Mining Prevalent Co-location Patterns Based on Global Topological Relations

Spatial co-location pattern mining is an important branch in the spatial data mining area, which discovers subsets of spatial features whose instances are frequently located together in the geographic space. The proximity between instances is defined by a distance threshold given by the user in traditional spatial co-location pattern mining. However, the user doesn't know which distance threshold is appropriate in most cases, even for experts. Besides, different densities of instance distribution are not considered in a dataset when using a unified distance threshold to measure the proximity. Also, global topological relations of instances are ignored in mining. In this paper, we consider the global topological relations by constructing Delaunay triangulation of spatial instances and calculate a distance constraint for each instance based on the constructed Delaunay triangulation. We redefine the proximity of instances according to the distance constraint so that users don't have to worry about giving an appropriate distance threshold when mining prevalent co-location patterns. We propose a new algorithm PTB based on a proximity relationship tree P-tree which stores the proximity relationships between instances. The experimental evaluation of several real-world datasets shows that our algorithm can get better results. We also evaluate each parameter and the number of features and instances affecting the efficiency of the algorithm by using synthetic datasets.