AbstractHigh-speed trains running over sea-crossing railway bridges can be subjected to earthquake action in deepwater. In analyzing hydrodynamic-induced response on train–bridge interactions, the effect of earthquake-induced hydrodynamic pressure is a critical issue that still needs to be correctly modeled and understood. This study adopts a machine learning-based method for evaluating the earthquake-induced response on train–bridge interactions. A back-propagation neural network (BPNN) surrogate model is established by correlating the environmental parameters with the stochastic responses of train–bridge interactions to improve computational efficiency. Pintang’s bridge, located in China, is selected as a case study. The results show that the proposed method is robust and accurate. The error is less than 1% when compared with the Monte Carlo method (MCM). Moreover, the consideration of earthquake increases the dynamic indices of the bridge girder to about 18%, compared to the case applying only hydrodynamic pressure. Furthermore, this study performed parametric analysis and found that the hydrodynamic pressure reached the maximum value under the action of –90° of incident angle. These results will be helpful for the design of railway bridges in the coastal area.