Machine Learning-Based Prediction and Feature Interpretability Analysis of Ultra-High-Performance Concrete Compressive Strength

The compressive strength of ultra-high-performance concrete (UHPC) exhibits a highly nonlinear and complex relationship with its mix design components, making it difficult for traditional modeling methods to effectively capture its performance evolution and the nonlinear coupling effects among influencing factors. To address this challenge, this study develops a predictive model that integrates a genetic algorithm (GA) with an artificial neural network (ANN), based on collected data of UHPC mix proportions and compressive strength. The GA is employed to optimize the hyperparameters of the neural network, thereby enhancing the prediction accuracy of the model. The results show that the improved model achieves a coefficient of determination (R²) of up to 95.6% on the training set, indicating strong predictive performance. To tackle the issue of poor interpretability in deep learning models, the SHapley Additive exPlanations (SHAP) method is further introduced to perform interpretability analysis. This approach systematically reveals the influence of individual material components on UHPC compressive strength and provides theoretical insights into the underlying mechanisms. The findings of this study offer theoretical guidance and data support for the optimization of UHPC mix designs and their practical engineering applications.