A Novel Approach to Generating fuzzy Rules Based on Dynamic Fuzzy Rough Sets

We propose a novel approach to generating fuzzy rules which, different from most known fuzzy rules induction, is not based on attributes reduction (AR) but granulation order and variational universe. Most rule induction algorithms based on fuzzy rough sets (FRS) usually include two steps: AR and fuzzy rules induction. It's helpful to shorten the time of rule mining to some extent by AR. However, AR may make against the induction of fine rules due to its limitation. Avoiding AR in fuzzy rules induction permits to improve the adaptability of fuzzy rules and reduce computational complexity. In this paper, the dynamic FRS is presented in two different ways. Then, an algorithm based on dynamic FRS is put forward for decision rule mining. At last, an initial experimentation is conducted, comparing the new method with a conventional FRS-based rule mining.