Learning non-convex fuzzy classifiers using single-class SVMs

In this paper, we propose an approach for building tree-like structured fuzzy classifiers. In order to learn classes for non-convex shapes of data, basic building blocks modelling convex classes are composed. For this purpose, an idea for the integration of single-class support vector machines (SVMs) into fuzzy class learning is sketched. The leading thought of this hybrid approach is the creation of a robust and most notably well interpretable parametric fuzzy classification model. Its feasibility is demonstrated in the context of a machine diagnosis task and compared to standard soft-margin SVMs.