Accuracy-interpretability balancing in fuzzy models based on multiobjective genetic algorithm

The aim of this paper is to propose a general methodology to improve the linguistic-accuracy trade-off of fuzzy models, applicable to any rule-based fuzzy model. Here, the neuro-fuzzy system FasArt (Fuzzy Adaptive System ART based) is used to obtain rule-based fuzzy models, as shown in previous papers and works. FasArt, however, has the usual drawbacks, from the linguistic point of view, of most (precise) fuzzy modeling methods found in the literature, so any other fuzzy modeling algorithm can be used. Then, an improvement process is carried out based on a multi-objective genetic approach, in which interpretability and accuracy are taken into account. Two case studies have been considered: a DC motor and the Box-Jenkins gas furnace data. For each one, two fuzzy models with different complexity and fuzzy nature are generated by FasArt, whose performance is a good estimation of their behavior. Then this performance is improved by a better interpretability of the knowledge attained and stored by this fuzzy model. This is carried out by a two objectives genetic approach, in which accuracy and interpretability are taken into account. It is thus possible to find a fuzzy model with enough accuracy and an adequate capacity of explanation or interpretability of knowledge acquired from data.