Tuning of a Fuzzy Classifier Derived from Data by Solving Inequalities

Shigeo ABE

IEICE TRANSACTIONS on Information and Systems   Vol.E81-D       pp.224-235
Publication Date: 1998/02/25
Online ISSN: 
Print ISSN: 0916-8532
Type of Manuscript: Category: Artificial Intelligence and Cognitive Science
artificial intelligence,  pattern recognition,  fuzzy systems,  tuning of membership functions,  

Full Text: PDF>>
Buy this Article

In this paper, we develop a novel method for tuning parameters known as the sensitivity parameters of membership functions used in a fuzzy classifier. The proposed method performs tuning by solving a set of inequalities. Each inequality represents a range of the ratio of the sensitivity parameters between the corresponding pair of classes. The range ensures the maximum classification rate for data of the two corresponding classes used for tuning. First, we discuss how such a set of inequalities is derived. We then propose an algorithm to solve the derived set of inequalities. We demonstrate the effectiveness of the proposed tuning method using two classification problems, namely, classification of commonly used iris data, and recognition of vehicle licence plates. The results are compared with those obtained by using the existing tuning method and with those by neural networks.