Impact Analysis of Contributing Factors of NYC Flash Floods with Interpre table Machine Learning

In this paper, we analyze multi-modal data including environmental, geographic, infrastructure, demographic, and socioeconomic data from New York City (NYC), to examine their impact on flood risk. We applied XGBoost machine learning model combined with SHapley Additive exPlanations (SHAP) function to analyze and interpret the contributions of these factors to flood risks. We tested the model with a window of 1 to 7 days to determine optimum time scales to capture the essential information in the time series and factors' dependencies over various time periods. Our findings reveal that although environmental and geographic factors have a strong correlation with flood risk, socioeconomic factors, such as median age, median house value, population density, and race also play an important role in such risks. Low-income household areas have higher flood risks, which happen to also be minority communities, making them more vulnerable to flooding than higher-income communities. Furthermore, population density and level of education may also have impact on flood risks. These insights can be used to facilitate planning and preparing future flood prevention and mitigation strategies.