Data-Driven Theory Weighting in Space-Time Crime Models

• In Event: Spatial Risk Assessment, I

Abstract

Crime research often studies crime from a single vantage point resulting in multiple theories that each explain crime patterns via a different analytic technique. This process results in competing theories that often require a practitioner to choose which theoretical path to apply to a given problem based upon their beliefs of the most applicable explanation of crime. This paper will explain a methodology for the incorporation of multiple theories within a space-time model of crime based upon non-parametric machine learning techniques. The methodology results in automatic incorporation of theory interactions and measurement of the relative contribution of each set of variables. The application of this technique across various crime types and jurisdictions will be compared from two perspectives: (1) how do the trained models differ across data sets and (2) how does predictive accuracy vary?

Author

• Jeremy Heffner, Azavea