A Local Distribution Based Spatial Clustering Algorithm

All existing spatial clustering methods only utilize the spatial distances and the numbers of entities in the spatial nearest neighborhood to find the spatial clusters in the spatial database,without taking the spatial local distribution characters into account.Hence,the clustered results are unreasonable in many cases.To overcome such limitations,we propose a new spatial clustering algorithm based upon the local distribution among the entities in certain spatial neighborhood,where median angle for each entity is employed to measure its property of local distribution.In the process of spatial clustering,a series of recursive search were implemented for all the entities so that those entities with its median angle being very similar are clustered.Two tests were implemented to demonstrate that the proposed method is more prominent than DBSCAN,and can be used to find the clusters with arbitrary shapes.