Fuzzy rule generation by hyperellipsoidal clustering

摘要

This paper considers the generation of fuzzy rules each of which consists of a fuzzy proposition as the premise part and linear models as the consequence part. The fuzzy modeling involves some interdependent problems that are the detection of linear substructures in a given multi-dimensional data set, and the identification of the consequence parts and the membership functions of premise variables. Because it is difficult to solve these problems simultaneously, we treat these problems separately in this paper. The proposals are related to the detection of linear substructures in data by the hyperellipsoidal clustering, and the development of multi-dimensional member-ship functions that take into account the correlation between variables.