A feature set for spatial behavior characterization

Collection of GPS data is becoming a standard experimental method for studies ranging from public health interventions to studying the browsing behavior of large non-human mammals. However, the millions of records collected in these studies do not lend themselves to traditional geographic analysis. Standardized feature sets likely to produce distinct classes or clusters may be a tool that is powerful in both end-use utility and model describability. In this paper we present a feature set drawn from three different mathematical heritages: the convex hull of activity space, the fractal dimension of the recorded GPS traces, and the entropy rate of individual paths. We analyze these features against three human mobility datasets. Taken together these features can distinguish datasets with known demographic or geographic differences, while equating datasets which have similar demography and geography.