Mining fuzzy rules based on pattern trees

We develop a method to construct a fuzzy rule-based classifier which makes use of the pattern trees and Axiomatic Fuzzy Set (AFS) theory. The AFS framework supports a way on how to convert the information present in databases into the membership functions and their fuzzy logic operations. A selection index used for quantifying the discriminatory capabilities of the fuzzy concept was proposed. Being guided by the selection index, the antecedents of the fuzzy rules are selected from the fuzzy concepts which are found when using the pattern trees. The performance of the proposed classifier is compared with the results produced by classifiers commonly encountered in the literature when using ten datasets taken from the UCI Machine Learning Repository. It has been found that the accuracy on test data is higher than the ones produced by the other classifiers.