DEVELOPING A COPULA-BASED FRAMEWORK FOR MODELLING PEDESTRIAN CRASH FREQUENCY

Abstract

Road traffic accidents cause significant damage to property and loss of life. The incidence of vehicular collisions on roadways presents a substantial hazard to human life and constitutes a primary cause of preventable deaths worldwide. A matter of noteworthy apprehension pertains to occurrences that entail vehicular-pedestrian collisions transpiring within demarcated crosswalks. The utilization of crash data-driven research as a reactive approach to assess road safety is not without limitations, as it is subject to sporadic crash incidence and inconsistent timing of crash reporting. The aim of this study is to construct a predictive framework that estimates the incidence rate of pedestrian collisions at unsignalized intersections through an analysis of the conflicts that emerge from interactions between pedestrians and vehicles. The current study involved a comprehensive examination of crash hotspots to identify suitable sites for data collection. Based on the findings of this analysis, Ludhiana was selected as the site for data collection. The video-based traffic data underwent additional analysis to derive appropriate Surrogate Safety Measures (SSMs), namely, Post-Encroachment Time (PET) and Time to Collision (TTC). The PET and TTC values were subsequently employed in the development of univariate and bivariate Extreme Value Theory (EVT) models, utilizing Copula based techniques. The study determined that the univariate models produced inadequate outcomes with regard to the projected frequency of pedestrian accidents. The study further revealed that the bivariate Copula models demonstrated remarkable outcomes, particularly with the Gumbel-Hougaard and Joe Copulas, which exhibited exceptional fits for sites 1 and 2, respectively. The Copula models exhibited exceptional predictive accuracy, as evidenced by their remarkably low Mean Absolute Percentage Error (MAPE) values of 0.6% and 2.64% for both study sites.