Self-generating Interpretable Fuzzy Rules Model from Examples

Abstract

In this paper, we propose a powerful method for automatically generating interpretable fuzzy rules model from a set of given training examples (i.e. numerical data) which are sampled from an unknown function. Self-generating fuzzy rules from examples can be used as a common method for simulation such as behavior simulation for virtual humans and CGF. Our method consists of two steps: Step 1 automatically extracts a fuzzy rule base which can approximate the unknown function with an approving accuracy by introducing a homologous Gaussian-shaped membership function. Step 2 improves its interpretability by deriving linguistic rules from fuzzy if-then rules with consequent real numbers. In this way, we achieve the balance between the accuracy and interpretability of the generated rules. Finally, we show the availability of our method by applying it to the problem of function approximation.