Modeling Human Decision-Making during Hurricanes: From Model to Data Collection to Prediction

Hurricanes are devastating natural disasters. To effectively plan to help people at risk during a hurricane, a model of human decision-making is needed to predict people's decisions and to potentially identify ways to influence those decisions. In this work, we propose a generative model of human decision making based on a Markov Decision Process where we explicitly model concerns, risk perception, and information. As a first step toward evaluating the model, the work presented here focuses on one step of the decision part of the model. We created a questionnaire based on the model and collect data from 2018 Hurricanes, Florence and Michael. The results show that, across hurricane data-sets that we collected, the features of the models correlate well with evacuation decisions and our model outperforms existing methods in most cases, demonstrating the validity of the proposed model.