Towards More Specific Estimation of Membership Functions for Data-Driven Fuzzy Inference Systems

Many fuzzy inference systems are built estimating their parameters from data. In particular, Takagi-Sugeno systems have been used a lot in data-driven fuzzy modeling. In this paper, we investigate one step in the data-driven identification of these models, namely the antecedent estimation when fuzzy clustering is used for estimating antecedent memberships and fuzzy rules. We propose removing noise coming from cluster membership values to obtain more specific antecedent sets, which is important for the interpretability of the models. The results obtained and presented in this paper show that this additional step leads to improved performance of the fuzzy model and higher specificity of the antecedent sets.