Log-Gaussian Cox Processes for Spatiotemporal Traffic Fatality Estimation in Addis Ababa

Abstract

We investigate the spatiotemporal dynamics of traffic accidents in Addis Ababa, Ethiopia, using 2016--2019 data. We formulate the traffic accident intensity as a log-Gaussian Cox Process and model it as a spatiotemporal point process with and without fixed and random effect components that incorporate possible covariates and spatial correlation information. The covariate includes population density and distance of accident locations from schools, from markets, from bus stops and from worship places. We estimate the posterior of the state variables using integrated nested Laplace approximations with stochastic partial differential equations approach by considering Matèrn prior. Deviance and Watanabe - Akaike information criteria are used to check the performance of the models. We implement the methodology to map traffic accident intensity over Addis Ababa entirely and on its road networks and visualize the potential traffic accident hotspot areas. The comparison of the observation with the model output reveals that the covariates considered has significant effect for the accident intensity. Moreover, the information criteria results reveal the model with covariate performs well compared with the model without covariates. We obtained temporal correlation of the log-intensity as 0.78 indicating the existence of similar traffic fatality trend in space during the study period.