Finite Mixtures of Matrix Variate t Distributions

Abstract

Finite mixture of multivariate t distributions (Peel and McLachlan, 2000) was introduced as an alternative to the finite mixture of multivariate normal distributions to model datasets with heavy tails. In this study, we define the finite mixture of matrix variate t distributions as an extension of finite mixture of multivariate t distributions. Mixture of matrix variate t distributions can provide an alternative robust model to the mixture of matrix variate normal distributions (Viroli, 2011) for modeling matrix variate datasets with heavy tails. We give an Expectation Maximization (EM) algorithm to find the maximum likelihood (ML) estimators for the parameters of interest. We also provide a small simulation study to illustrate the performance of the proposed EM algorithm for finding estimates.