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

Abstract

Network intrusion detection systems (NIDS) are essential to mitigate the increasing cyber security risks since they protect computer networks from a wide range of attacks. Class imbalances in the datasets frequently impede the effectiveness of these systems, leading to incorrect classification of minority classes. In addition, improper feature selection may lead to a degradation of results and higher computing time. This research combines the Jaya optimization technique with the Synthetic Minority Oversampling Technique—Edited Nearest Neighbors (SMOTE-ENN) method to address the problems of feature dimensionality and class imbalance, offering an efficient approach to identifying network intrusions. The Jaya optimization algorithm selects the most suitable features after the preprocessing stage. We test the proposed method on two widely used IDS datasets, namely UNSW-NB15 and NSL-KDD. We perform the classification task using five machine-learning approaches: decision trees (DT), random forests (RF), bagging, J48, and extra-tree classifiers for a comprehensive analysis. The Extra Tree Classifier emerged as the best performer, achieving exceptional accuracy values of 99.94% and 99.13% for NSL-KDD and UNSW-NB15, respectively. In both datasets, our proposed model outperforms the existing solutions proposed in the literature. The proposed methodology effectively addresses the problems of feature dimensionality and class imbalance, offering an efficient approach to identifying network intrusions.

Keywords

• Intrusion detection system
• Jaya optimization
• SMOTE-ENN
• machine learning
• class imbalance