Prediction of shear strength of studs embedded in UHPC based on an interpretable machine learning method

The headed stud is a critical component in steel-UHPC composite structures, with its shear strength significantly affecting overall structural performance. Therefore, accurately predicting the shear capacity of headed stud is of paramount importance. This study examines the shear strength of headed studs embedded in ultra-high-performance concrete (UHPC) and develops a predictive model using interpretable machine learning techniques. A dataset of 577 push-out tests was established and employed the advanced unsupervised Isolation Forest method to eliminate outliers. Then, five machine learning models including the Linear Regression (LR), Decision Tree (DT), Random Forest (RF), Gradient Boosting Decision Tree (GBDT), and Extreme Gradient Boosting (XGBOOST) were trained to predict the shear strength of headed studs embedded in UHPC. The XGBoost model achieved the highest accuracy, with R2 value of 0.97. It outperformed the other models, thereby ensuring the reliability of shear strength predictions for headed studs. To address the interpretability challenges associated with machine learning models, feature importance was analyzed using Partial Dependence Plots (PDP), Individual Conditional Expectation (ICE), and Shapley Additive explanations (SHAP). The results indicate that the cross-sectional area of headed stud has the greatest influence on the shear performance, followed by the characteristics of UHPC and the tensile strength of headed studs. Finally, utilizing the XGBOOST model for parameterized study of input features, a new equation for predicting the shear strength of headed studs in steel-UHPC composite structures was established, combined with curve fitting methods. This equation not only enhances the accurate prediction of shear performance but also provides insights into the machine learning models.