<p>Seismic fault rupture can extend to the surface, and the resulting surface deformation can cause severe damage to civil engineering structures crossing the fault zones. Coseismic Surface Rupture Prediction Models (CSRPMs) play a crucial role in the structural design of fault-crossing engineering and in the hazard analysis of fault-intensive areas. In this study, a new global coseismic surface rupture database was constructed by compiling 171 earthquake events (<i>M</i>_w: 5.5–7.9) that caused surface rupture. In contrast to the fault classification in traditional empirical relationships, this study categorizes earthquake events as strike-slip, dip-slip, and oblique-slip. CSRPMs utilizing Bayesian ridge regression (BRR) were developed to estimate parameters such as surface rupture length, average displacement, and maximum displacement. Based on Bayesian theory, BRR combines the benefits of both ridge regression and Bayesian linear regression. This approach effectively addresses the issue of overfitting while ensuring the strong model robustness. The reliability of the CSRPMs was validated by residual analysis and comparison with post-earthquake observations from the 2023 Türkiye earthquake doublet. The BRR-CSRPMs with new fault classification criteria are more suitable for the probabilistic hazard analysis of complex fault systems and dislocation design of fault-crossing engineering.</p>