Comparative Analysis of Type-1 Fuzzy Inference Systems with Different Sugeno Polynomial Orders Applied to Diagnosis Problems

Abstract

Fuzzy Logic has been implemented successfully for different kind of problems. One of the interesting problems that had been solved with Fuzzy Logic is the classification problem, however, there exist an opportunity to improve this system to be competitive in the realm of classifications problems with respect another kind of methods for example Artificial Neural Networks. The present paper is focused in a specific application of classification problems, the diagnosis systems, this problem consists in training an intelligent system to learn the relationship between symptoms and diagnosis. This kind of problems are usually based in powerful non-linear methods for example Modular Neural-Networks or complex hybrids models, however, in this paper are applied the Type-1 Takagi Sugeno Fuzzy Systems (TSK) but analyzing the improvement of their performance by increasing the order of the Sugeno polynomial, the objective is to evaluate if is possible to improve the performance of the TSK systems applied in diagnosis problems. The conventional Takagi-Sugeno Fuzzy Systems are based in the aggregation of first-order polynomial but it is interesting to observe the effect of increase the order of this polynomial, the TSK Fuzzy Diagnosis Systems are evaluated by their accuracy obtained in ten benchmark dataset of the UCI Dataset Repository, for different kind of diseases and different difficult levels.