Simulating Panic in Crisis Situations Using a Neural Network

This paper proposes a neural network that can simulate panic during crisis situations. When and how people experience anxiety and panic can be different even if they are in similar situations due to several factors. These factors can include availability of exits, anxiety of family around them, anxiety of people around them and, actual events in the environment they are in. A neural network, if trained properly, could recognize these patterns and simulate when and how much anxiety an agent would experience. Initial training sets shows that the network was able to learn the patterns and give correct responses, but still gave incorrect responses showing a need to revise the current training set to give more accurate results. This research project has application to Homeland Security evacuation, natural disaster and socio-cultural modeling scenarios.

This Project is supported by a grant from the U.S. Department of Homeland Security.