Análisis y predicción de accidentes mensuales de tránsito en la provincia de Chimborazo en el período enero 2015 – diciembre 2021

Abstract

The objective of this research was to predict the number of monthly traffic accidents that enter the Provincial Prosecutor's Office of Chimborazo (FPCH), through generalized linear models of ARIMA which were compared with Poisson models for reliable decision making and to improve the service of the FPCH. For the development of the study, monthly data, provided by the Procedural Management unit of the FPCH, of the period January 2015 – December 2021 was used, of which the 80-20 rule was applied, where the most significant variables were obtained: Material Damage (DM) since it represents 58% of accidents and injuries caused by traffic accidents (LCPATR) with 28% of traffic accidents. As results in the exploratory analysis of univariate data, the presence of atypical data was observed, the decomposition of the time series was done, thus verifying the presence of seasonality. It is concluded that between the ARIMA and POISSON models using the Akaike Information Criterion (AIC) and the Bayesian Information Criterion (BIC) as the model with the best predictive quality ARIMA (0,0,0) (2,1,2) [12] for the DM variable and ARIMA (1,0,1) (1,0,1) [12] for the LCPATR variable with which the respective predictions were made. It is recommended that the FPCH use more variables in order to apply multivariate methods that adjust to real situations.