Clasificador difuso para diagnóstico de enfermedades

Abstract

This paper presents the application of a new fuzzy identification method to solve classification problems. The model or fuzzy classifier, obtained after training process, contains triangular sets with 0.5 overlapping to the antecedent and singleton sets for the consequent. In the evaluation of the rules is used an average operator instead of a T-norm. The consequent are adjusted using recursive least squares. The proposed method achieves higher accuracy than others methods, using a small number of rules and parameters, without sacrificing the interpretability of the fuzzy model. The proposed approach is applied in two classic classification problems: Pima Indian Diabetic and Dermatology Problem, to show the performance of the proposed method and compare the results with other researchers.