Using Recursive Feature Elimination Feature Selection based Machine Learning Classifier for Attack Classification on UNSW-NB 15 dataset

In modern computer networks, intrusion detection systems (IDS) play a crucial role by detecting and notifying administrators about malicious activities. Ensuring network security and identifying unauthorized access to network resources is paramount in constructing an effective IDS. Traditionally, intrusion detection research relied heavily on the KDDCUP99 dataset, but its limitations in evaluating network intrusion detection systems (NIDS) compared to the UNSW-NB15 dataset have become apparent. This article introduces a novel approach that incorporates feature selection techniques to enhance the performance of three classification methods: K-Nearest Neighbors (KNN), Decision Trees (DT), and Logistic Regression (LR). By utilizing the recursive feature elimination (RFE) technique, irrelevant features are eliminated, allowing the model to focus on the most discriminative attributes for intrusion detection. This not only improves classification accuracy but also reduces computational complexity. The study compares the performance of KNN, DT, and LR algorithms using various metrics such as accuracy, precision, recall, and F1-score. The findings provide a comprehensive analysis of each algorithm's strengths and weaknesses, offering valuable guidance to practitioners and researchers when selecting appropriate algorithms and feature selection techniques for real-world intrusion detection in network environments.