2026-09-21 ノースカロライナ州立大学（NC State）
＜関連情報＞
- https://news.ncsu.edu/2026/09/post-hurricane-rescue-prioritization/
- https://www.sciencedirect.com/science/article/pii/S2212420926004176
社会人口統計データと機械学習を用いてハリケーン発生時の家庭救助需要を予測する Anticipating household rescue demand in hurricanes using socio-demographic data and machine learning
Patrick Leavitt, Fred Livingston, Brandon McConnell, Benjamin Rachunok
International Journal of Disaster Risk Reduction Available online 18 September 2026
DOI:https://doi.org/10.1016/j.ijdrr.2026.106405
Highlights
- Develops a two-stage machine learning framework to predict rescue occurrence and severity.
- Combines Hurricane Harvey rescue requests, ACS indicators, and pre-Harvey NFIP claims.
Abstract
Hurricanes pose severe threats to human life, particularly for residents who shelter in place and later require emergency rescue. While significant existing work has focused on evacuation behavior and long-term recovery, less attention has been devoted to predicting rescue demand during the response phase of disasters. In this work, we develop a two-stage machine learning framework to identify where household-level rescues are most likely to be needed and, conditional on rescue demand, to estimate the likely severity of demand. We combine geocoded Hurricane Harvey rescue-request data with socio-demographic indicators from the American Community Survey and pre-Harvey National Flood Insurance Program claims data. The Stage 1 Random Forest model using compact pre-Harvey NFIP predictors achieved a ROC-AUC of 0.732 for binary rescue/no-rescue prediction. A severity-aware second stage further classified positive-demand tracts into low, moderate, and high rescue-count bins. Results indicate that historical flood-loss experience, transportation access, language, household composition, and poverty-related variables are important predictors of rescue demand. By shifting focus from hypothetical intentions to observed outcomes, this work provides actionable insights to support emergency managers in allocating scarce life-saving resources.
