Development and field based validation of a predictive model for concrete compressive strength using fresh state properties in a large scale construction project

Estimation of a compressive strength of concrete earlier than for a standard 28 day test remains a highly challenging issue in construction industry on a large scale, where the quality control is merged with efficient scheduling and sustainable construction. This work presents a predictive model validated under real field conditions, through the processing of data collected in 4872 concrete cylinders produced in an infrastructure work with a total cast volume of 25,000 m³. The models are based on fresh state parameters temperature, slump, air content, and density that are commonly tested in the field, and are used to predict 7 and 28 days compressive strength for 35 and 40 MPa designs. Selected were three statistical methods including multiple linear regression as the base method, principal component analysis for dimensionality reduction, and LASSO regularization with interaction terms. Cross validation results showed the robust performance and the adjusted R² value was as high as 0.71 and RMSE was less than 3.4 MPa. While LASSO tightened model parsimony and did not reduce prediction efficiency, PCA increased precision in the context of multicollinearity and only slightly at the cost of precision. A strong correlation between early and late strength (0.84 and 0.74 for 35 MPa and 40 MPa, respectively) supports the use of early age results as practical predictors. The compressive strength prediction module was validated with real project data, providing ±RMSE confidence bands for operational use. These findings demonstrate that statistical modeling can be integrated into quality control workflows, enabling data driven decisions in concrete production and placement.