Spatial Scan Statistic Method for Discovering Regional Network Co-location Patterns

  Objectives  Currently, most methods for mining regional co-location patterns focus on the planar geospatial space which can hardly support the analysis on the network space such as urban roads. Therefore, a regional network co-location pattern mining method is proposed based on spatial scan statistics.

  Methods  A network-constrained path expansion method is developed to detect the candidate paths where co-location patterns could occur. These candidates are further validated using significance tests, where the null model is constructed using a network-constrained bivariate Poisson distribution.

  Results  Experiment results of simulated data and taxi datasets show that the proposed method is more effective for discovering regional co-location patterns on the network space than a baseline method.Compared with traditional methods for discovering regional co-location patterns, our proposed method fully considers the network constraint properties of events, in both phases of the candidate path detection and the significance testing.

  Conclusions  Our method can effectively assist the resource allocation of taxis within the urban road network in Beijing by analyzing the taxi supply-demand patterns in different urban districts.