PERFORMANCE COMPARISON OF SVM KERNELS FOR INTRUSION DETECTION SYSTEM USING UNSW-NB15 DATASET

Given the proliferation of internet security concerns, the Intrusion Detection System has become an essential part of the IoT network. The growing demands for study in the realm of cyberattacks necessitate the availability of datasets. UNSW-NB15 is a publicly accessible security dataset. Since its inception in 2015, numerous researchers have used this dataset to elucidate successful models for threat classification and prediction-based machine learning. Nevertheless, there is a deficiency of research specifically examining the comparison of kernels in relation to the SVM classifier. This paper presents a performance comparison of four SVM kernels. The model's outputs are assessed using execution time and false positive rate, along with four assessment metrics: accuracy, precision, recall, and F1 score. The results demonstrate that the Poly kernel attains the maximum performance, with an accuracy of 98.78%, precision of 97.98%, recall of 98.27%, and an F1 score of 98.12. Nevertheless, the execution duration of the RBF kernel is the most rapid among other SVM kernels, totaling 10 minutes and 23 seconds. Regarding the False Positive Rate, the Linear kernel exhibits optimal performance at 20%.