Abstract:

Using a given data set (the Kefallinia erosion data) with only 3 dimensions and with fractal correlation dimension rGP ¼ 1:60, we wanted to see, what really by the kernel methods is provided.We have used Gaussian kernels with various kernel width ¾. In particular we wanted to ¯nd out, whether the GDA (Generalized Discrimination Analysis) as proposed by Baudat and Anouar (2000), permits to distinguish better the high, medium and low erosion classes as compared to the
classical Fisherian discriminant analysis. The general result is that the GDA yields discriminant variates permitting for better differentiation among groups, however the calculations are more lengthy.