Hierarchical GIS clustering using principal components

GIS point-clustering is an important feature in any GIS based data visualization application. Clustering not only condenses data into visualizable units, it also presents analytical tools for data study. In this paper we use principal components analysis of underlying point data to recursively find appropriate split boundaries and partition point data into a hierarchy of cluster regions. We then group each region into a cluster point using a closest-mean approach. The result is a very efficient O[n log(n)] algorithm of linear spatial complexity to build the hierarchy. Resulting hierarchy can give instant clusters at varying map scales with logarithmic complexity. Moreover, the output cluster points represent more naturally the point density.