Simulation‐Tested Spatial Association Mining of Co‐Location Patterns From Multiple Point‐Feature Classes

Are certain types of point features frequently close to or away from each other or in a recurring motif? When identified, it piques curiosity to inquire about the drivers, contextual factors, and the functional relations responsible for and the prediction of such a spatial association. Co‐location, characterized by feature classes that frequently occur in proximity, is a fundamental type of spatial association. Spatial statistical approaches (e.g., K‐function or Poisson regression methods) quickly reach their computational capacity to cross‐examine co‐location for many‐to‐many relationships among feature classes. While data mining offers computational efficiency, it lacks statistical significance on identified co‐located feature classes. By combining strengths from both statistical and mining approaches, we developed the Simulation‐tested Spatial Association Mining (SimSAM) algorithm. Two controlled simulations and an empirical study demonstrated the algorithm's capabilities in identifying co‐location patterns with statistical significance for simulated point‐feature classes and for real‐world Points of Interest and social event types in the City of Dallas.