While cyberattacks get increasingly more sophisticated and common, traditional Intrusion Detection Systems (IDS) usually lack the capacity to catch new kinds of attacks, especially infrequent ones like User to Root (U2R) and Remote to Local (R2L) attacks. This paper suggests an upgraded Double-Layered Hybrid Approach (DLHA) with the addition of Gradient Boosting (or, specifically, eXtreme Gradient Boosting or XGBoost) alongside Naive Bayes classifiers. This strategy substitutes Support Vector Machine (SVM) in the second layer with XGBoost and adds synthetic augmentation with Generative Adversarial Networks (GANs) to counter class imbalance. Experimental comparisons on the NSL-KDD dataset show better detection rates, particularly for sparse classes, as well as lower false alarm ratios and computational complexity. This architecture offers a scalable, precise, and effective solution to contemporary Intrusion Detection System (IDS) challenges.