The simple spatial disaggregation approach to spatio-temporal crime forecasting

Reported criminal damage incidences in the City of Cardiff, UK were modelled using an avant-garde statistical modelling technique, namely the Hierarchical Profiling Approach (HPA). With HPA, salient events affecting crime levels are identified and their influences modelled, thus providing a catalogue of profiled events. These profiles are incorporated into crime model. The model is subsequently spatially disaggregated using the crime forecasts over the entire study area. The Simple Spatial Disaggregation Approach (SSDA), developed within a Geographical Information System (GIS) environment, utilises four cluster forecasting techniques to provide a best cluster forecast. The hot spot detection routine procedure utilised was the Spatial and Temporal Analysis of Crime (STAC) technique which ignores fixed administrative boundaries. The spatio-temporal errors of the forecasts were then calculated and compared to one another using novel two-dimensional error evaluation techniques, namely the Spatio-temporal Mean Absolute Percentage Error (STMAPE) and the Spatio-Temporal Mean Root Squared Error (STMRSE). An average prediction error of +/- 3.51 crimes per day for the city as a whole was obtained using HPA alongside an optimum STMRSE of 1.131 errors per weekday per cluster with SSDA.