Multi-objective genetic learning of Fuzzy Rule-Based Systems (FRBSs) is a very prolific investigation trend. The use of more optimization objectives to cover more aspects of the fuzzy model is very convenient, but also leads to a many-objective problem that is intractable with classical algorithms. This paper proposes three distinct categories of interpretability measures that can be used for optimization. Moreover, it introduces a new interpretability measure for fuzzy tuning. The proposed metric is implemented into a state-of-the-art algorithm that includes many-objectives techniques which allow the use of more objectives without substantial degradation. The new algorithm is tested in a set of real-world regression problems with successful results.
展开▼
