MACHINE LEARNING-BASED FRAMEWORK WITH ENHANCED FEATURE SELECTION AND RESAMPLING FOR IMPROVED INTRUSION DETECTION

Abstract

Intrusion Detection Systems (IDS) play a vital role in protecting modern computer networks against cyber threats, including botnet attacks, Man-in-theMiddle (MITM) attacks, malware, and other malicious activities. However, conventional intrusion detection approaches face significant challenges due to high-dimensional network traffic data, irrelevant and redundant features, and imbalanced class distributions. These limitations can reduce detection performance, increase computational complexity, and result in poor identification of minority attack classes. To address these challenges, this paper proposes a machine learning-based framework that integrates advanced feature selection and data resampling techniques for improved intrusion detection. The framework employs Recursive Feature Elimination, Sequential Feature Selection, and statistical feature selection to identify relevant network traffic attributes. Furthermore, resampling techniques, including Adaptive Synthetic Sampling , SMOTE-ENN, and SMOTE-Tomek, are considered to address class imbalance and improve the representation of minority attack categories. The selected and balanced features are then used to train machine learning classifiers. The performance of the models is evaluated using accuracy, precision, recall, F1- score, and false-positive and false-negative rates. The proposed framework aims to improve the reliability of intrusion detection, enhance minority attack identification, reduce the influence of irrelevant features, and support efficient network security monitoring.