Title: Mixture modelling with scale mixtures of skew normal distributions
Authors: Sharon Lee - University of Adelaide (Australia) [presenting]
Abstract: In recent years, mixture models with skew component distributions have received increasing attention. The literature now offers a wide variety of non-normal distributions with different properties suitable for a range of applications. An overview of existing skew models is provided, focusing on those adopted in the model-based clustering literature. We then consider a very general family of skew distributions, namely, the scale mixture of canonical fundamental skew normal (SMCFUSN) distributions. This family encapsulates many important and commonly used symmetric and skew distributions including the normal, t, hyperbolic, slash, and their skew variants. Mixtures of SMCFUSN distributions can be fitted by maximum likelihood via an EM-type algorithm. Dimension reduction via a factor version of mixtures of SMCFUSN distributions is also considered. The usefulness of the approach will be demonstrated via clustering applications to some real datasets.