Tropical cyclones are one of the most destructive natural disasters, frequently occurring in the southeastern coastal areas of China, causing huge losses. Accurate characterization and prediction of negative impacts are necessary to minimize impact on coastal cities. However, a complex interaction between hazard intensity and other drivers is involved in the impact of some extreme tropical cyclones, which are often worthy of attention. In this study, a quantitative vulnerability assessment model of tropical cyclones considering the mutual relation between hazard intensity and environmental drivers was developed using the generalized additive model (GAM), random forest (RF), and extreme gradient boosting (XGBoost) for the case of Hainan Province, China. In the testing dataset, the root mean squared error of the optimal model was 0.1393, and the mean absolute percentage error of the optimal model was 35.2 % for spikes in the affected population rate. The spikes in the affected population rate were predicted well by the hybrid model jointly considering nonlinear multivariable regression models and undersampling machine learning methods.