Predicting Human Mobility during Natural Disasters: A Deep Learning Approach

Abstract

There has been an increasing body of research utilizing human trajectory data to capture and analyze human movements. Previous studies demonstrated inherent transition regularities of human mobility with potential high predictability. However, we still have very limited understanding of how the predictability could change under the influence of natural disasters. In this study, we utilize mobility dataset with ample sample size and high resolution under catastrophic events to quantify their disruption on human mobility. We develop a framework adopting information theory metrics and long short-term memory (LSTM)-based predictive model to collectively quantify the perturbation effect on human mobility during Hurricane Irma. Moreover, we associate such effect with the evacuation behavior and investigate the heterogeneity of individuals. Despite observing a universally instant spatial constraint for all individuals, we found that the evacuees were able to withstand their spatial-temporal patterns with higher resilience compared to non-evacuees. We also show that the displacement distance was not significantly correlated with predictability change, hinting that additional economical and sociodemographic factors should be considered to understand the mobility recovery pattern more thoroughly.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 4 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.