Interpretability issues in data-based learning of fuzzy systems

Abstract

This paper presents a method for an automatic and complete design of fuzzy systems from data. The main objective is to build fuzzy systems with a user-controllable trade-off between accuracy and interpretability. Whereas criteria for accuracy mostly follow straightforwardly from the application, definition of interpretability and its criteria are subject to controversial discussion. For this reason, a set of interpretability criteria is given which guide the design process. Consequently, interpretability is maintained by structural choices regarding the type of membership functions, rules, and inference mechanism, on the one hand, and by including interpretability criteria in the rule/rule base evaluation, on the other hand. An application in Instrumented Gait Analysis, to characterize a certain group of patients in comparison to healthy subjects, illustrates the proposed algorithm.