Improved feature selection and classification by the 2-additive fuzzy measure

This paper focusses on the investigation of a pattern recognition method ba sed on the fuzzy integral. Until now this method has used a general fuzzy m easure, which is characterized by exponential complexity. Naturally this le d to some difficulties in practical applications of this pattern recognitio n method. In this paper, a heuristic algorithm for the identification of th e 2-additive fuzzy measure, which is a particular type of k-additive fuzzy measures, is proposed. This algorithm can be used to reduce complexity of f eature selection and classifier design. A further topic considered in this paper is the development of a feature selection algorithm for the fuzzy int egral classifier. The proposed heuristic algorithm is based on two feature- evaluation criteria such as the importance and the interaction indexes. The y were earlier defined in the literature using the semantic interpretation of the fuzzy measure. To validate the proposed algorithms, the feature sele ction algorithm and the pattern recognition method based on the fuzzy integ ral are applied to a problem of acoustic quality control. (C) 1999 Elsevier Science B.V. All rights reserved.