Speaker: Xin Zhang
Abstract: In this work, we propose a model-based clustering approach for spatial data. Specifically, a partial linear model is considered, in which the non-parametric term is shared by all the locations to absorb the spatial random effect, while the linear parametric term is dependent on the cluster membership. To estimate the model, we use the bivariate spline method (Lai and Wang, 2013) to approximate the non-parametric term and a multiple-tree penalty to cluster the parametric coefficients. We conduct thorough numerical experiments to examine our method, which show that our approach outperforms existing works in terms of estimation and clustering. Lastly, we apply our method on a temperature dataset collected from remote sensors. It shows that our method could provide more reasonable clustering results compared with some existing methods and have a better prediction.