Data-driven modelling of police route choice

In recent years, increasing digitisation of police patrol activities has enabled new insights into police patrol behaviour. This paper explores digitised traces of police patrol journeys in order to understand police routing preferences and to propose models that enable simulations of patrol behaviour. The models assume that police journeys are undertaken as series of “topics”, which are inferred from vehicle GPS data using a widely used topic modelling technique called latent Dirichlet allocation. Initial experiments have shown that they are capable of reproducing police coverage patterns that would not be captured by alternative models assuming optimal behaviour.