Predicting Evacuation Destinations due to a Natural Hazard using Mobile Network Data

Abstract

Exposed to a variety of natural hazards, Vanuatu is one of the most disaster-prone countries in the South Pacific. The Government plays a central role in disaster response and has articulated a need for information on disaster-induced displacement in order to target resources. This paper aims to inform preparation and planning by developing a method to predict evacuation destinations before a disaster happens by applying machine learning approaches to mobile network data. In this study, the eruption of Mount Monaro in 2017 is chosen to test the prediction performance of the model in a real disaster scenario. We explored 273 features, extracted from over one-hundred-million anonymized mobile network records, to describe (a) basic phone usage, (b) active user behavior, (c) spatial behavior, (d) regularity, and (e) diversity. Our results show that supervised machine learning methods produce promising results in predicting evacuation destinations.