A data-driven fuzzy system for the automatic determination of fuzzy set type based on fuzziness

Abstract

Existing research mainly uses prior knowledge to set all fuzzy sets in a fuzzy system to the same type. In data-driven fuzzy modeling, automatically determining the type of fuzzy set is still a challenging problem. In this paper, a novel data-driven design method is proposed to automatically determine the types of fuzzy sets (type-1 fuzzy sets (T1-FSs) or interval type-2 fuzzy sets (IT2-FSs)) based on fuzziness. In the hybrid-type fuzzy system (HTFS), the fuzzy set type is determined by fuzziness to improve the performance of the fuzzy system, and the interpretability of the fuzzy system is determined by the integrity, distinguishability, and redundancy of the fuzzy sets. First, fuzzy clustering based on fuzzy entropy is used to initialize the rule base and the types of fuzzy sets. Second, social learning particle swarm optimization (SLPSO) is used to optimize the parameters of the HTFS, and the types of the fuzzy sets are determined by fuzziness, which means that the types of the fuzzy sets will change dynamically in the optimization process. Because the fuzzy set types change, a parameter change strategy is also proposed for adaptive fuzzy set types. Finally, the HTFS is compared with the T1 fuzzy system on 10 real-world datasets, and it is shown that the HTFS has better performance than the T1 fuzzy system. Comparing the HTFS with other advanced methods, it is shown that the HTFS has better prediction performance while taking into account the interpretability of the fuzzy set distribution.