Forecasting crime for strategic crime analysis: a comparison of methods

• In Event: Predicting crime in space and time: Examining and understanding input, throughput and output

Abstract

Being able to forecast the future frequency of crime at city or police-district level is useful for answering a variety of questions in strategic crime analysis. For example, being able to estimate if the number of violent crimes is a city is likely to increase in the following year can be useful in helping police prepare budgets or decide on investment in particular teams or capabilities. Many methods exist for forecasting the frequency of events and are widely used in business applications, but little used in policing or crime analysis.
This paper uses a large sample of crime data from 12 different US cities to compare the accuracy of 10 different forecasting methods when used in three different realistic crime-analysis scenarios. The results show that forecasting methods can be significantly more accurate than the default approach of assuming that crime will continue at the same frequency as in the past, and that an ensemble model that aggregates multiple forecasting models produces the most accurate forecasts.

Author

• Matt Ashby, UCL