Ant Colony Optimization-Driven Ensemble Learning for Carbon Emission Modelling in Fly Ash-Slag Geopolymer Concrete

Area of Science:

• Civil and Environmental Engineering
• Materials Science
• Computational Science

Background:

• Machine learning (ML) is widely used for geopolymer concrete (GPC) mechanical properties but less for environmental impact prediction.
• Estimating carbon emissions in GPC is crucial for sustainable construction materials.
• Advanced ensemble ML techniques offer potential for accurate carbon emission prediction in GPC.

Purpose of the Study:

• To investigate and compare the performance of various ensemble ML models for predicting carbon emissions in GPC.
• To identify key factors influencing carbon emissions during GPC production.
• To evaluate the robustness and interpretability of different ML models for sustainable material design.

Main Methods:

• Employed six ensemble ML models: Random Forest, Gradient Boosting, Extreme Gradient Boosting (XGB), CatBoost, and Light Gradient Boosting Machine (LGBM).
• Optimized models using Ant Colony Optimization (ACO).
• Conducted feature importance analysis, cross-validation, uncertainty quantification, Monte Carlo simulation, and Gaussian white noise analysis.

Main Results:

• The ACO-enhanced XGB model achieved the highest predictive accuracy (R²=0.97, MAE=3.92, RMSE=6.17).
• Curing parameters (initial curing time, temperature, NaOH dosage) significantly influenced carbon emissions.
• CatBoost and ACO-gradient boosting demonstrated superior robustness and stability under noisy conditions compared to XGB-based models.

Conclusions:

• A data-driven framework for quantifying GPC carbon emissions was established.
• Model robustness and interpretability are as critical as predictive accuracy for sustainable material design.
• Intelligent modeling advances sustainable material development by accurately assessing environmental impacts.