همایش بین المللی هوش مصنوعی

صفحه اصلی / دومین همایش بین المللی هوش مصنوعی

Cross-Dataset Empirical Evaluation of NSGA-II for Multi-Objective Feature Selection in Intrusion Detection Systems

نویسندگان :

Mahdis Rahmani¹ Fereshteh-Azadi Parand²

1- Allameh Tabataba’i University 2- Allameh Tabataba’i University

کلمات کلیدی :

Non-dominated Sorting Genetic Algorithm II (NSGA-II)،Feature Selection،Intrusion Detection Systems (IDS)،Multi-objective Optimization

چکیده :

Abstract—Intrusion Detection Systems (IDS) play a vital role in network security, as they monitor and detect cybersecurity attacks. However, the increasing complexity and dimensionality of IDS datasets pose significant challenges for training intrusion detection models. Redundant or irrelevant features within these datasets can decrease classifier performance and increase computational overhead. As a result, feature selection becomes an essential step in the preprocessing phase. Many existing feature selection methods used in IDS datasets rely on single-objective optimization, which often struggles to effectively balance the trade-off between minimizing feature count and maximizing accuracy. This study conducts a cross-dataset empirical evaluation of the Non-dominated Sorting Genetic Algorithm II (NSGA-II) for multi-objective feature selection in IDS datasets. We evaluate the proposed framework using seven benchmark datasets namely NSL-KDD, UNSW-NB15, CIC-IDS2017, UKM-IDS20, UNR-IDD, N-BaIoT and BoT-IoT, which encompass traditional networks, modern networks, and IoT environments. Experimental results demonstrate that NSGA-II reduces the feature space by an average of 72.36% while maintaining detection accuracy. In many cases, feature reduction results in improved performance, with the best instance achieving a 1.28% increase in accuracy. These findings highlight NSGA-II as a generalizable and effective method for feature selection in high-dimensional IDS applications.