Predicting Mechanical Strength of Alkali-Activated High-Performance Concrete Using Machine-Learning Methods

Area of Science:

• Materials Science
• Sustainable Construction
• Artificial Intelligence

Background:

• Growing concrete demand presents environmental challenges.
• Alkali-activated high-performance concrete (AA-HPC) offers a sustainable solution.
• Accurate prediction of AA-HPC compressive strength is crucial for efficient use.

Purpose of the Study:

• To explore machine learning (ML) applications for predicting AA-HPC compressive strength.
• To identify the most effective ML model for optimizing experimental costs, construction time, and environmental impact.
• To develop a user-friendly tool for practical application of the optimized ML model.

Main Methods:

• Evaluated nine different machine learning models.
• Utilized the African vultures optimization algorithm (AVOA) for hyperparameter tuning of extreme gradient boosting (XGBoost).
• Developed a graphical user interface (GUI) for accessible model implementation.

Main Results:

• The XGBoost model optimized with AVOA (XGB-AVOA) demonstrated superior performance.
• XGB-AVOA achieved R² of 0.994 and RMSE of 2.368 on the training set, and R² of 0.975 and RMSE of 5.664 on the testing set.
• AVOA showed higher efficiency in parameter optimization compared to other tested algorithms (GWO, WOA, SSO, GTO).

Conclusions:

• The XGB-AVOA model provides accurate, cost-effective, and time-saving predictions for AA-HPC compressive strength.
• AVOA is a highly effective optimizer for XGBoost in this context.
• The developed GUI facilitates practical application of advanced ML for sustainable construction materials.