Investigating the impacts of training data set length (T) and the aggregation unit size (M) on the accuracy of the self-exciting point process (SEPP) hotspot method

This study examines the impacts of two variables; the training data lengths (T) and the aggregation unit sizes (M); on the accuracy of the self-exciting point process (SEPP) model during crime prediction. A case study of three crime types in the South Chicago area is presented, in which different combinations of values of T and M are used for 100 daily consecutive crime predictions. The results showed two important points regarding the SEPP model: first is that large values of T are likely to improve the accuracy of the SEPP model and second is that, a small aggregation unit, such as a 50m x 50m grid, is better in terms of capturing local repeat and near-repeat patterns of crimes.