Integrating explainable hybrid machine learning and evolutionary optimization for high-fidelity prediction of ultra-high-performance concrete compressive strength

Accurate prediction of compressive strength (CS) in ultra-high-performance concrete (UHPC) remains challenging due to complex nonlinear interactions among mixture components and curing conditions, limiting the effectiveness of conventional predictive models. From an applied perspective, reliable strength prediction is essential for optimizing UHPC mix design, enhancing structural performance, and reducing material consumption and environmental impact. To address these challenges, this study proposes a hybrid machine learning (ML) framework for predicting CS of UHPC using ensemble and tree-based models (extra trees, XGBoost, LightGBM, decision tree, bagging, histogram gradient boosting, and random forest) optimized through grid search, random search, bayesian optimization, differential evolution, genetic algorithms, and particle swarm optimization. The models, developed from 810 literature-derived datasets, were rigorously evaluated using an 80/20 train–test split, 10-fold cross-validation, and a 60/20/20 training–testing–validation scheme, with predictive performance quantified by R 2 , RMSE, and MAE, and model interpretability examined through SHAP, ALE, and ICE analyses. Evolutionary hyperparameter optimization markedly improves predictive accuracy and generalization, with XGBoost (grid search and bayesian optimization) achieving the highest performance (testing R 2 = 0.960, RMSE = 7.687 MPa, MAE = 5.492 MPa; cross-validation R 2 = 0.970, RMSE = 6.913 MPa, MAE = 4.903 MPa), while extra trees and XGBoost with grid and random search consistently exhibit robust prediction across all phases. Explainability analyses identified curing age as the primary factor influencing CS. From a practical standpoint, the proposed framework constitutes a robust and interpretable predictive tool that enables engineers and researchers to design and optimize sustainable UHPC mixtures, facilitating informed decision-making, improving material efficiency, and reducing environmental impact in advanced concrete technologies. This study thus advances scientific understanding while providing practical utility, bridging advanced predictive modeling and real-world engineering applications.