Network Intrusion Detection Using UNSW-NB15 Dataset: Stacking Machine Learning Based Approach

Abstract

Network intrusion has become a prime concern issue for the industry and government organizations in the domain of the cyber-threat landscape. To counter this threat, a network intrusion detection system has been considered to be vital in identifying network traffic as normal or anomaly. Correct identification of the potential threat as an anomaly depends on the accuracy of the Network Intrusion Detection System (NIDS). Several approaches like single classical, hybrid, and ensemble methods are in practice to develop a network intrusion detection model. In this paper, two different stacking Machine Learning (ML) models with Extra Tree (ET) Classifier and Mutual Information Gain feature selection methods are proposed for better accuracy of the NIDS. We applied the models on the UNSW-NB15 packet-based dataset which contains the most recent attack types and experimentally proved that the testing accuracy of the stacking models is better than all individual models. Comparative results also depict that one of our proposed models shows better accuracy (96.24%) than any other existing competing models.