A fuzzy regression based support vector machine (SVM) approach to fuzzy classification

The objective of this study is to develop a fuzzy regression model using support vector machine (SVM) to problems of classifying patterns belonging to two overlapping classes. The design of the regression model consists of two phases. Phase I uses a fuzzy linear regression to separate linearly two classes of patterns. As a result, the fuzzy linear regression may separate the feature space into three main regions, that is (a) a region occupied by patterns belonging to class 1, (b) a region occupied by patterns belonging to class 2 and (c) the region, in which we encounter a mixture of the patterns belonging to the two classes. In Phase 2, we develop an SVM to non-linearly separate the mixture of the patterns. It will be shown that the proposed fuzzy regression comes with a significant advantage of shortening the processing time associated with the realization of the SVM.