Measuring the rationality in evacuation behavior with deep learning

TL;DR

This paper introduces a deep learning framework using CNNs to quantify deviations from rationality in evacuation behavior, demonstrating robustness to incomplete data and potential for broader application in virtual experiments.

Contribution

A novel deep learning approach to measure rationality deviations in evacuation models, with robustness to incomplete information and generalizability to other virtual scenarios.

Findings

CNN accurately predicts rational factors from CA model images

Framework remains effective with incomplete or partial images

Method can be applied to other virtual experimental setups

Abstract

The bounded rationality is a crucial component in human behaviors. It plays a key role in the typical collective behavior of evacuation, in which the heterogeneous information leads to the deviation of rational choices. In this study, we propose a deep learning framework to extract the quantitative deviation which emerges in a cellular automaton (CA) model describing the evacuation. The well-trained deep convolutional neural networks (CNNs) accurately predict the rational factors from multi-frame images generated by the CA model. In addition, it should be noted that the performance of this machine is robust to the incomplete images corresponding to global information loss. Moreover, this framework provides us with a playground in which the rationality is measured in evacuation and the scheme could also be generalized to other well-designed virtual experiments.