Multivariate data approximation with preprocessing of data

Abstract

An adaptive information system is constructed in order to approximate a set of multidimensional data. To get better approximation properties a pre-processing stage of data is proposed in which the set of points, forming the multidimensional data base and called a training set TRE, undergoes a clustering analysis. In the analysis two independent clustering algorithms are used; on each cluster a feed-forward neural network is trained and a membership function of a fuzzy set is constructed. The constructed system contains a module of two-conditional fuzzy rules consequent parts of which are of the functional type. Each rule is designed on a pair of clusters.