This report presents a novel method for crime forecasts that selects an optimal grid size and orientation in combination with a scoring function that directly maximizes the Predictive Accuracy Index (PIA).
This method was one of the top performing submissions for the 2017 National Institute of Justice's (NIJ's) Crime Forecasting Challenge, winning 9 of the 20 PAI categories under the team name of PASDA. The model's performance is shown using data from the Portland Police Department, which were used by NIJ for the Crime Forecasting Challenge. These data were used by participants in the Challenge to forecast crime hotspots for four offense types (burglary, motor vehicle theft, street crime, and all calls for service) for the months of March, April, and May of 2017. Participants were asked to define a grid subject to area and geometrical constraints and to rank grid cells for each crime type over several forecasting periods. Forecasts were made for 1-week, 2-week, 1-month, 2-month, and 3-month periods. The forecasts were scored on PAI accuracy. This report first provides details on the contest, including the data used, the submission guidelines, and the evaluation metrics. The Rotational Grid PAI-Maximizing (RGPM) methodology is then presented, along with the feature engineering and models used within the RGPM framework. This is followed by an analysis of the results of the competition and the accuracy of the RGPM model. The report concludes with a discussion of the competition and directions for future research. 5 tables, 3 figures, and 20 references
Report (Technical Assistance)
Report (Grant Sponsored)
Date Published: October 1, 2017