Hurricane risk assessment in Texas using machine learning and remote sensing data

Advancements in remote sensing and robust machine learning models enable the development of reliable hurricane risk assessments. This study demonstrates the potential of insurance data, where building replacement cost was used as the response variable. A stacked ensemble model was developed, with MAE of 6.2% and model error of 4.6% in coastline regions. The partial dependency plots provided interesting insights regarding the positive impact of building codes on damage mitigation. Furthermore, the final risk assessment was at a ZCTA (ZIP Code Tabulation Areas) level, and hence captured the spatial variation in hurricane risk at a granular level across Texas.