NID-Shield: Improving Intrusion Detection thr...

NID-Shield: Improving Intrusion Detection through CAPPER Feature Optimization

Author Name : Jai Mungi, Harsh Mathur

ABSTRACT The escalating sophistication of network-based threats has necessitated the development of more effective Intrusion Detection Systems (IDSs). This paper presents NID-Shield, a hybrid framework that integrates a novel feature selection method, termed CAPPER (Correlation and Wrapper-based Adaptive Performance-oriented Predictor Enhancement and Refinement), with stacking-based ensemble learning for enhanced network intrusion detection. CAPPER synergizes Correlation-based Feature Selection (CFS) with wrapper evaluation techniques to eliminate redundant and irrelevant features systematically, employing Symmetrical Uncertainty (SU), C4.5 (J48), and Naive Bayes classifiers as appraisers. NID-Shield leverages Random Forest and stacking ensemble learning to perform two-tiered attack classification at both the attack-category level (e.g., DoS, R2L, U2R, Probe) and the granular attack-name level. Experimental evaluation on the NSL-KDD and UNSW-NB15 benchmark datasets demonstrates that NID-Shield achieves a weighted average accuracy of 99.90% and 99.89%, respectively, with an MCC of 0.992, outperforming several recently reported approaches.