Kernel Method for Building Fuzzy Classifiers

This paper investigates the connection between modified fuzzy basis function (MFBF)-based classifiers and support vector classifiers, establishes a link between fuzzy rules and kernels, and proposes a new approach to build MFBF-based classifiers. Under some minor constrains, the equivalence of the two seemingly quite distinct classifiers is proved. Moreover, the kernel method has the inherent advantage that the MFBF-based classifiers do not have to determine the number of rules in advance. The designed classifier can be represented as a decision function consisting of series expansion of MFBFs, and this also makes itself to be interpretable. The performance of the proposed approach is illustrated by IRIS data sets and comparisons with other methods are also provided