SVM Based Network Intrusion Detection for the UNSW-NB15 Dataset

Due to the growth of internet security issues, Network Intrusion Detection System (NIDS) becomes an integral part of the IoT environment. In the past, most research on intrusion detection was experimented with the KDDCUP99 dataset. However, the KDDCUP99 dataset lacks some typical examples when evaluating NIDS compared with the UNSW-NB15 dataset. In this paper, we propose Support Vector Machine (SVM) with a new scaling method for binary-classification and multi-classification experiments. The performance of our method is evaluated through accuracy, detection rate and false positive rate. Compared with other methods, the superiority of the proposed SVM method is shown by the experimental results. The accuracy of the proposed method reaches 85.99% for binary-classification, compared to 78.47% by Expectation-Maximization (EM) clustering. For multi-classification, the proposed SVM method can achieve the testing accuracy of 75.77%, which is 6.17% higher than that of Naïve Bayes (NB).