Global and local two-sample tests via regression (Q2283577)

The objective of this paper is to report on global and local tests to determine if two samples are from different multivariate distributions. Such tests have applications in a variety of machine learning areas, e.g. to detect differences in healthy and cancerous tissue, in database attribute matching and many other classification and regression problems. Under condition that two populations only differ in their means it is proved that the regression test based on Fisher's LDA achieves the same local optimality as the Hotelling's \(T^2\) test. The simulation studies are fulfilled to examine the empirical performance of the proposed tests. The empirical performance of proposed tests is validated at the datasets from Hubble Space Telescope: it is shown that the proposed approach can identify galaxies with specific features of star-forming galaxies.