Comparison of Alternative Trip Generation Models for Hurricane Evacuation

Abstract

The purpose of this study was to compare the relative accuracy of alternative forms of trip generation of evacuation traffic. Conventional participation rate, logistic regression, and various forms of neural network models were estimated and tested using a data set of evacuation behavior collected in southwest Louisiana following Hurricane Andrew. The data set was divided into a 350-household data base on which the logistic regression and neural network models were estimated, and a separate 60-household data base on which all models were tested. Limited and comprehensive model inputs were tested among the neural network models to determine whether more comprehensive specifications enhance the performance of the models. It was found that the limited specification performed almost as well as the more detailed specification. Comparison of the performance of the models considered in this study showed that the logistic regression and neural network models were able to predict evacuation more accurately than the participation rate model.