A DESIGN METHOD OF SCALABLE FUZZY RULE-BASED SYSTEMS FOR SOLVING REGRESSION PROBLEMS

Abstract

This paper proposes an approach for handling linguistic words directly to develop an evolutionary method for designing fuzzy rule-based systems interpretable in Tarski et al.’s sense and scalable to solve dataset regression problems. This interpretability requires that the constructed fuzzy multi-granularity structures representing the currently used word sets of dataset’s attributes must be the isomorphic images of their respective semantic word sets’ structures. Furthermore, in practice, human domain knowledge are accumulated and grown over time, leading to the requrements of expanding the currently used word sets to solve their encountered problems more effectively. It suggests studying behaviors of fuzzy rule-based systems when allowing the currently used word sets of dataset’s attributes to grow while requiring the already constructed fuzzy sets based semantics of the existing linguistic words are reused. Experiments were conducted with 15 regression datasets to show the performance and advantages of the proposed method compared to the existing methods.