Predicting Autogenous Shrinkage of High-Performance Concrete Utilizing Advanced Machine Learning Techniques

This study utilized gene expression programming (GEP) and multi-expression programming (MEP) machine learning techniques to predict the autogenous shrinkage (AS) of concrete incorporating superabsorbent polymers (SAPs) and cementitious supplements, eliminating the need for labor-intensive and costly experimental procedures. Both the MEP and GEP models demonstrated strong performance in predicting the shrinkage and swelling of concrete containing SAPs and SCMs, with GEP emerging as the most effective. GEP achieved an impressive R2 of 0.98, underscoring its enhanced forecasting capability over MEP, which recorded an R2 of 0.94. The SHapley Additive exPlanations along with partial dependence plots underscored the notable impact of several parameters on the AS of concrete. Notably, the aggregate-to-cement ratio (Agg/C) and SAP proportion were identified as the most influential factors, demonstrating a significant decrease in shrinkage with their increase. In contrast, both the duration of the observation period and SAP size contributed to an increase in shrinkage. The partial dependence plots revealed that shrinkage decreases sharply with an increase in both the Agg/C ratio and SAP content, while it increases significantly as the SAP size grows. A graphical user interface has been developed, enabling users to input the necessary parameters and obtain predictions for AS in high-performance concrete without conducting experiments.