Practical prediction of concrete compressive strength using an interpretable enhanced linear regression model

Reliable estimation of concrete compressive strength is essential for quality control in civil engineering. While complex machine learning models offer high accuracy, their lack of transparency limits practical adoption. This study develops an interpretable linear regression model in which cement, slag, and fly ash are represented as total binder content and slag-to-binder ratio, raw age is replaced by its natural logarithm, and water-to-cement ratio is retained alongside superplasticizer and aggregate contents. Using a publicly available dataset of 1,030 concrete mixtures, the final model achieved a full-data R 2 of 0.821 and, under nested GroupKFold cross-validation grouped by unique mixture design, a mean outer-fold R 2 of 0.805 (SD = 0.018), RMSE of 7.30 MPa, and MAE of 5.62 MPa. Multicollinearity was controlled, with a maximum variance inflation factor of 6.35 and a condition number of 4.97. All predictors were statistically significant ( p < 0.001), and the maximum Cook’s distance was 0.0217. Benchmarking against tuned Random Forest (nested R 2 = 0.884) and XGBoost (nested R 2 = 0.882) indicated that while ensemble methods offer higher accuracy, the proposed linear model retains interpretable coefficients and provides a transparent screening tool for preliminary strength estimation within the range of the studied dataset.