outstanding performance, with Decision Tree attaining perfect accuracy (100%) but with risks of overfitting, SVM recording stronggeneralization at 99.63% accuracy, and Random Forest emerging as the most effective with 99.99% accuracy, balanced precision, recall, and F1-score of 1.00, and excellent ROC-AUC values. The findings highlight the superiority of ensemblebased approaches in achieving high detection rates while minimizing false alarms. This research presents a reliable and scalable hybrid framework that enhances IDS capabilities, offering a promising pathway for deployment in real-world cybersecurity environments.