Using Stacking Machine Learning Models to Predict High-Performance Concrete Compressive Strength

One of the most important mechanical performance indicators for concrete is its compressive strength, which differs depending on the mix proportions used to make the concrete. Traditional methods mainly rely on compression tests to obtain compressive strength and combine it with practical applications for mix design and optimization. In this study, machine learning models for predicting compressive strength were developed using seven algorithms, including LightGBM (LGBM), HistGradientBoosting (HGB), XGBoost (XGB), Random Forest (RF), K-Nearest Neighbors (KNN), Support Vector Machine (SVM), and Linear Regression (LR). Grid search and cross-validation were used to choose the most effective hyperparameter combinations. In order to forecast the outcomes, the stacking approach was then used with HGB, LGBM as the base learners and LR as the meta-learner. The models were compared and selected based on evaluation metrics, and the best stacking model (HGB+LGBM+LR) was obtained. Finally, SHAP analysis was used to explain the variables affecting concrete's compressive strength. The study's findings show that the stacking model, with its MAE score of 2.43, MSE score of 11.296, RMSE score of 3.361, and R2 value of 0.96, outperforms all others on the test set. The three main variables that have the biggest effects on high-performance concrete's (HPC) compressive strength are age, cement, and water. For forecasting the compressive strength of concrete, the trained stacking model HGB+LGBM+LR is a highly accurate and stable machine learning model. The research's conclusions offer useful information for predicting concrete's compressive strength and enhancing mix design.