Discriminant analysis for two data sets in IRd with probability densities f and g can be based on the estimation of the set G = {x : f(x) ≥ g(x)}. We consider applications where it is appropriate to assume that the region G has a smooth boundary. In particular, this assumption makes sense if...