Crashworthiness of automotive materials is critical, However, obtaining mechanical properties data at high strain rates requires a high cost and cycle time. In this paper, an ensemble neural network model is proposed to establish the mapping relationship between material basic information and dynamic ultimate tensile strength of automotive steels. The results show that the model has a good predictive performance, and its predicted ultimate tensile strength of automotive steels at 100/s has a fivefold cross-validated average R2 and MAE of 0.9 and 51 MPa, respectively. In addition, mechanical properties at quasi-static conditions, Si, and the width of the material are the main factors affecting the crash strength of the material as analyzed by SHAP. This study contributes to the reduction of tensile experiments to save cost and cycle time on one hand, and to the selection and design of materials to improve the structural strength and safety of automobiles on the other hand.