Machine learning driven optimization of biomedical waste ash concrete for sustainable construction

Area of Science:

• Materials Science
• Civil Engineering
• Environmental Science

Background:

• Cement production significantly contributes to environmental pollution.
• Biomedical waste ash (BMWA) presents a potential sustainable alternative for cementitious materials.
• Developing reliable methods to predict BMWA concrete properties is crucial for its adoption.

Purpose of the Study:

• To develop a database of BMWA concrete properties from existing literature.
• To apply advanced Machine Learning (ML) techniques for predicting BMWA concrete mechanical properties.
• To evaluate the sustainability of BMWA concrete using a comprehensive index.

Main Methods:

• Compilation of a BMWA concrete properties database.
• Implementation and evaluation of four ML models: Random Forest (RF), SAINT, TabNet, and an Ensemble model.
• Hyperparameter optimization and ten-fold cross-validation for model performance assessment.
• External validation and calculation of a Sustainability Index (SI).

Main Results:

• RF and TabNet models achieved the highest predictive accuracy (R² = 0.82) for strength parameters.
• SAINT showed stable generalization but lower accuracy for specific strength parameters.
• The ensemble model underperformed compared to standalone models, highlighting the effectiveness of robust individual models.
• External validation confirmed the reliability of the developed ML models.
• BMWA concrete at 15% substitution demonstrated positive sustainability indicators.

Conclusions:

• An integrated data-driven framework combining ML, interpretability, and sustainability indexing is proposed.
• The study validates the potential of BMWA concrete as a sustainable construction material.
• This approach supports the circular economy by utilizing biomedical waste and reducing the cement industry's environmental footprint.