Hybrid-optimized ensemble learning models and GUI for predicting compressive strength of fly ash and slag concrete

This study investigates the application of ensemble machine learning models Gradient Boosting (GB), Extreme Gradient Boosting (XGB), Adaptive Boosting (ADA), and Categorical Boosting (CAT) optimized using Bayesian Optimization (BO), Water Cycle Algorithm (WCA), and Social Emotional Optimization Algorithm (SEOA) to predict the compressive strength of fly ash and slag concrete. Utilizing a dataset of 1030 concrete samples with eight input parameters (cement, blast furnace slag, fly ash, water, superplasticizer, coarse aggregate, fine aggregate, and age) and compressive strength as the output, the models were evaluated using metrics such as R², RMSE, and A20-index. The CAT model optimized with BO achieved the highest performance (training R² = 0.993, testing R² = 0.939, A20-index = 99.03 % train, 86.89 % test), outperforming traditional empirical models (e.g., Abrams’ law, R² ≈ 0.60–0.70; Feret's equation, R² ≈ 0.65–0.75) by up to 50 % in accuracy, as well as other ensemble methods. Sensitivity analysis via the Cosine Amplitude Method highlighted cement, age, and superplasticizer as key predictors. SHAP analysis further revealed nonlinear feature interactions. Statistical tests and Monte Carlo simulations confirmed CAT-BO's precision, with narrow 95 % confidence intervals across the full dataset, while a user-friendly graphical user interface (GUI) was developed to facilitate practical application in civil engineering. The GUI, tested by engineers, supports mix optimization, reducing cement usage by 10–15 %. These findings enhance the accuracy of compressive strength prediction, supporting sustainable concrete design by optimizing mix proportions and promoting the use of industrial by-products.