Hybrid machine learning for SCC strength prediction using metaheuristic optimization

Abstract

Self-Compacting Concrete (SCC) represents a significant innovation in modern concrete technology owing to its excellent flowability, workability, and mechanical performance. Accurate prediction of SCC compressive strength is essential for optimizing mix design, minimizing experimental effort, and promoting sustainable material usage. In this study, a hybrid Machine Learning (ML) framework was developed by integrating Gradient Boosting (GB), Adaptive Boosting (ADA), and Random Forest (RF) models, optimized through metaheuristic algorithms such as Grey Wolf Optimizer (GWO), Mountain Gazelle Optimizer (MGO), Brown Bear Optimizer Algorithm (BBOA), and Fox Optimizer (FO). A dataset containing 691 SCC samples with varied mix proportions and curing conditions was utilized for model development and validation. Statistical analysis using the coefficient of determination (R²), Root Mean Square Error (RMSE), Weighted Mean Absolute Percentage Error (WMAPE), and Willmott’s Index of Agreement (WI) revealed that the BBOA-GB model exhibited the highest predictive accuracy, with R² values of 0.9955 and 0.9645 for training and testing, respectively. SHapley Additive exPlanations (SHAP) analysis indicated that slag content, cement content and curing age were the most influential factors affecting compressive strength. The proposed hybrid approach demonstrated improved performance compared to the considered baseline ensemble models. The novelty of this research lies in the comparative benchmarking of multiple metaheuristic–ensemble integrations and the development of a Graphical User Interface (GUI) for practical SCC strength prediction. The study contributes a reliable and efficient data-driven framework that enhances decision-making in SCC design and supports the advancement of sustainable concrete technology. However, the study is limited by the use of literature-based datasets and a restricted range of mix compositions. Future research will focus on incorporating larger and real-time datasets, exploring deep learning approaches, and extending the framework toward sustainability-driven mix optimization.