Inconsistency measure associated discretization methods to network-based intrusion detection

Network intrusion is one of the most serious threats to network infrastructure and digital assets. Intrusion-detection systems detects un-authorized network access which violates the permissions settings of computer systems, networks, and information systems. Data mining is the one of the businesslike techniques amongst which are available for intrusion detection. Many data mining methods, including decision trees and rough set theory, would work better on discrete or binarized data. In data sets, on account of the existence in very large numbers, data discretization is deemed to be the data pretreatment method and such technology is very important. In this paper, the discretization methods of DFSFDP, DBSCAN, CAIM and K-means are implemented and applied to the KDD intrusion detection data set. Then four evaluation indexes are used to evaluate the effect of discretization. Moreover, with regard to the four discretization methods, their performance and accuracy are put forward and contrasted. The results of experiment acquired through using these methods on NSL-KDD dataset prove that used DFSFDP which is selected by the index of inconsistency measure performs well in terms of classification accuracy.