Spatial entropy-based clustering for mining data with spatial correlation

Due to the inherent characteristics of spatial datasets, spatial clustering methods need to consider spatial attributes, nonspatial attributes and spatial correlation among non-spatial attributes across space. However, most existing spatial clustering methods ignore spatial correlation, considering spatial and non-spatial attributes independently. In this paper, we first prove that spatial entropy is a monotonic decreasing function for non-spatial attribute similarity and spatial correlation. Then we propose a novel density-based spatial clustering method called SEClu, which applies spatial entropy in measuring non-spatial attribute similarity and spatial correlation during the clustering process. The experimental results from both the synthetic data and the real application demonstrate that SEClu can effectively identify spatial clusters with spatial correlated patterns.