Design and comparison of two rule-based fuzzy classifiers for computer security

Abstrakt

This paper describes the design and comparison of two rule based fuzzy classifiers, which are subsequently enhanced with respect to comprehensibility. The first approach is based on random forests. The second is a specific kind of a fuzzy decision tree which is built on information granules extracted from results of a fuzzy clustering algorithm. Membership functions of fuzzy sets are fitted to the outputs of these two tree based classifiers. This gives us logical rules, which have a better comprehensibility. These two approaches are compared to each other and with a support vector machine classifier as a representative of precise classifiers. The comparison is performed on data concerning network security.