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

Abstract

This work proposes a machine-learning-based approach to enhance the selection and accessibility of emergency evacuation centres in urban environments. Urban areas prone to natural disasters often lack adequately distributed evacuation facilities; in cities highly vulnerable to seismic activity, many neighbourhoods do not have sufficient access to safe evacuation locations, and existing planning methods are static, inefficient, and unable to adapt to real-time conditions, leading to increased risk, congestion, and potential failure during emergencies. An Artificial Neural Network (ANN) model is developed to identify optimal locations for evacuation centres, reducing the average distance between residents and centres while improving per-capita accessibility. In addition, a dynamic navigation system is implemented using geospatial data and routing algorithms to guide users to the nearest safe location in real time, considering distance and accessibility. By integrating machine learning with Geographic Information Systems (GIS), the system supports proactive disaster management—improving preparedness, resource allocation, and emergency-response planning—and provides a map-based interface that displays the user’s location, evacuation centres, earthquake events, and the recommended route. Fifteen test cases covering data validation, preprocessing, site prediction, route optimisation, real-time updates, visualisation, and security were executed and all passed. The methodology is scalable and adaptable to other earthquake-prone cities with similar urban challenges.