Abstract
Although various analyses have been conducted on the resident evacuation behavior by
various viewpoints, it is still difficult to say that the problems related to the resident evacuation
have been solved because of the number of victims caused by heavy rain disasters. Therefore,
the authors have made a new attempt to analyze the factors of resident evacuation behavior
using XAI (eXplainable AI). However, there are still issues such as the estimation accuracy of
the behavioral model. In addition, previous analyses of resident evacuation behavior have often
attempted to determine whether certain factors influence evacuation behavior.
In this study, the prediction accuracy was improved by converting questionnaire survey data
into image data and constructing a resident evacuation behavior model using a convolutional
neural network (CNN). we analyzed the factors of resident evacuation behavior during heavy rain
disaster by using Grad-CAM that is a method of XAI in image recognition technology. As a result,
the combination of the following factors were found to influence evacuation behavior during a
disaster: obtaining evacuation information from a trusted person, understanding disaster prevention
information deeply, preparing emergency supplies, and discussing with family members.
