New Approach of Ensemble Method to Improve Performance of IDS using S-SDN Classifier

Abstract

The application of Machine Learning (ML)-based Intrusion Detection System (IDS) has been widely used. The advantage of ML-based IDS is that it can detect intrusions in the network. However, in its application, there are still false positive detections on the IDS. False positive detection occurs due to improper ML techniques. This research applies an S-SDN model based on Ensemble Learning (EL) to overcome this problem. The S-SDN model is built from three base-learners, namely SVM, Decision Tree, and Naïve Bayes with the Stacking technique. Furthermore, the S-SDN model is used as a classifier on the IDS to detect intrusions. S-SDN was validated using the UNSW-NB15 dataset. Based on the experiment, S-SDN's performance was superior to the old method based on a single classifier. The performance of S-SDN can achieve an accuracy of 83.19%. In comparison, the old method based on a single classifier (SVM) can only achieve an accuracy of 75.89%, and the ensemble classifier (Bagging-DT) is only 80,09%. As for further research, the development of EL-based IDS still needs to be improved. For example, it builds an EL-based model with feature selection techniques and different base learners.