Accurate and reliable tunnel boring machine (TBM) penetration rate prediction has significant engineering value for improving construction efficiency and ensuring construction safety. Aiming at the limitations of the existing TBM penetration rate prediction model with poor accuracy and insufficient consideration of uncertainty during construction, an interpretable TBM penetration rate interval prediction method based on machine learning was proposed. Firstly, several sets of domestic TBM tunnel engineering data were collected, with rock uniaxial compressive strength (UCS), rock integrity coefficient (Kv), thrust force (TF) and cutter speed (RPM) were selected as the input features. An extreme gradient boosting (XGBoost) point prediction model was developed through the Newton-Raphson optimization (NRBO) algorithm and cross-validation strategy. The Shapley additive explanation (SHAP) framework was introduced to analyze the contribution of the feature parameters to the prediction results. Then, the uncertainty of the point prediction results was quantified based on the adaptive bandwidth kernel density estimation (ABKDE) method, and achieve the interval prediction of penetration rate. Finally, the validity of the model was verified by a case study of the Kerman water transfer tunnel project in Iran. The results of the study show that compared with the XGBoost model without NRBO algorithm, the prediction error mean square error (MSE), mean absolute error (MAE), and mean absolute percentage error (MAPE) of the NRBO-XGBoost model have been reduced by 13.9%, 19.1%, and 0.7%, respectively, and the coefficient of determination, R², has been improved by 0.015 1; the feature importance rankings are UCS (0.415 6)>TF (0.155 4)>RPM (0.104 5)>Kv (0.004 7), revealing that rock strength is the dominant influencing factor of penetration rate; the proposed model outperforms the adaptive boosting (AdaBoost) and the random forest (RF) model in the interval prediction performance, the predicted interval coverage probabilities (PICP) of NRBO-XGBoost, AdaBoost and RF models, reach 92.1%, 88.4% and 90.2%, respectively, with better uncertainty quantification ability; in the engineering example validation, the predicted R² reaches 0.967 6 and the predicted intervals cover the measured values, which confirms that the model has a good applicability to engineering.