Efficiently Mining Colocation Patterns for Range Query

Abstract

Colocation pattern mining finds a set of features whose instances frequently appear nearby in the same geographical space. Most of the existing algorithms for colocation patterns find nearby objects by a user-provided single-distance threshold. The value of the distance threshold is data specific and choosing a suitable distance for a user is not easy. In most real-world scenarios, it is rather meant to define spatial proximity by a distance range. It also provides flexibility to observe the change in the colocation patterns with distance and interprets the result better. Algorithms for mining colocations with a single distance threshold cannot be applied directly to the range of distances due to the computational overhead. We identify several structural properties of the collocation patterns and use them to propose an efficient single-pass colocation mining algorithm for distance range query, namely Range−CoMine. We compare the performance of the Range−CoMine with adapted versions of the famous Join-less colocation mining approach using both real-world and synthetic data sets and show that Range−CoMine outperforms the other algorithms.