Wind-driven rain (WDR) significantly affects the thermal and moisture performance of building walls, poses challenges due to the dynamic interplay of wind speed, wind direction, and rainfall intensity. This paper introduces a machine learning (ML)-based approach for predicting WDR on building facades, aiming to overcome the limitations of traditional experimental methods, semi-empirical models, and CFD simulations. The study presents two distinct ML models tailored to different weather scenarios. For the normal weather scenario, CFD simulations on standardized residential building models generate a comprehensive dataset incorporating wind parameters, raindrop sizes, and building geometries. Eight ML algorithms—including ANN, LightGBM, and XGBoost—are evaluated. Among these, the ANN model demonstrates superior performance with an RMSE of 0.009 and achieves predictions in just 7 s, making it over 300 times faster than traditional CFD simulations. For the extreme weather scenario, which focuses on high-risk areas during typhoon conditions, the study employs symbolic regression to develop a formula-based model using a limited dataset. This model, which considers only building height, wind speed, and rainfall intensity, attains an RMSE of 0.095 on the validation dataset. The resulting formula not only provides high predictive accuracy but also encapsulates the physical relationships between the key variables, highlighting the dominant effect of wind speed on WDR and the coupling effects with building height and rainfall intensity. Overall, the proposed ML models leverage the high accuracy of CFD simulations and the computational efficiency of semi-empirical models. They offer a robust alternative for predicting WDR, enhancing both the speed and precision of building envelope assessments under a range of weather conditions. This work opens new avenues for integrating ML techniques into urban environmental studies and building safety evaluations, especially in the context of increasingly frequent extreme weather events.