Performance comparison and interpretability of machine leaning models for TBM penetration rate prediction

To achieve accurate prediction of tunnel boring machine (TBM) penetration rate (PR) under complex geological conditions, this study proposes an interpretable machine learning (ML) framework optimized using Bayesian optimization (BO). A dataset comprising 411 samples collected from two TBM projects was established, incorporating key geological parameters (uniaxial compressive strength, elastic modulus, and Poisson’s ratio) and operational parameters (thrust force and rotation speed). Seven ML models, including decision tree, Gaussian process regression, k-nearest neighbors, random forest (RF), support vector regression, extreme gradient boosting, and light gradient boosting machine, were systematically optimized using BO under five-fold cross-validation. The results demonstrate that ensemble learning models exhibit superior predictive performance. Among them, the RF model achieved the highest accuracy on the testing set with an R 2 of 0.9582, followed by the light gradient boosting machine with an R 2 of 0.9565. SHapley Additive exPlanations (SHAP) analysis indicates that uniaxial compressive strength is the dominant factor controlling PR, while thrust force act as the primary controllable parameters. The proposed framework effectively integrates high predictive accuracy with strong interpretability, providing a reliable methodological basis for intelligent TBM performance prediction and construction parameter optimization.