Abstract
Spatiotemporal patterns of repeat victimization in residential burglaries provide valuable insights for predictive applications. However, these patterns become increasingly complex in large and heterogeneous environments.
When the assumption of homogeneous areas surrounding prior burglary sites is reduced, the near-repeat pattern becomes more restricted. This suggests that burglars may shift their activity to other similar regions rather than remaining close to the original location. Such movement appears to be driven by opportunity-based criteria consistent with established criminal behaviour theories.
Under this proposed strategy, certain target areas remain consistently active throughout a burglary wave. This sustained activity leaves behind a detectable “hot trail” when visualized on heat maps, indicating persistent spatial patterns rather than isolated events.
To identify these patterns, Principal Component Analysis (PCA) with varimax rotation was applied for dimensional reduction. This approach revealed “hot trails” as weighted groupings of spatial cells, referred to as burglary constellations.
The weekly time series associated with these constellations exhibit non-random behavior, making them suitable for predictive modelling. The similarity between cells and burglary profiles within each constellation, combined with their spatial representation, provides valuable information for improving risk prediction.
Overall, these findings offer practical insights for preventive policing strategies in large, diverse areas. They are consistent with recent criminological research and suggest a novel approach for integrating spatiotemporal structure into predictive policing frameworks.