A Gradient Boosting Approach to Identify Behavioral and Policy Determinants of Flood Resilience in the Continental US

Abstract

Annual flood losses in the US currently amount to more than 30 billion US$. A common way for individuals to strengthen household flood resilience is purchasing a flood insurance. However, only less than 5% of all US households currently have a flood insurance. Understanding the determinants of flood insurance purchase is key to foster preparedness and community resilience. Here, we formulate the problem of identifying the determinants of flood insurance purchase as a feature selection problem. We develop a gradient boosting modelling approach to select the most relevant predictors of flood insurance for the continental US. First, we train the gradient boosting model on a large open dataset containing publicly available records on more than 50 million flood insurances purchased between 2009 and 2020, socio-demographic data extracted from more than 25000 census variables, and data on the participation in the Community Rating System, a voluntary incentive program of the US National Flood Insurance Program. Then, we analyze feature importance metrics to identify the most relevant determinants. Numerical results indicate that our model successfully approximates the functional relation between some of the candidate predictors and flood insurance purchase rates in the continental US. Householders purchase flood insurances with a reactive attitude: flood insurances are purchased particularly in areas where householders have already experienced the impacts of flooding. More proactive behaviors can be effectively encouraged by community policies and public incentive schemes.