A Novel IDS Based on Jaya Optimizer and Smote-ENN for Cyberattacks Detection

Abstract

Network Intrusion Detection Systems (NIDS) are pivotal in addressing the increasing cybersecurity threats by protecting computer networks from various types of cyberattacks. However, class imbalances in datasets often reduce the effectiveness of detection algorithms, leading to misclassification of minority classes. Additionally, poor feature selection can hinder performance and increase computational load. In this study, we utilize the NSL-KDD and UNSW-NB15 datasets to develop a robust IDS model. We applied Jaya Optimization for feature selection to enhance relevant data features and utilized the SMOTE-ENN method for oversampling to balance the dataset. Various algorithms were tested, including Decision Tree, Random Forest, Bagging with Decision Trees, J48, ExtraTree, and a Voting Classifier combining ExtraTree with Boosted Decision Tree. To achieve higher accuracy, ensemble methods were incorporated, combining predictions from multiple individual models. Among the tested algorithms, the Voting Classifier exhibited superior performance, achieving an accuracy rate of 100% in NSL-KDD (with SMOTE-ENN and without SMOTE-ENN) and 100% in UNSWNB15 (with SMOTE-ENN) while achieving 92% accuracy in UNSW-NB15 (without SMOTE-ENN), highlighting its robustness and precision in cyberattack detection. This approach promises improved network security by addressing class imbalances and optimizing feature selection, making NIDS systems more effective in real-world applications.