Kernel based hybrid fuzzy clustering for non-linear fuzzy classifiers

In this paper, an objective function based approach is presented to characterize a fuzzy classifier system via a kernel learning algorithms for non-linear data. We combine the distance based kernel fuzzy clustering and the non-linear support vector classification (SVC) with a conjoint objective based fuzzy clustering method in a novel way in order to learn a fuzzy classifier system. The two objectives are balanced with a regularization term. An additional merit of the novel method is that the information on natural groupings of the data samples i.e., the membership values, are utilized as additional predictors of each fuzzy classifier function learnt from the non-linear SVC to improve the accuracy of the classifier model. The comparative experiments demonstrate the effectiveness of the proposed method in building a classifier model for a detection system.