Fuzzy genetic algorithm approach to feature selection problem

Author

Fung, George S K ; Liu, James N K ; Chan, K.H. ; Lau, Rynson W.H.

Author_Institution

Dept. of Comput., Hong Kong Polytech., Hung Hom, Hong Kong

Volume

1

fYear

1997

fDate

1-5 Jul 1997

Firstpage

441

Abstract

The primary role of a genetic algorithm is in the selective breeding of a population of individuals. A suboptimal solution can be obtained by such an algorithm which applies the paradigm of various evolutionary selection and searches through many generations via different genetic operators. The selection of features associated with each individual is based on the fitness-proportionate selection in which parents are chosen from the population. This fitness is a problem-specific property that describes an individual´s performance upon some chosen features quantitatively. This paper describes a formal fuzzy genetic algorithm to overcome the traditional problems in feature classification and selection and provides fuzzy templates for the identification of the smallest subset of features. Simulation results demonstrate that the operation using soft crossover significantly improves the searching power through the multi-dimensional feature space

Keywords

feature extraction; fuzzy set theory; genetic algorithms; pattern classification; search problems; evolutionary selection; feature selection problem; fitness-proportionate selection; fuzzy genetic algorithm; fuzzy templates; multi-dimensional feature space; selective breeding; suboptimal solution; Computational efficiency; Computational modeling; Data analysis; Error analysis; Facial features; Genetic algorithms; Handwriting recognition; Image classification; Monte Carlo methods; Robustness;

fLanguage

English

Publisher

ieee

Conference_Titel

Fuzzy Systems, 1997., Proceedings of the Sixth IEEE International Conference on

Conference_Location

Barcelona

Print_ISBN

0-7803-3796-4

Type

conf

DOI

10.1109/FUZZY.1997.616408

Filename

616408