Modeling Pedestrian Evacuation Response for Tsunami Events

摘要

Most pedestrian evacuation models have restrictive assumptions that do not allow one to accurately depict the wide array of responses in the event of a tsunami. This work will describe the use of survey data on the at-risk population in Rincon, PR in order to relax an assumption found in most pedestrian evacuation models-all individuals evacuate immediately. It is well known that there are a variety of factors that will prevent this from happening. For example, a single mother separated from her children might choose to pick them up at school before evacuating, as opposed to evacuating immediately if she were with them at the time. In this work, a prediction model will be used to predict the evacuation response of individuals. The factors that will be considered in the model include: age group, gender, resident status (resident of Rincon vs. tourist), the individual's total number in household, and whether the individual is with his or her dependent population in the event of an evacuation. After evaluating a variety of standalone classifiers and meta learners, results in a five-fold cross validation setting indicate the random forest learner is the top performer.