Three-Step Performance Assessment of a Pedestrian Crossing Time Prediction Model

Abstract

Pedestrian behavior and safety are emerging issues in current transportation. One way to safely study pedestrian dynamics, especially at potential conflict points such as crosswalks, is through micro-simulation. This tool provides the opportunity to repeatedly study pedestrian behavior and safety under different scenarios of interest. However, to obtain reliable results, micro-simulation models need to be calibrated and their parameters fine-tuned. One way to methodically calibrate these models is to identify the outcomes of interest, develop a predictive model for those specific outcomes, and use it as a tool to fine-tune the input parameters of the micro-simulation model. To be reliable, the results of the predictive model should be comparable to those of the micro-simulation model, and these should be validated. The aim of this research is to present a predictive model of pedestrian behavior and to evaluate this model and a conventional micro-simulation model developed using Vissim/Viswalk, given that the chosen common output is pedestrian crossing time. To achieve this goal, a multi-step procedure is followed, which is part of a more general methodological framework for calibrating the Vissim/ Viswalk micro-simulation model. This evaluation consisted in a threestep validation procedure, i.e. visual, conceptual and operational validation. Operational (statistical) validation was performed by comparing the variances of the results to understand whether the predicted sample is representative of the simulated sample. A correlation of 97% have been found between the predicted and micro-simulated crossing time values, with mean values of 6.41s and 6.32s for the simulated and predicted crossing times, respectively. Furthermore, both the predicted and simulated crossing time values fall within the ranges found in the literature for field measurements of this variable, indicating good agreement with real observed pedestrian behavior.