Cohesion based co-location pattern mining

Because of a wide range of applications, e.g., GPS applications and location based services, spatial pattern discovery is an important task in data mining. A co-location pattern is defined as a subset of spatial items whose instances are often located together in spatial proximity. Current co-location mining algorithms are unable to quantify the spatial proximity of a co-location pattern. We propose a co-location pattern miner aiming to discover co-location patterns in a multidimensional spatial structure by measuring the cohesion of a pattern. We present two ways to build the co-location pattern miner, FromOne and FromAll, in an attempt to find a balance between accuracy and runtime. Additionally, we propose a method named Fre-ball to transform a structure into a transaction database, after which any existing itemset mining algorithm can be used to find the co-location patterns. An experimental evaluation shows that FromOne and Fre-ball are more efficient than existing methods. The usefulness of our methods is demonstrated by applying them on the publicly available geographical data of the city of Antwerp in Belgium.