Abstract
Accurate prediction of concrete compressive strength remains a critical challenge in structural engineering due to the complex, non-linear relationships between mixture components and mechanical properties. This study presents a comprehensive machine learning framework for predicting concrete compressive strength, incorporating both deterministic and probabilistic approaches. We analyzed 1030 laboratory-tested concrete specimens with eight input variables to evaluate four distinct algorithms: linear regression, support vector regression, random forest, and artificial neural networks. Additionally, we developed a Monte Carlo dropout neural network to quantify prediction uncertainty through Bayesian approximation. Results demonstrate that ensemble tree-based methods achieved superior predictive performance with a root mean square error of 5.47 MPa and coefficient of determination of 0.884, significantly outperforming conventional approaches at the 0.01 significance level. Feature importance analysis revealed that curing age and cement content collectively account for 66% of the predictive variance, while water content exhibits the expected negative correlation with strength development. However, uncertainty quantification through Monte Carlo dropout revealed systematic undercalibration, achieving only 38.4% coverage of the nominal 95% prediction interval. These findings provide engineers with quantitative insights for mixture design optimization while emphasizing the importance of uncertainty awareness in computational predictions for structural applications.