CONSTRUCTING FUZZY CLASSIFICATION SYSTEMS FROM WEIGHTED TRAINING PATTERNS

摘要

In this paper, we propose a fuzzy rule-generation method for pattern classification problems. We consider a situation where each training pattern has a weight. The weight is considered as a cost of misclassification/rejection of classification. Our fuzzy classification system consists of a set of fuzzy if-then rules. The antecedent part of fuzzy if-then rules linguistically specifies a subarea of a pattern space. Thus, fuzzy if then rules are linguistically interpretable. The main aim of this paper is to construct fuzzy classification systems that reflect weighted training patterns. That is, we make the cost of misclassification and rejection of classification for unseen patterns as small as possible. In computer simulations, we also propose two methods for assigning appropriate weights from the distribution of given training patterns. The performance of the proposed fuzzy rule-generation method is examined for several real-world pattern classification systems that have been used in literature.