A Brief Review of Performance and Interpretability in Fuzzy Inference Systems

Abstract

Business analytics allows understand the current state of an organization and identify the needs of the company. Currently, there are commercial and scientific tools, which focus on trying to solve some of the stages of a business analytics system based on machine learning that helps in decision making through the discovery of knowledge. Network-based learning methods solve problems that were challenging for many years but the results they produce are not interpretable. On the other hand, methods based on fuzzy logic allow the discovery of predicates and their evaluation of unseen data, which is called inference. Due to their construction based on predicates, fuzzy logic facilitates the presentation of results in an interpretable way, but with less precision than network-based methods. A general line open to research is to integrate learning methods that compensate for accuracy and interpretation. This work presents a scoping review on compensatory fuzzy logic, Archimedean compensatory fuzzy logic as well as works based on inference, accuracy, and interpretability. Also, it presents a list of lines open to research on these topics.