Abstract
Spatial co-location and de-location patterns represent subsets of Boolean spatial feature types whose instances are often located in close/separate geographic proximity. Existing literatures pay more attention on mining colocation patterns based on distance threshold spatial relation. In this paper, we proposed a novel co-location and de-location patterns mining algorithm (CODEM) to discover useful co-location and de-location patterns in large spatial datasets. We used k nearest features (k-NF)to measure spatial close/separate relationships of colocation/de-location patterns in spatial datasets. The k-NF set of one feature type's instances was used to evaluate the close/separation relationship between other features and one feature. Then, a correlation checking operation was adopted to filter the uninteresting patterns, and moreover a grid index method was used to accelerate the k nearest features query, while a T-tree (Total support tree) structure was also used to compress the candidate frequent and infrequent item sets, and generate patterns efficiently. Experimental results prove that the algorithm is accurate and efficient, has a time complexity of O(n).