Interpretable predictive modeling, sustainability assessment, and cost analysis of cement-based composite containing secondary raw materials
Asif U. Memon S.A.
25 April 2025Elsevier Ltd
Construction and Building Materials
2025#473
The use of secondary cementitious raw materials (SCMs) in concrete, such as silica fume (SF) and ground granulated blast furnace slag (GGBS), can be a worthy option to reduce CO₂ emissions arising from the use of cement in concrete. Previously, several experimental and machine learning (ML)-based research studies have been conducted to promote the use of SCMs in concrete. However, prior studies have the following shortcomings: (a) a smaller database, (b) use of individual ML models, which often suffer from overfitting, poor robustness, and limited predictive accuracy, (c) non-availability of experimental validation, raising concerns about the reliability of ML-based predictions, (d) absence of sustainability and economic assessments, which are essential for the large-scale adoption of SCMs, and (e) non-existence of a comprehensive predictive model for estimating the compressive strength (CS) of binary and ternary blended concrete containing SF and GGBS. Therefore, to address these shortcomings, this study presents an innovative approach that employs a larger database, comparative analysis of individual and ensemble learning algorithms (ELAs), and experimental validation to estimate the CS of concrete while considering the sustainability and economic aspects of SCMs. A comprehensive database with eight input parameters of concrete was used to develop ML-based models such as multi-linear regression (MLR), artificial neural network (ANN), gene expression programming (GEP), and ELAs, including modified random forest (RF), bagging, and boosting techniques to anticipate the CS of concrete. The efficacy of the proposed models was validated by performing experimental investigations, statistical evaluations, K-fold tests, and external validation criteria. Shapleys Additive explanations (SHAP) analysis was also performed to explain the impact of each individual and the combined influence of input parameters on the CS of concrete. Comparative analysis shows that GEP ranks as the best predictive individual model, while MLPNN-XGBoost proves to be the most accurate among ELAs. Notably, the GEP method not only demonstrates superior results in predicted accuracy but also gives interpretable equations, enhancing the transparency and practical applicability of the model. The experimental results demonstrate strong evidence for all the developed models with lower error values and comparable R2 values with the original models. The SHAP analysis reveals that cement, followed by age and water content, are the primary parameters that significantly influence the CS of concrete containing SCMs. Likewise, sustainability and economic assessments show that the optimum mix design of SCM concrete results in a 37.8 % and 7.4 % reduction in global warming potential (GWP) and cost compared to conventional concrete. The findings of this study can contribute to sustainable concrete practices, providing useful insights into optimizing mix designs and promoting environmentally friendly construction procedures.
Compressive strength , Ensemble learning algorithms , Gene expression programming , Secondary cementitious raw materials , Shapleys Additive explanations , Sustainability assessments
Text of the article Перейти на текст статьи
Department of Civil and Environmental Engineering, School of Engineering and Digital Sciences, Nazarbayev University, Nur-Sultan, 010000, Kazakhstan
Department of Civil and Environmental Engineering, University of Maryland, College Park, 20742, MD, United States
Department of Civil and Environmental Engineering
Department of Civil and Environmental Engineering
10 лет помогаем публиковать статьи Международный издатель
Книга Публикация научной статьи Волощук 2026 Book Publication of a scientific article 2026