Optimising parameter selection for predicting volume crime using hotspot mapping

Mr Timothy Mashford1, Mr Scott Davidson1

1Victoria Police, Melbourne, Australia

Policing agencies across the world regularly utilise hotspot mapping processes to identify risk and predict future crime events. The most common hotspot technique, kernel density estimation (KDE), takes a number of parameters for which intelligence practitioners often accept the default values suggested by the software. This study examines two of these parameters – search radius and input data period – and tests their impact on KDE’s ability to predict future crime events. Using a volume crime category (theft from motor vehicle), a variety of parameter value combinations are tested, with the predictive accuracy measured on a weekly basis across a 12 month period. The results demonstrate that the choice of parameter values can have a significant impact on the predictive accuracy of hotspot maps, suggesting more effort is required to educate intelligence practitioners on how these parameter values should be considered.


Biography:

Tim has worked as a geospatial analyst at Victoria Police since 2004. He currently manages the Geospatial Analysis Unit, which focuses on the application of GIS to analysing crime and road policing data, and supports intelligence practitioners across the organisation to enhance geospatial capabilities. Tim has qualifications in geospatial science and statistics, and is a certified Geographic Profiling Analyst. He has presented at crime mapping conferences in Australia, UK and the USA.