A Latent Variable Bayesian Approach to Spatial Clustering with Background Noise

We propose a finite mixture model for clustering of the spatial data patterns. The model is based on the spatial distances between the data locations in such a way that both the distances of the points to the cluster centers and the distances of a given point to its neighbors within a defined window are involved in the model. Nevertheless, we take into consideration the background noise as well. We resort to Classification Expectation-Maximization (CEM) algorithm for both estimating the parameters and clustering the data points. We test the algorithm on some simulated data sets with different background noises and apply it to a real earthquake data recorded in Kashmir in 2005.