Predictive Crime Mapping: Arbitrary Grids or Street Networks?

Area of Science:

• Criminology
• Geographic Information Systems (GIS)
• Predictive Analytics

Background:

• Empirical research consistently shows crime concentrates at various spatial scales, including street segments.
• The 'law of crime concentration at places' highlights the stability of this clustering.
• Existing crime forecasting methods primarily use area-level or grid-based predictions, neglecting street-level analysis.

Purpose of the Study:

• To develop and test the accuracy of network-based crime forecasting at the street segment level.
• To compare the predictive performance of network-based methods against grid-based alternatives.
• To address the gap in street-segment-level crime prediction, crucial for resource deployment.

Main Methods:

• Utilized property crime data from a large UK metropolitan police force.
• Introduced and calibrated a network-based prospective crime mapping model.
• Compared the network model's performance against grid-based alternatives, assessing predictive accuracy measures.

Main Results:

• The calibrated network-based model demonstrated substantially higher predictive accuracy than the grid-based alternative.
• Approximately 20% more crime was identified at a 5% coverage level using the network model.
• The improvement in predictive accuracy was statistically significant across all tested coverage levels (1-10%).

Conclusions:

• Network-based crime forecasting methods are superior to grid-based alternatives for property crime prediction.
• The findings support the operational implementation of network-based methods in policing.
• Future research should explore and test more sophisticated variations of the network-based model.