Title: Flexible mixtures of factor models based on skew component distributions
Authors: Sharon Lee - University of Adelaide (Australia) [presenting]
Abstract: Flexible mixtures of skew factor analyzers are gaining increasing attention, being exploited as powerful tools for the modelling, clustering, and dimension reduction of high-dimensional data that exhibit non-normal distributional features. These models have emerged as robust generalizations of the traditional mixtures of factor analyzers, where the assumption of normality for the latent factors is relaxed to cater for skewness in the observed data. We discuss several different formulations of skew factor models and propose to adopt a very flexible form of skew distribution as the density for the component latent factors. This allows the model to accommodate various types of skewness and asymmetry in the data, including multiple arbitrary directions of skewness. As such, it encompasses a number of commonly used models as special cases, such as some versions of the skew normal and skew t-factor analyzers. Parameter estimation can be carried out by maximum likelihood via an EM-type algorithm. The usefulness and potential of the proposed model are demonstrated using both real and simulated datasets.