Handling fuzzy systems' accuracy-interpretability trade-off by means of multi-objective evolutionary optimization methods - selected problems

The paper addresses several open problems regarding the automatic design of fuzzy rule-based systems (FRBSs) from data using multi-objective evolutionary optimization algorithms (MOEOAs). In particular, we propose: a) new complexity-related interpretability measure, b) efficient strong-fuzzy-partition implementation for improving semantics-related interpretability, c) special-coding-free implementation of rule base and original genetic operators for its processing, and d) implementation of our ideas in the context of well-known MOEOAs such as SPEA2 and NSGA-II. The experiments demonstrate that our approach is an effective tool for handling FRBSs' accuracy-interpretability trade-off, i.e, designing FRBSs characterized by various levels of such a trade-off (in particular, for designing highly interpretability-oriented systems of still competitive accuracy).