Specifying Distinct Groups in Multivariate Data using Machine Learning
When working with model-based classifications, finite mixture models are utilized to describe the distributions of the resulting groupings of observations within a data set. These groupings are derived using minimal prior knowledge of this group structure. Mixtures of multiple scaled multivariate-t and SAL distributions possess properties that have been shown to provide excellent classification and improved inferential capabilities for several data sets. However, the number of parameters in these models scale quadratically, leading to models becoming highly parametrized. Hence, the intention of this research project is to develop an efficient variation of an expectation maximization (EM) algorithm to fit mixtures of these distributions. The classification of the models resulting from this algorithm will be compared to other state-of the-art mixtures using simulated and real data sets.
Faculty Mentor: Brian Franczak