On a multivariate homogeneity test based on one-class support vector machines

Abstract

In this paper a new multivariate homogeneity test (whether two sets of observations arise from the same distribution) is proposed. The test is based on concepts of minimum volume sets and one-class support vector machines. The test statistic as a matter of fact is test statistic for binomial proportions. We conducted experimental comparison our test with some statistical tests such as Hotelling test (the classical test for testing the mean difference for two multivariate populations) [1], multivariate rank test [8] and kernel Cramer test [2]. For experiments we used package R [9] — open source environment for statistical computing and graphics which is freely available for most computing platforms.