Formulation and analysis of pedestrian safety problems using Bayesian network model
Date
2010
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
University of Delaware
Abstract
Causes of pedestrian road accident have been a major concern to transportation engineers and other road safety professionals despite all efforts being applied to alleviate this problem. Although studies have aimed at modeling and analyzing the causes of pedestrian road accidents, the bulk of these studies have been found to be too stochastically oriented and more macroscopic than it is necessary. Consequently, the existing models seldom incorporate the interactions between pedestrians and their immediate environment. In this study, pedestrian crossing behavior during spring and summer season has been thoroughly investigated using Bayesian network modeling technique. The model was constructed with variables known to influence pedestrian crossing behavior either directly or indirectly. Stages of the model building process including Graphical Level (GL), Information or Qualitative Level (IL) and Quantitative Level (QL) have been discussed and implemented to extract useful information from both observed data and data elicited from stakeholders‘ opinion as well as experts‘ experience. The robustness of the Bayesian network model is compared based on its ability to produce physically meaningful results that truly reflects realistic behavior of a system. The model‘s results show that pedestrians often exhibit rational crossing behavior than they do irrationally and such an attitude is found to be influenced mostly by their own motives and less by external factors even though roadway environment did not favored them. Also, a sensitivity analysis carried out revealed that signal timing phase length is the most influential parameter that affects pedestrian crossing behavior.