Geo-spatial Clustering with Non-spatial Attributes and Geographic Non-overlapping Constraint: A Penalized Spatial Distance Measure

摘要

In many geography-related problems, clustering technologies are widely required to identify significant areas containing spatial objects, particularly, the object with non-spatial attributes. At most of times, the resultant geographic areas should satisfy the geographic non-overlapping constraint. That is, the areas should not be overlapped with other areas. If without non-spatial attributes, most spatial clustering approaches can obtain such results. But in the presence of non-spatial attributes, many clustering methods can not guarantee this condition, since the clustering results may be dominated in non-spatial attribute domain which can not reflect the geographic constraint. In this paper, a new spatial distance measure called penalized spatial distance (PSD) is presented, and it is proofed to satisfy the condition which can guarantee the constraint. PSD achieves this by well adjusting the spatial distance between two points according to the non-spatial attribute values between them. The clustering effectiveness of PSD incorporated with CLARANS is evaluated on both artificial data sets and a real banking analysis case. It demonstrates that PSD can effectively discover the non-spatial knowledge and contribute more reasonably to spatial clustering problem solving.