Prediction of compressive strength of high-performance concrete using optimization machine learning approaches with SHAP analysis

Forecasting the compressive strength of high-performance concrete (HPC) is crucial for its practical applications. However, conducting experimental tests for this purpose demands significant resources and time. In recent times, data-driven modeling, particularly machine learning, has become prominent in this field. Thus, this study employed machine learning methods, including Gradient Boosting, Random Forest, and CatBoost algorithms, to predict the compressive strength of high-performance concrete (HPC). Furthermore, interpretable algorithms like SHapley Additive exPlanations (SHAP) were utilized in this study to assess the contribution of input variables to predictions. The dataset includes parameters such as cement, blast furnace slag, fly ash, water, coarse aggregate, sand, and age as input variables, with compressive strength as the output variable for models development. The CatBoost model performance was superior to GBM and RF models, with R² values of 0.979 for the training stage and 0.959 for the testing stage. Moreover, it demonstrated minimal mean absolute errors (MAE) of 2.37 MPa and 3 MPa for compressive strength prediction in the training and testing stages, respectively. Furthermore, the SHAP analysis revealed that age, cement, and superplasticizer are the primary influencers in determining compressive strength. The machine learning models presented in this study exhibit promising potential for practical implementation in different engineering applications.