Performance increase of intrusion detection systems utilising support vector machine (SVM) classification method by feature selection

While Internet is making our life easy in many aspects, which has been one of the most concrete sample of the technological improvements in this era, misuse of the deficiencies in information technology components creates negative results seriously by people with malicious intensions. Systems used in network security like antivirus software, firewalls, authentication verification, access and environment controls are supplemented by intrusion detection systems. Feature selection methods are utilized to increase the performance of support vector machines which are used in anomaly intrusion detection system model mainly for classification. In this research, it is shown that classification performance of support vector machine on KDD99 data set can be increased by Fisher Score feature selection method. In the study, different subsets are created of the attributes ordered by importance by the feature selection method. It is seen that the less is the number of the attributes in a subset, the more affected is the performance by the replacement of any attribute in the subset, and that the more gets the number of the low-value- attributes in a subset, either the accuracy score decreases and execution time increases or accuracy score does not change notably and execution time increases significantly.