Towards Efficient Multi-objective Genetic Takagi-Sugeno Fuzzy Systems for High Dimensional Problems

Multi-objective genetic Takagi-Sugeno (TS) fuzzy systems use multiobjective evolutionary algorithms to generate a set of fuzzy rule-based systems of the TS type with different trade-offs between, generally, complexity/interpretability and accuracy. The application of these algorithms requires a large number of TS system generations and evaluations.When we deal with high dimensional data sets, these tasks can be very time-consuming, thus making an adequate exploration of the search space very problematic. In this chapter, we propose two techniques to speed up generation and evaluation of TS systems. The first technique aims to speed up the identification of the consequent parameters of the TS rules, one of the most timeconsuming phases in TS generation. The application of this technique produces as a side-effect a decoupling of the rules in the TS system. Thus, modifications in a rule do not affect the other rules. Exploiting this property, the second technique proposes to store specific values used in the parents, so as to reuse them in the offspring and to avoid wasting time. We show the advantages of the proposed method in terms of computing time saving and improved search space exploration through two examples of multi-objective genetic learning of compact and accurate TS-type fuzzy systems for a high dimensional data set in the regression and time series forecasting domains.