Unification of Fuzzy SVMs and Rule Extraction Methods through imprecise Domain Knowledge

Christian Moewes, Rudolf Kruse.

In this paper, we want to motivate the combination of kernel-based methods with fuzzy rule extraction methods to describe uncertain domains by fuzzy models. We thus introduce and motivate the concept of a fuzzy support vector machine (FSVM) to incorporate impreciseness into kernel machines. Furthermore, we present the idea of a positive definite fuzzy classifier (PDFC), the rules of which are obtained by kernel-based models. We conclude with two vague conceptions to associate FSVM with PDFC to finally obtain understandable and meaningful fuzzy rules.

PDF full paper