Enhanced spatial clustering for crime analysis: Novel advances in Ward-like and SKATER algorithms for Brazilian public security

This study investigates howspatially constrained clustering can reveal actionable crime patterns to support public–security planning. Using neighbourhood–level crime records for Recife, Brazil (2007–2015), we provide new versions of two established methods —Ward–like hierarchical clustering and the graph-based spatial kluster analysis by tree edge removal (SKATER) algorithm– so that they: (i) enforce geographical contiguity and (ii) handle mixed (qualitative and quantitative) variable types through the Gower dissimilarity. Validation with the Calinski–Harabasz, Dunn, and Davies–Bouldin indices shows that the new Ward–like method achieves the strongest compactness and inter–cluster separation, whereas the Gower– based SKATER method achieves the highest variance-based discrimination. The resulting clusters partly, but not fully, coincide with Recife’s official Integrated Security Areas (or Áreas Integradas de Segurança, AIS, in Portuguese), suggesting that a data-driven redrawing of boundaries could improve police resource allocation and the monitoring of local safety policies. All computations were performed in the R software and both the code and methods are readily transferable to other cities facing similar crime-mapping challenges.