Power system resilience plays a crucial role in their safe and reliable operation during extreme weather events. However, the unpredictability of typhoons and variability of wind turbines bring high computational challenge for the traditional model-based resilience assessment approaches. To address this issue, this paper proposes a hybrid regression-classification learning approach to improve the efficiency of resilience assessment. The Deep Natural Network (DNN) model is utilized to estimate the optimal amount of load shedding in each scenario, while the Random Forest (RF) model is employed to detect the occurrence of load shedding events. By leveraging the mutually corrective mechanism between the two models, the efficiency of resilience assessment is further enhanced. Finally, the accuracy and rapidity of the proposed method is validated through a comprehensive analysis conducted on the adapted IEEE RTS-79 system.