Machine Learning for Earthquake Emergency Evacuation: Site Selection and Neighborhood Navigation

This research is first to introduce a machine learning-based method to enhance the quality and speed of selecting emergency evacuation centers in Tehran, optimizing the use of the city's current capacities. Tehran, with a population of 8.7 million, is located on multiple active faults, where 258 out of its 373 neighborhoods lack sufficient emergency evacuation centers. This inadequacy poses a significant risk during emergencies, potentially leading to failures in the evacuation process. By employing an artificial neural network model informed by data from San Francisco—a city with similar seismic risks, urban structure, building types, and population density—this study achieved a 23% reduction in the average minimum distance from each parcel to the nearest emergency evacuation center compared to the current conditions. Additionally, per capita adequacy at the neighborhood level increased from 27% to 59%. The second goal of the research is to streamline the emergency evacuation process through the development of an application that dynamically navigates users to the nearest evacuation center identified by the model. This application is built using OpenStreetMap and the OpenRouteService routing algorithm. With the successful achievement of both goals, the emergency evacuation process in Tehran is now smarter and more comprehensive. The model here can also be applied to other cities that are similar to the cities studied in terms of site-selecting criteria. For cities with less similarity, the model needs to be slightly trained with the data of the target city to achieve desirable results.