Neuro-fuzzy versus non-parametric approach to system modeling and classification

Abstract

This paper presents the comparative study concerning selected neuro-fuzzy systems and non-parametric methods. Moreover, a new idea of rough-neuro-fuzzy systems is suggested to solve the problem of missing features. The main applications of methods under study are system modeling and classification. The non-parametric methods are based on density and regression estimates. They converge to the optimal solution when the sample size grows large. The neuro-fuzzy structures do not possess convergence properties however they are applied successfully in modeling and classification problems. The methods are illustrated on several simulation examples. © Springer-Verlag Berlin Heidelberg 2004.