The joint roughness coefficient (JRC) of rock significantly influences the mechanical response and stability of rock joints, and it can be predicted based on the joint profile line geometry. This study developed a model, XGBoost-BO, which integrates eXtreme Gradient Boosting (XGBoost) and Bayesian Optimization (BO) to predict the JRC of rock joint profiles and investigated the impact of sample size and feature indicators on prediction accuracy and efficiency. Specifically, it includes: (1) Publicly available data from 112 rock structural surface profiles were used to expand the sample dataset using various interpolation algorithms, and the effects of expanding the dataset based on different interpolation algorithms were compared for prediction accuracy. (2) Analytical studies were conducted using the expanded sample dataset (448 samples in total, with 336 training samples and 112 test samples), and the model was analyzed using Shapley Additive Explanations (SHAP) for machine learning. R², Mean Absolute Percentage Error (E_MAPE), Mean Absolute Error (E_MAE), and Root Mean Square Error (E_RMSE) were used as performance metrics to evaluate the model's prediction accuracy. The results indicate that: The XGBoost-BO model performs well in predicting the JRC coefficients of rock joints. The prediction accuracy of the XGBoost-BO model (R² = 0.9121, E_MAPE = 0.1023, E_MAE = 0.7557, E_RMSE = 1.2653), expanded by interpolation algorithm from the original sample data, surpasses that of the original sample data (R² = 0.8627, E_MAPE = 0.1895, E_MAE = 1.3418, E_RMSE = 1.7513).