A single-objective genetic-fuzzy approach for multi-objective fuzzy problems

Article type: Research Article

Authors: Kaya, Ersin | Koçer, Barış | Arslan, Ahmet

Affiliations: Department of Computer Engineering, Faculty of Engineering and Architecture, Selçuk University, Konya, Turkey

Note: [] Corresponding author. Ersin Kaya, Department of Computer Engineering, Faculty of Engineering and Architecture, Selçuk University, Konya 42075, Turkey. Tel.: +90 332 223 33 33; Fax: +90 332 233 00 64; E-mail: [email protected]

Abstract: In this paper, a genetic algorithm-based search method, which builds ideal rule set for fuzzy rule-based classification systems (FRBCSs), is developed. In FRBCSs, ideal rule set means a set of rules which ensure high classification accuracy with small rule count and small rule length. The related studies in the literature point out that rule set grows exponentially with input attribute count. This growth complicates the searching process and lowers the success rate. Through the proposed method, successive results are obtained for datasets with large input attribute counts using a special coding technique. The proposed method is tested for various datasets and results are compared against the method which uses candidate rule set.

Keywords: Fuzzy rule-based classification, genetic algorithm, generating fuzzy rules

DOI: 10.3233/IFS-120661

Journal: Journal of Intelligent & Fuzzy Systems, vol. 25, no. 3, pp. 557-566, 2013