Hybrid machine learning for SCC strength prediction using metaheuristic optimization.

Area of Science:

• Materials Science and Engineering
• Civil Engineering
• Computational Intelligence

Background:

• Self-Compacting Concrete (SCC) offers superior workability and mechanical properties, crucial for modern construction.
• Accurate prediction of SCC compressive strength is vital for efficient mix design, cost reduction, and sustainable practices.

Purpose of the Study:

• To develop and validate a hybrid Machine Learning (ML) framework for predicting SCC compressive strength.
• To optimize ML models using metaheuristic algorithms and identify key factors influencing concrete strength.
• To provide a practical tool for SCC mix design and sustainable material development.

Main Methods:

• Integration of Gradient Boosting (GB), Adaptive Boosting (ADA), and Random Forest (RF) models.
• Optimization of ML models using metaheuristic algorithms: Grey Wolf Optimizer (GWO), Mountain Gazelle Optimizer (MGO), Brown Bear Optimizer Algorithm (BBOA), and Fox Optimizer (FO).
• Utilized a dataset of 691 SCC samples; employed statistical metrics (R², RMSE, WMAPE, WI) and SHapley Additive exPlanations (SHAP) for analysis.

Main Results:

• The BBOA-GB model demonstrated superior predictive accuracy, achieving R² values of 0.9955 (training) and 0.9645 (testing).
• SHAP analysis identified slag content, cement content, and curing age as the most significant factors influencing compressive strength.
• The hybrid framework outperformed baseline ensemble models, indicating enhanced predictive capabilities.

Conclusions:

• The developed hybrid ML framework offers a reliable and efficient data-driven approach for SCC strength prediction.
• The study highlights the potential of metaheuristic-optimized ensemble models in advancing concrete technology and sustainability.
• Limitations include reliance on literature data; future work will focus on larger datasets and deep learning approaches.