Credit assignment by fuzzy rule-based systems in fuzzy classifier ensembles

This paper examines credit assignment methods in fuzzy classifier ensembles. Multiple fuzzy rule-based classification systems are included in a fuzzy classifier ensemble. Fuzzy rule-based classification systems in a fuzzy classifier ensemble are constructed from given training patterns. In addition to fuzzy rule-based classification systems, we construct a credit assignment system that assigns a weight vector for each input pattern. For an input pattern, a class is suggested by each fuzzy rule-based classification system, and the final classification of the input pattern is determined by considering both the suggested class by the fuzzy rule-based classification systems and the weight vector from the credit assignment system. Fuzzy and non-fuzzy rule-based system is used for the credit assignment systems. The credit assignment systems are constructed so that large weights can be assigned to the fuzzy rule-based classification systems with high classification power. In computer simulations, we illustrate the effect of credit assignment systems for a two-dimensional data set.