Predicción espacio temporal de delitos en la provincia de Chimborazo (2015 - 2018)

Abstract

An alternative to estimate criminal events through spatio-temporal regression models by areas was proposed, to provide information to optimize decision making. The aim was to characterize crimes in the province of Chimborazo through spatio-temporal point processes, which allowed to identify the behavior of crime patterns, to finally provide predictions with the CARBayes package, which implements generalized linear mixed models through statistical inference of the theoretical modeling and its explanatory capabilities in crime data. The univariate method was used for unit area data, with inference in a Bayesian setting, using Markov chain Monte Carlo (MCMC) simulation and R tool, where it was found that robberies have a high clustering pattern in urban areas, and the adjustment predicts with a 95% reliability by using spatial and temporal covariates surrounding the observed parishes, thus determining that the highest number of crimes occur in Riobamba, with 82% of the data recorded, with high temporal periods in the months of January, March and June, on Wednesdays, Thursdays, Fridays and Saturdays, especially in the afternoon with a 35% occurrence rate; followed by other cantons respectively, where 33% are robberies to homes, 27% to individuals, 22% theft of goods, 9% theft of economic units, etc., urban areas are more likely to suffer from crime than rural areas, the spatial covariates (number of ecu-911 cameras, number of community police units, number of financial institutions, number of educational units, etc.) allow to estimate very good models with minimal error, thus, the use of this statistical technique and the aforementioned spatial covariates is recommended for the explanation of crime data due to, they contribute significantly to the explanation of crime events.