Network Intrusion Detection Using Kernel-based Fuzzy-rough Feature Selection

The purpose of the intrusion detection systems is to detect attacks on computer systems and networks. Many technologies can be used for intrusion detection, and one of the most effective technologies is data mining. The rapid development of network technology and internet of things makes network intrusion detection become one of the hot topics for research. Various classifiers have been applied in the field of network intrusion detection, but the performance of such approaches highly depends on the features used. Therefore, feature selection approaches have been usually used along with classifiers for network intrusion detection, including the fuzzy-rough feature selection. The fuzzy-rough sets is an extension of the classical rough sets, which can deal with the imprecision and uncertainty of discrete, real value or noise data. It can be seen from the practical applications that there are some shortcomings. Therefore, researchers combine fuzzy-rough sets with kernel methods in order to solve these problems. In this paper, the kernel-based fuzzy-rough feature selection method is used to select the feature subset for the intrusion detection. The proposed approach is validated and evaluated using the KDD 99 dataset with the support of different common classifiers. The experimental outcomes obtained by applying the kernel-based fuzzy-rough feature selection method on KDD data set demonstrate that it performs well in terms of reduction effect and accuracy.