Extracting Prevalent Co-location Patterns from Historic Spatial Data

Abstract

A co-location pattern is a set of spatial features whose instances frequently appear in a spatial neighborhood. The data from which the patterns are derived is often historic, yet little or no attention has been paid to the time aspects of this data. In this paper we study the problem of finding prevalent co-location patterns from time constrained spatial data. We define spatial instances with time constraints brought to a net present value, and then define weighted row instances, weighted table instances and a weighted participation index for the spatial co-location patterns. We propose two algorithms to extract prevalent co-locations from spatial data with time constraints, a w-join-based algorithm that can find all prevalent patterns, and a top-k-w algorithm to find the top k most prevalent co-location patterns. Optimization strategies for the two algorithms are presented. Finally, we show the performance of the proposed algorithms using “real+synthetic” data sets, including the effect of various parameters on the algorithms.