A Multivariate Method for Comparing N-dimensional Distributions
Jim Loudin · 2003
We propose a new multivariate method for comparing two N-dimensional distributions. We first use kernel estimation to construct probability densities for the two data sets, and then define two discriminant functions, one appropriate for the null hypothesis and another appropriate for the actual data. Distributions of the two discriminant functions at random test points are then compared using the one-dimensional K-S test. The performance of the method is illustrated with Monte Carlo data. 1.