Model-Based Clustering for Spatio-Temporal Data

The explosive growth in spatial and temporal data has lead to the emergence of spatio-temporal clustering. However, very few methods adopt model-based approaches for such data. We introduce a new mixture for clustering spatio-temporal data, called STM, for which the generic identifiability is proved. The resulting model defines each mixture component as a mixture of autoregressive polynomial regressions in which the weights consider the spatial and temporal information with logistic links. Under the maximum likelihood framework, parameter estimation is carried out via an expectation-maximization algorithm while the classical information criteria can be used for model selection. To illustrate the benefits of the proposed approach, two challenging applications are conducted.