Cluster detection in uncertain point distributions: a comparison of four methods

Abstract

Uncertainty is unavoidable in spatial data. Though this fact is widely recognized in the GIS community, it is often assumed in spatial analysis that spatial data are accurate and thus analyses of the data are reliable, which is usually not the case. It is necessary to discuss from both theoretical and practical viewpoints how uncertainty in spatial data affects the results of spatial analysis. To fill the gap in the research, this paper discusses the accuracy of spatial analysis based on uncertain spatial data, focusing on cluster detection in point distributions. Four methods of cluster detection in uncertain point distributions are proposed: (1) centroid method; (2) minimum method; (3) maximum method; and (4) statistical method. They are evaluated in terms of their accuracy and efficiency of computation through numerical simulations. Some empirical findings are shown which are useful for choosing a method of cluster detection and estimating its accuracy.