Slump retention as a predictor of ready-mixed concrete strength: Evidence from industrial production data

Reliable compressive-strength prediction is important for quality control in ready-mixed concrete production, particularly when workability evolves during transportation and waiting. This study evaluates whether time-dependent workability descriptors, including slump measured at 0, 60, and 120 min and slump retention, improve prediction of 7-day and 28-day compressive strength from an anonymized industrial dataset. Linear Regression, Random Forest, and Gradient Boosting regression models were evaluated using an auditable retained dataset of 893 records with a 714/179 train/test source-order split. The retained dataset represents medium-strength ready-mixed concretes within approximately the C35–C50 range; therefore, the findings should not be generalized to low-strength, high-strength, self-compacting, lightweight, or specialty concretes without additional validation. Because the archived retained workbook did not preserve calendar dates, validation was reported as an internal source-order split rather than external plant-level validation. The 28-day Gradient Boosting model achieved an RMSE of 1.53 MPa and an MAE of 1.21 MPa on the 179-record test subset; however, R 2 was negative because the source-order test subset had a narrow and shifted strength range. Ablation analysis showed that adding the full time-dependent workability descriptor set reduced MAE by 0.18 MPa relative to the initial-slump feature set in a paired bootstrap test ( p = 0.0004 ). Correlation, VIF, and permutation-importance analyses further showed that slump retention is highly collinear with 120-min slump and should be interpreted as a predictive descriptor rather than an independent causal variable. The revised results therefore support the practical value of monitoring workability evolution for pre-placement or post-transport strength-risk assessment while emphasizing the need for external validation with date-preserved and plant-diverse records.