A Comparative study of machine learning models for Network Intrusion Detection System using UNSW-NB 15 dataset

In recent days, the Intrusion Detection System (IDS) has become a fundamental component of network security for an organization. Several approaches have been proposed and developed for IDS to protect the perimeter network and resources from different cyber-attacks. In this work, we concentrated on machine learning (ML) techniques to build IDS, because of the effectiveness of ML models to provide better accuracy in anomaly detection. However, we used UNSW-NB 15 dataset as an offline dataset to evaluate the binary classification-based ML models. Performance analysis was conducted by training and testing the Decision Tree (DT), Random Forest (RF), Gradient Boosting Tree (GBT), and Multi-Layer Perceptron (MLP). Using the Chi-Square test, we removed the features which were independent of response. However, DT was found out as the best classifier with maximum accuracy and lowest False Positive Rate (FPR). Except for RF, overall performance for all models was improved with feature elimination. Experimental analysis revealed that our proposed approach is superior to other existing ML methods in terms of accuracy.