Machine learning algorithm for predicting the compressive strength of sustainable ultra-high-performance concrete (UHPC) containing supplementary cementitious materials (SCMs)

The reliable prediction of concrete compressive strength is essential for advancing sustainable construction practices, particularly when incorporating supplementary cementitious materials (SCMs) such as fly ash, silica fume, ground granulated blast furnace slag, limestone powder, and recycled concrete powder. These materials enhance sustainability but introduce significant variability and nonlinear interactions that complicate traditional mix design approaches, which typically depend on extensive laboratory testing. To address this challenge, this study employs machine learning (ML) techniques, viz, Support Vector Machines (SVM), Gaussian Process Regression (GPR), and Boosted Trees (BT), to develop predictive models based on a dataset of 101 experimental concrete mixtures. Bayesian optimization was applied to refine model hyperparameters, aiming to improve predictive accuracy and generalization. Model performance was assessed not only through statistical indices but also by scatter plots, residual error analysis, and sensitivity assessment to identify influential mix parameters. Results show that the optimized boosted tree model (OBT-M3) achieved the highest predictive accuracy (R2 = 0.92, RMSE = 0.0579, MAE = 0.0334), outperforming both baseline and alternative ML approaches. The findings provide practitioners with a data-driven tool to support mix design decisions, reduce reliance on costly trial-and-error experiments, and promote the use of SCMs in concrete production for sustainable infrastructure.