GENERATING FUZZY RULES FOR CLASSIFICATION PROBLEMS USING GENETIC ALGORITHMS

摘要

The problem of extracting fuzzy rules for classification purposes is very important. In this paper we propose a coding scheme for chromosomes that will allow for the generation of classification rules such that the number of rules required to cover each class and the appropriate features to be included in each rule is determined by the optimal genetic solution for the problem under investigation. The coding scheme proposed does not require the clustering of features as input to the genetic solution proposed. The rules obtained are expressed in terms of linguistic variables. The main advantage of the method proposed is that it overcomes the problem of exponentially growing number of possible conjunctive prepositional rules. The results obtained are very encouraging and further refinements of the genetic solution may increase the performance significantly.