Machine learning prediction of climate-induced disaster property damages considering hazard- and community-related attributes

The rapid increase in the earth’s average temperature has led to an unpreceded surge in the frequency and impacts of Climate-Induced Disaster (CID) across the globe. Subsequently, the costs of CID damages have been growing, and climate action failure and extreme weather events were identified among the most severe global risks over the next decade. Within this context, machine learning-based models are developed to predict CID property damages. The models integrate both community- and hazard-related characteristics as inputs to predict CID property damages. The models are trained and tested using wind-related property damage data in New York State through integrating the Federal Emergency Management Agency’s community data and the National Atmospheric and Oceanic Administration’s hazard data. The current study utilizes different supervised machine learning techniques to develop several CID property damage prediction models. The developed models yielded a coefficient of determination of 0.66, 0.81, 0.72, 0.77, and 0.79 for the regression trees, random forest, bagging, gradient boosting, and extreme gradient boosting respectively. The developed models are expected to aid community stakeholders in developing urban center preparedness plans under CID, which can facilitate strategic urban resilience planning under different climate-induced hazards.