In recent years, machine learning methods have been widely applied to predict stratum displacement induced by tunnel excavation. However, machine learning methods suffer from deficiencies such as poor model generalization ability resulting from insufficient learning capabilities for small-sample events and the neglect of geotechnical mechanical mechanisms. In this work, SVR is chosen as the data-driven ML model. The hyperparameters of the SVR model are determined via the PSO algorithm. When the non-uniform deformation pattern is employed as the displacement boundary condition for the tunnel cross-section, the analytical solution of stratum displacement is derived by means of complex variable method. The derived analytical solution is utilized as the physics-driven model. Integrating the field-monitored stratum displacement data, the data-driven prediction results, and the analytical calculation results, a dual-loss function is formulated. Subsequently, a dual-driven prediction model for stratum displacement induced by tunnel excavation is established. The impact of the number of samples on the dual-driven prediction model is analyzed. Moreover, a comparison is made between the prediction results of the dual-driven model and the data-driven model for the same data samples. The results show that for small sample datasets, the prediction accuracy of the dual-driven model is higher. As the number of data samples increases, the prediction accuracy of the pure data-driven model is higher than that of the dual-driven prediction model.