Development of a Tree-Based Machine Learning Model for Predicting the Compressive Strength of Accelerated Carbonation-Cured Cementitious Composites

Key Points

• The aim is to predict the compressive strength of carbonation-cured cement composites using machine learning.
• Collected experimental data from 23 studies totaling 333 samples
• Constructed a dataset including mixture composition, CO₂ curing conditions, and aging variables
• Applied Random Forest and Gradient Boosting algorithms for prediction
• Conducted SHAP-based sensitivity analysis to interpret influential factors
• CatBoost and XGBoost models exhibited high prediction accuracy and stable generalization performance
• Identified CO₂ curing time, aggregate-to-binder ratio, and CO₂ concentration as key influencing factors
• Found that these variables demonstrated nonlinear relationships with compressive strength