Learning in rough-neuro-fuzzy system for data with missing values

Abstract

Rough-neuro-fuzzy systems offer suitable way for classifying data with missing values. The paper presents a new implementation of gradient learning in the case of missing input data which has been adapted for rough-neuro-fuzzy classifiers. We consider the system with singleton fuzzification, Mamdani-type reasoning and center average defuzzification. Several experiments based on common benchmarks illustrating the performance of trained systems are shown. The learning and testing of the systems has been performed with various number of missing values. © 2012 Springer-Verlag.