Spatiotemporal analysis of homicides in Ecuador: Application of DBSCAN and Random Forest for hotspot detection

Abstract

The objective of this research was to develop a geospatial and predictive analysis model to identify high-incidence homicide areas in Ecuador for strategic planning in public security. The geospatial distribution of homicides in the country during 2024 was analyzed using the official database provided by the Ministry of the Interior, with a particular focus on the provinces of Guayas and Manabí, which report the highest rates of violence. A combined methodology was implemented, integrating the unsupervised clustering algorithm DBSCAN to detect spatial homicide clusters and the supervised prediction model Random Forest to anticipate the likelihood of new cases belonging to the identified hot zones. The analysis revealed significant crime clusters in urban and peri-urban areas, as well as the presence of scattered homicides classified as noise, suggesting complex or atypical patterns. The predictive model achieved high levels of accuracy (F1-score of 0.9987), highlighting the importance of geospatial and demographic variables in classification. The results are consistent with national and international research and confirm that spatiotemporal analysis using machine learning techniques provides valuable tools for targeting security interventions.