Optimized Ensemble Intrusion Detection: Balancing Data with SMOTE-ENN and Feature Selection via Jaya Algorithm

Abstract: Network Intrusion Detection Systems (NIDS) are highly significant in ensuring that computer networks are not exposed to emerging cyber threats. However, non-balanced datasets can also reduce the accuracy of detection, particularly on minority attack classes, and non-important features can also obstruct this. This paper proposes a superior way of detecting intrusions within groups. It applies SMOTE-ENN, which equalizes the classes and Jaya Optimization, which selects the most suitable features. To test the system, the NSL-KDD and UNSW-NB15 datasets are used. Pretrained data is trained by individual classifiers such as Decision Tree, Random Forest, ExtraTree, J48 and Bagging with Decision Tree. It is further followed by an ensemble Voting Classifier which composes ExtraTree and Boosted Decision Tree. Tests show that the suggested model works better than others; it gets 100% accuracy, precision, recall, and F1-score on both sets of SMOTE-ENN data, and up to 92% accuracy on unbalanced UNSW-NB15 data. The approach is effective in fixing the imbalance between classes, simplifying calculations, and enhancing the identification of minority types of attacks. This renders it a viable and solid solution to network security challenges in the real world.

Keywords: Network Intrusion Detection System (NIDS), SMOTE-ENN, Jaya Optimization, Feature Selection, Ensemble Learning, Voting Classifier.