Natural Hazards Research Summit 2024: Hurricane Wind Loss Modeling Using Insurance Claims Data

This study aims to develop a predictive model for estimating hurricane-induced property loss using variables such as wind speed, building attributes, land cover, and rainfall from insurance claims data. The dataset includes insurance claims data from four major hurricanes (Isaias, Florence, Dorian, and Matthew) that affected Eastern North Carolina. Exploratory data analysis is employed to gain insights into the relationships between the variables and the observed losses. Several statistical and machine learning models are developed to predict hurricane-induced losses. Two response variables, namely Dollar loss and Building loss ratio, are considered during model development. The appropriate model is selected and used for further analysis. Ongoing work focuses on analyzing the model prediction errors on aggregated data (e.g. hurricane-level, county-level) and refining the modeling process.