Increasing the explainability and trustiness of Wang-Mendel fuzzy system for classification problems

Abstract

Fuzzy systems are the perfect systems for knowledge representation-the fuzzy rules on which they are based can be interpreted and explained during the presentation of the system’s results. In this paper, a new method for fuzzy system optimization is proposed in which the fuzzy rules are built with the Wang–Mendel approach. The proposed method allows taking into account a new additional criterion during optimization - trustiness, which is aimed at obtaining better trust in the created knowledge base. In addition, a new criterion of explainability and explainable interface are proposed, which are aimed at presenting fuzzy rules in an explainable form. The proposed approach was tested in many variants and compared to methods from the literature. The results of the simulations showed that not only the accuracy of the classification was improved, but also the resulting rules had less significant conflicts, and the explanation of the classification was more transparent. Such significant results confirm the effectiveness of the proposed method-it not only allowed for the simultaneous improvement of the three mentioned elements without any trade-offs compared to known methods but also provided further development potential in this direction.