To accurately predict the peak dilation angle of rock discontinuities, a hybrid data-driven model based on the improved sparrow search algorithm is proposed. The main influencing features such as normal stress, basic friction angle, uniaxial compressive strength and three-dimensional roughness parameters are used as the inputs of the model, and the peak dilation angle is used as the output of the model. The coefficient of determination (R2), mean absolute error (MAE), root mean square error (RMSE), and mean absolute percentage error (MAPE) are introduced to evaluate the prediction performance of the proposed model and compare its performance with existing classical models. Then, the effect of different normalization methods on data-driven model is discussed. Finally, the shapley additive explanations (SHAP) method is adopted to explain the contribution and relative importance of each input feature. The results indicate that the proposed model can well describe the complex nonlinear relationship between influencing features and peak dilatation angle. Compared with classical models, the proposed model has the highest accuracy and the smallest error. R2, MAE, RMSE and MAPE are 0.966, 1.029, 1.184 and 3.99 % respectively. SHAP results show that normal stress and uniaxial compressive strength are the most important input features for predicting peak dilation angle. Compared with mean normalization method and arc-tangent normalization method, the min-max normalization method is more suitable as a pre-processing method for the proposed model.