Mining Co-locations from Continuously Distributed Uncertain Spatial Data

A co-location pattern is a group of spatial features whose instances tend to locate together in geographic space. While traditional co-location mining focuses on discovering co-location patterns from deterministic spatial data sets, in this paper, we study the problem in the context of continuously distributed uncertain data. In particular, we aim to discover co-location patterns from uncertain spatial data where locations of spatial instances are represented as multivariate Gaussian distributions. We first formulate the problem of probabilistic co-location mining based on newly defined prevalence measures. When the locations of instances are represented as continuous variables, the major challenges of probabilistic co-location mining lie in the efficient computation of prevalence measures and the verification of the probabilistic neighborhood relationship between instances. We develop an effective probabilistic co-location mining framework integrated with optimization strategies to address the challenges. Our experiments on multiple datasets demonstrate the effectiveness of the proposed algorithm.