Fuzzy classification using probability-based rule weighting

Abstract

Design of fuzzy classifiers based on probabilistic fuzzy systems is considered. It is shown that the statistical properties of the training data can be used for the design of fuzzy rule based classification systems. Takagi-Sugeno type fuzzy systems are designed for estimating the underlying conditional probability density function for the data. Probabilistic rule weighting is introduced, and classifiers based on the discriminant function approach are formulated. It is shown that some of the fuzzy classifiers that have been proposed in the literature can be formulated in terms of probabilistic rule weighting. Furthermore, the relation to certainty factor approach to fuzzy classifiers is considered.