Comparison of fuzzy robust Kernel C-Means and support vector machines for intrusion detection systems using modified kernel nearest neighbor feature selection

Abstract

Intrusion detection Systems (IDS) can be used to monitor and analyze user activities in a computer. One of the most important tasks of IDS is to protect the target of the attack: user password, file systems and kernel systems. The attack itself can be classified into two categories, which are host-based attacks, and also network based attacks. This study proposes a new method that is FRKCM (Fuzzy Robust Kernel C-Means) to solve IDS problems. For our empirical study, we use dataset from KDD99, which contains five classes: Normal, Probe, DOS, U2R and R2L. This paper also discusses Feature Selection procedure because it may improve the performance of classification algorithm. For the Feature Selection, MKNN (Modified Kernel Nearest Neighbor) method has been chosen in this paper. MKNN is a new method for feature selection. There will be an accuracy comparison between FRKCM method and SVM (Support Vector Machine) method. Our results indicate that the Fuzzy Robust Kernel C-Means provides better results better than SVM method in terms of classification accuracy because the highest accuracy of FRKCM method using Poli Kernel reaches approximately 99.26 % while SVM method using RBF Kernel was only 99.22 %.