Enhancing shear strength predictions of UHPC beams through hybrid machine learning approaches

Abstract Ultra-high-performance concrete (UHPC) beam shear strength prediction is a complicated process due to the involvement of numerous parameters. The accuracy needed for precise predictions is frequently lacking in current empirical equations and traditional machine learning (ML) techniques. This study proposes hybrid ML models that integrate three nature inspired metaheuristic algorithms—Giant Armadillo Optimization (GOA), Spotted Hyena Optimization (SHO) and Leopard seal optimization (LSA)- Extreme Gradient Boosting (XGB) to predict the shear strength of UHPC beams. A comprehensive dataset was created from extensive literature reviews and trained and tested on the models using multiple input parameters that affect UHPC’s shear capacity. For model assessment, performance metrics, such as coefficient of determination (R2), root mean square error (RMSE), mean absolute error (MAE), and variance accounted for (VAF), were utilized. Results showcased high accuracy, with R2 values approaching 0.9912 in training and 0.9802 in testing phases using the LSA-XGB algorithm, indicating excellent model fit and predictive reliability. To improve the model’s transparency and interpretability, the study also incorporates shapely additive explanations (SHAP), which reveal how each dataset attribute affects the predictive results. The LSA-XGB algorithm performed better than prior studies and empirical equations in predicting the shear strength of UHPC beams. More sophisticated machine learning techniques that improve the precision of predicting the shear capacity of UHPC beams are demonstrated in the study. Further, the use of a graphical user interface (GUI) helps researchers and engineers to make quick, well-informed decisions in real-time. These findings offer a reliable, interpretable, and accessible approach to predicting shear strength in UHPC beams, contributing to safer structural engineering practices.