A0295
Title: Mixtures of skewed regression models for clustering spatial data
Authors: Michael Gallaugher - Baylor University (United States) [presenting]
Abstract: Over the years, data has become increasingly complex, creating the need for new analytical tools. This is especially the case in the areas of clustering and classification, as well as spatial analysis. In the case of a large and complex spatial domain with predictor-response relationships between variables, a single regression model is unlikely to hold. To address this challenge, we propose a mixture of regression models on a Markov random field combined with skewed distributions for clustering spatial data. The model identifies clusters of locations with similar predictor-response relationships. Overfitting in the number of groups is addressed by integrating skewed distributions into the error term of each component of the mixture of regressions. Parameter estimation via an EM approach will be discussed and assessed via simulation. Insurance data and water basin data will be used for illustration.
