A Constraint Neighborhood Based Approach for Co-location Pattern Mining

Driven by the ever increasing amount of spatial data collected by observations and GPS-enabled devices, mining such data for interesting or previously unknown patterns has become a major challenge. Among the many possible patterns, co-location patterns describing the frequently occurring spatial proximity of objects possessing some features are of particular interest. While several approaches have been proposed to discover such patterns, so called self co-location patterns where objects having the same feature (among others) are in spatial proximity, however, have not been effectively addressed. Furthermore, most of the co-location discovery methods suffer from expensive computations, such as spatial joins. To address these problems, in this paper, we propose a novel constraint neighborhood based approach to find co-location patterns. This approach can discover both star and clique co-location patterns, including single and complex self co-locations. Based on the constraint neighborhood idea, our method neither needs to perform spatial or instance joins nor checks for cliques to find co-location instances. To demonstrate the effectiveness of our proposed framework, we conducted experiments using both real-world and synthetic data sets. As our evaluations show, the constraint neighborhood based approach outperforms the well-known joinless approach with respect to the types of co-location patterns discovered and runtime complexity.