A self-exciting point process model for predictive policing: implementation and evaluation

The self-exciting point process (SEPP) model has recently been shown to perform well in predicting spatiotemporal crime patterns. However, this model has not been widely applied to crime data and many open questions remain about how best to implement it in a real setting. In this work, we consider a range of practical implementation details relating to the application of SEPP models to real crime data. We propose a robust protocol that optimises the performance of the method, and suggest guidelines for parameter selection.