A reliable approach for classifying metakaolin content of high-Performance concrete via machine learning algorithm

Cement production is energy-intensive and emits significant CO2, making sustainable alternatives essential. Pozzolanic materials, such as fly ash, silica fume, and metakaolin, can partially replace cement, enhancing strength and durability while reducing costs and environmental impact. Metakaolin, derived from calcined kaolin clay, has gained interest in concrete applications. This study introduces a classification method for High-Performance Concrete based on metakaolin-to-cement ratios: Free (0%), Low (0%-20%), and High (20%-50%). Three tree-based classifiers (Decision Tree, Random Forest, and Stochastic Forest) were trained on 80% of 241 samples. Additionally, hybrid models were developed by fine-tuning hyperparameters using Swarm Magnetic, Sea-horse, and Exponential distribution optimizers. Classification performance was evaluated using various metrics, comparative visual plots, and Fourier Amplitude Sensitivity Tests. The Random Forest model optimized with the Swarm Magnetic Optimizer achieved the highest accuracy (95.8%), precision (97.9%), and F1-score (96.5%) across training, validation, and testing phases. These findings highlight the effectiveness of machine learning in optimizing concrete mixture designs and improving classification accuracy.