The effect of biased sampling in radial basis function networks for data mining

Abstract

Radial basis function (RBF) networks are known to have very good performance in the task of data mining of classification, and k-means clustering algorithm is often used to determine the centers and radii of the radial basis functions of the networks. Among many parameters the performance of generated RBF networks depends upon given training data sets very much, so we want to find some better classification models from the given data set. We used biased samples as well as conventional samples to find better classification models of RBF networks. Experiments with real world data sets showed successful results that biased samples could find some better knowledge models in some classes and conventional samples also could find some better knowledge models in some other classes so that we can take advantage of the results. © 2011 Springer-Verlag.