Data‐Driven and SHAP‐Guided Framework for Seismic‐Resilient Concrete Proportioning and Compressive Strength Prediction

ABSTRACT High‐performance concrete (HPC) plays a critical role in hydraulic and seismic‐prone infrastructure, where both mechanical strength and resilience under earthquake loading are essential. Accurate estimation of uniaxial compressive strength (UCS) is crucial for ensuring structural safety and durability, yet direct testing is often time‐consuming, costly, and difficult to integrate into construction workflows. In this study, we propose a novel data‐driven framework that combines a Dung Beetle Optimization (DBO)‐enhanced LightGBM model with SHapley Additive exPlanations (SHAP) to predict the compressive strength of HPC and elucidate the influence of mix design parameters on structural performance under seismic loads. The proposed framework is validated using real‐world datasets from large‐scale hydropower projects. Results demonstrate a high predictive accuracy, and SHAP analysis identifies key parameters–cement content, curing age, water‐to‐cement ratio, and supplementary materials such as blast furnace slag–that significantly influence both strength development and earthquake resilience. The framework provides actionable insights for designing seismic‐resilient HPC, optimizing material proportions, and supporting construction decisions under complex environmental onditions.