Abstract
Spatial cluster is the set of geographical units where events concentrate. Spatial clusters provide useful information for understanding mechanism and characteristic of socioeconomic activity. A lot of methods have been proposed for cluster detection in spatial epidemiology and criminology. However, there is no existing methods that relax constraint on adjacency of geographical units that compose a spatial cluster. Constraint that requires exact adjacency may have significant impact on detected clusters, especially in the case of detailed data. In this study, we propose a new cluster-detection method that relaxes constraints on shape and adjacency of spatial clusters. Along the lines of model-based clustering, we assume spatial data arise through a probabilistic model. Employing Potts model on probabilistic model, we can embed constraints on shape in probabilistic model and relax constraints on shape and adjacency. The results of case study on mesh data of Japanese economic census show that our method can detect clusters consist of non-adjacent geographical units.
