A0582
Title: Modelling and clustering via finite mixtures of skew factor analyzers
Authors: Sharon Lee - University of Queensland (Australia) [presenting]
Geoffrey McLachlan - University of Queensland (Australia)
Abstract: Mixtures of factor analyzers (MFA) provide a popular and powerful tool for simultaneous clustering and dimension reduction. The traditional MFA assumes a joint normal distribution for the latent factors and errors. We consider a robust generalization of the MFA model where the latent factors and errors jointly follow the so-called canonical fundamental skew t-distribution (CFUST). This approach provides an effective way for modelling and clustering high-dimensional heterogeneous data that exhibit non-normal features. The adoption of a CFUST distribution allows the mixtures of CFUST factor analyzers model to accommodate flexible skewness, encompassing a number of commonly used models as special and/or limiting cases. Parameter estimation can be carried out by maximum likelihood via an EM-type algorithm. The usefulness of the proposed methodology is demonstrated using simulated and real datasets.
