Precast Ultra-High-Performance Concrete (UHPC) segmental bridges (PUSBs) are increasingly recognized as a promising class of next-generation structural systems due to their superior durability, high strength, accelerated construction, and reduced life-cycle costs. These systems form continuous bridge structures by connecting precast segments through mechanical or bonded joints, with integrated prestressing tendons ensuring structural integrity. However, the presence of discontinuities at the joints can significantly influence global performance and durability, necessitating a comprehensive understanding of the shear capacity of different joint configurations. Thus, this study develops accurate empirical models for predicting joint shear capacity using Extreme Gradient Boosting (XGB), trained on a dataset compiled from published experimental studies. The model hyperparameters were optimized using two techniques: the Tree Parzen Estimator (TPE) and Taguchi Optimization (TO). A comparative evaluation demonstrated that the TO-XGB model achieved superior predictive accuracy ( R 2 = 0.996 ) and faster convergence relative to TPE. Additionally, the TO-XGB model exhibited a lower average prediction error (16 kN) compared to the TPE-XGB model (27 kN), further validating its effectiveness. The reliability of the proposed models was verified through comparisons with existing design equations and simpler optimization strategies. To enhance model transparency, SHAP (Shapley Additive Explanations) and individual conditional expectation analyses were conducted, identifying joint area, confining stress, and number of shear keys as the most influential parameters. Furthermore, a user-friendly computational tool integrating SHAP-based visualization was developed to enable practical and transparent predictions. Overall, this study facilitates data-driven design of safe and reliable precast segmental bridge systems.
Waleed Bin Inqiad (Tue,) studied this question.