GEnI-FR: um método para assegurar a interpretabilidade de regras fuzzy por meio de granularidade.

Abstract

The computational representation of human knowledge, when composed of imprecise data, is a task facilitated by the use of systems that use resources from set theory and fuzzy logic. Such systems use rules that allow mapping the knowledge obtained from data into an easy-to-understand linguistic representation. The fuzzy rule base that constitutes this type of system can be generated by a specialist or through methods that consider the characteristics of the data itself, reducing the need for a specialist in this creation process. With the appropriate adjustments and use of automated methods, it is possible to increase the interpretability of these rules without reducing the system’s accuracy. In this sense, the rule base of a fuzzy system can be improved through the principle of justifiable granularity to adjust the fuzzy sets representative of the data. Therefore, in this master’s thesis, the method called GEnI-FR (Granularity to Ensure Interpretability of the Fuzzy Rules) is proposed, which makes adjustments and refinements in the process of generating fuzzy rules, achieving a balance between interpretability and precision, adjusting sets fuzzy based on the characteristics of the data itself. GEnI-FR presents itself as a promising method as it provides a reduction in the number of fuzzy rules while maintaining the same levels of accuracy when compared to other state-of-the-art methods.