Uma abordagem evolutiva multiobjetivo com TOPSIS para geração de sistemas de classificação fuzzy baseados em regras

Abstract

Fuzzy Rule-Based Classification Systems (FRBCSs) are widely used in classification problems. A relevant issue to be considered when generating FRBCSs is the accuracy-interpretability tradeoff, which can be addressed in the context of multiobjective optimization. Thus, in this work, we propose a new evolutionary approach to design FRBCSs in which the accuracy and the interpretability (number of rules) of the FRBCSs are considered objectives to be treated with the Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS). In order to validate the proposed approach, we applied our method to several classification datasets. The results of the first experiment indicated that the normalization procedure used in TOPSIS technique influences just the accuracy of the FRBCSs, having no significant effect on the interpretability of the classifiers. The results of the second experiment indicated that the performance of the proposed method is similar to the performance of a method from literature based on a multiobjective evolutionary algorithm.