This paper proposes a comprehensive analysis framework, combining three-dimensional (3D) numerical model and machine learning, to investigate probabilistic performance of retrofit actions on coastal bridges subjected to extreme wave forces. Specifically, a 3D Computational Fluid Dynamics (CFD) model is developed to calculate extreme wave load on the bridge superstructure, which could provide more accurate results as compared with traditional twodimensional (2D) model. The established 3D model is validated by laboratory experiments. The characteristics of wave forces are parametrically investigated, and an Artificial Neural Network (ANN) model is utilized to quantify the loading effects with multiple surge and wave parameters. Such numerical-based ANN model could predict wave forces under variable scenarios accurately, and significantly reduce the high computational cost of the 3D numerical model. Based on the numerical and machine learning results, the bridge fragility curve is derived by considering uncertainties associated with structural demand, capacity, and hurricane hazard. Long-term failure risk is assessed under different climate change scenarios. Furthermore, different retrofit methods to improve structural performance and reduce failure risk are examined according to the proposed framework, including inserting air venting hole, enhancing connection strength, and elevating bridge structure. The proposed framework could facilitate the optimal and robust design and maintenance of coastal infrastructures under hurricane effects in a long-term time interval.