Crime forecasting in small city blocks using a general additive spatio-temporal model

Spatio-temporal modeling is widely recognized as a promising means for predicting crime patterns. Despite their enormous potential, the available methods are still in their infancy. A lot of research focuses on crime hotspot detection and geographic crime clusters, while a systematic approach to include the temporal component of the underlying crime distributions is still under-researched. In this paper, we gain further insight in predictive crime modeling by including a spatio-temporal interaction component in the prediction of residential burglaries. Based on an extensive dataset, we show that including additive space-time interactions leads to significantly better predictions.