Skip to Main Content
Spatial Co-location patterns are similar to association rules but explore more relying spatial auto-correlation. They represent subsets of Boolean spatial features whose instances are often located in close geographic proximity. Existing co-location patterns mining researches only concern the spatial attributes, and few of them can handle the huge amount of non-spatial attributes in spatial datasets. Also, they use distance threshold to define spatial neighborhood. However, it is hard to decide the distance threshold for each spatial dataset without specific prior knowledge. Moreover, spatial datasets are not usually even distributed, so a unique distance value cannot fit an irregularly distributed spatial dataset well. Here, we proposed a qualitative spatial co-location pattern, which contains both spatial and non-spatial information. And the k nearest features (k-NF) neighbourhood relation was defined to set the spatial relation between different kinds of spatial features. The k-NF set of one feature's instances was used to evaluate close relationship to the other features. To find qualitative co-location patterns in large spatial datasets, some formal definitions were given, and a QuCOM (Qualitative spatial CO-location patterns Mining) algorithm was proposed. Experimental results on the USA thesis map data prove that QuCOM algorithm is accurate and efficient, and the patterns founded contain more interesting information.