Explainable ensemble machine learning framework with particle swarm optimization for predicting compressive strength...

Area of Science:

• Materials Science and Engineering
• Sustainable Construction
• Computational Intelligence

Background:

• Growing demand for low-carbon construction materials drives interest in carbonation-cured concrete for CO₂ sequestration.
• Predicting the compressive strength of CO₂-cured concrete is challenging due to complex interactions between composition and curing.
• Existing studies often use simpler models and optimization, limiting predictive accuracy and interpretability.

Purpose of the Study:

• To develop a robust framework for predicting the compressive strength of carbonation-cured concrete.
• To integrate advanced ensemble learning models with metaheuristic optimization for enhanced predictive performance.
• To provide multi-level interpretability and probabilistic uncertainty estimation for model insights.

Main Methods:

• Seven ensemble learning models (Random Forest, CatBoost, LightGBM, NGBoost, AdaBoost, Extra Trees, DeepGBM) were employed.
• Particle Swarm Optimization (PSO) was used for hyperparameter tuning of the ensemble models.
• Interpretability techniques including SHAP, ALE, permutation importance, and counterfactual analysis were applied.
• A graphical user interface (GUI) was developed for real-time predictions.

Main Results:

• All seven PSO-optimized ensemble models demonstrated strong predictive performance (R² > 0.92).
• The Random Forest model achieved the highest accuracy with R² = 0.955 and RMSE = 3.9251 MPa.
• Cement, coarse aggregate, and water content were identified as key predictors of compressive strength.
• Carbonation curing parameters had a secondary influence on the predicted strength.

Conclusions:

• The integrated framework provides accurate and interpretable predictions for CO₂-cured concrete compressive strength.
• PSO-based optimization effectively enhances the performance of various ensemble learning algorithms.
• The study offers valuable insights into factors influencing concrete performance, supporting sustainable material design.