A spatial co-location mining algorithm based on a spatial continuous field with refined road-network constraints

Urban service facilities reflect the development status of a city from a micro level. Extracting useful spatial patterns from this type of data can assist planners with the reasonable allocation. Co-location pattern mining is valid to solve this problem. Current methods are mainly implemented in a homogenous spatial field with little constraints. However, the urban service facilities are mostly distributed in a manmade spatial field with refined road-network constraints. To address this problem, we improve the traditional methods from two aspects: (1) using a network kernel density model, we replace the Euclidean distance by an accessibility indicator to measure the proximity of two spatial instances. This indicator involves the direction and network constraints in urban space. (2) We introduce a reachability weight into the calculation of the prevalent index to replace the traditional discrete approach. The above two improvements regard the truth that the movement of human in city mainly depends on the road-network in a spatial continues field. The preliminary experiments show that the algorithm is more applicable than the current methods in solving urban facility problems.