Predicting Compressive Strength of Ultra-High-Performance Concrete with the Influence of Silica Fume, Fly Ash, and Marble Powder Using Machine Learning Models

Ultra-high-performance concrete (UHPC) is a high-strength, durable, and sustainable material. Machine learning (ML) is utilized in this study to predict its compressive strength (CS) based on 249 experimental observations. The mixed designs included materials such as silica fume, fly ash, marble powder, cement, aggregates, water, superplasticizers, and curing age. Four models—Gradient Boosting (GB), Extreme Gradient Boosting (XGBoost), Light Gradient Boosting Machines (LightGBM), and Categorical Boosting (CatBoost)—were compared based on performance metrics like root mean square error (RMSE), mean squared error (MSE), mean absolute error (MAE), coefficient of efficiency (CE), and coefficient of determination ($\mathrm{R}^{2}$). Curing age, fine aggregates, marble powder, and silica fume were the most significant factors influencing strength using SHAP (Shapley Additive exPlanations) and PDP (Partial Dependence Plot) analyses. 10-fold cross-validation was used to achieve reliability and prevent overfitting. Model comparisons and feature analyses offer practical insights for optimizing UHPC composition and performance. The study offers valuable recommendations for UHPC mix design improvement and sustainability enhancement using industrial by-products.