Mixtures of common t-factor analyzers for modeling high-dimensional data with missing values

Mixtures of common t-factor analyzers (MCtFA) have emerged as a sound parsimonious model-based tool for robust modeling of high-dimensional data in the presence of fat-tailed noises and atypical observations. This paper presents a generalization of MCtFA to accommodate missing values as they frequently occur in many scientific researches. Under a missing at random mechanism, a computationally efficient Expectation Conditional Maximization Either (ECME) algorithm is developed for parameter estimation. The techniques for visualization of the data, classification of new individuals, and imputation of missing values under an incomplete-data structure of MCtFA are also investigated. Illustrative examples concerning the analysis of real and simulated data sets are presented to describe the usefulness of the proposed methodology and compare the finite sample performance with its normal counterparts.