SCPP: A Point Process--based Clustering of Spatial Visiting Patterns

AbstractA collection of individuals is represented by point patterns. Each individual is a finite set of geographical locations representing their visiting pattern to places in a region. We present SCPP, an algorithm for clustering these individuals considering the spatial patterns of their visiting locations. We adopted a probabilistic framework based on the theory of point processes that allows us to derive a non-obvious distance metric between each individual point pattern and the underlying, unobserved continuous intensity function. This metric is the Kullback-Leibler divergence between the true data-generating point process distribution and the model-generating distribution. We also introduce a theoretically based framework for the cost function to be minimized, a functional T(P) taking as arguments the probability distributions underlying the unknown clusters. We present an extensive experimental analysis to show SCPP’s effectiveness using several synthetic datasets and spatial mobility patterns from geo-tagged social media.