Feature Selection and XGBoost for Enhanced Intrusion Detection: A Comparative Study Across Benchmark Datasets

This study evaluated machine learning and deep learning models for Intrusion Detection Systems (IDS) using the UNSW-NB15, NSL-KDD, and CIC-IDS 2017 datasets. Among the tested models, XGBoost consistently achieved the highest accu- racy, demonstrating its robustness in handling complex network traffic. Feature selection methods, including Lasso Regularization and SelectKBest (ANOV A), were employed to simplify the models, with Lasso maintaining near-optimal performance and SelectKBest showing dataset-specific sensitivity. These findings highlight the efficacy of XGBoost and the importance of feature selection in improving IDS performance while balancing model complexity.