Knowledge Base Generation Based on Fuzzy Clustering

Abstract

An approach for generating an optimal rule base of a fuzzy system is proposed, that relies on ellipsoidal clustering of observable data. Premises of fuzzy rules are formed by constructing projections of ellipsoids onto the coordinate axes while conclusions are formed either using ellipsoid axes or also based on projections. The idea of optimization is in using ellipsoids of the minimal volume that includes all points of the cluster. A comparative analysis of various ways to choose optimal parameters for ellipsoids covering the clusters is performed. The root-mean-square error is used to estimate the approximation accuracy of the resulting fuzzy system.