Fatigue Life Prediction of Ultrahigh Performance Concrete Using Gaussian Process Regression With SHAP Interpretation and Yeo–Johnson Transformation

Accurate prediction of the fatigue life of ultrahigh performance concrete (UHPC), due to its complex multiparameter dependent nonlinear behavior, presents a major challenge for ensuring structural safety and durability. To address this issue, this study integrates machine learning algorithms with interpretability analysis to develop a high‐precision prediction model. The Yeo–Johnson transformation was employed to correct the pronounced positive skewness of the UHPC fatigue life data. Through a systematic evaluation of seven machine learning algorithms, the hyperparameter‐optimized Gaussian Process Regression model exhibited the best overall performance ( R 2 = 0.9931, MAPE = 1.410%). Validation against the source data further confirmed the robustness of Gaussian process regression (GPR) with an R 2 of 0.939, significantly outperforming the existing model ( R 2 = 0.665). Furthermore, Shapley Additive exPlanations analysis revealed that coarse aggregate volume, cube compressive strength, and stress level are the dominant influencing factors, accounting for over 85% of the prediction variability. Consequently, this research provides a robust engineering framework for the assessment of UHPC fatigue life.