Discussion
In the last decade, research on the spatial (and temporal) distribution of crime has begun to move from the task of description to that of prediction. To date, for almost all attempts at prediction, the units of analysis considered have either been areas (e.g. police beats), or regular (often arbitrarily) sized grid cells. This, however, fails to account for the effect of a key element of the urban backcloth, upon which much human activity takes place: the street network. Such activity includes the movement of ordinary citizens through places, which influences the conditions for crime and the activities of offenders and the police. With this in mind, in this paper we have introduced a network-based method for the prospective identification of crime locations. The aim of our work was to build on existing predictive approaches, while taking advantage of the theoretical and practical advantages of the network setting.