Efficient Compressive Strength Prediction of Alkali-Activated Waste Materials Using Machine Learning.

Area of Science:

• Materials Science
• Civil Engineering
• Data Science

Background:

• Alkali-activated materials (AAMs) offer sustainable alternatives to traditional cementitious materials.
• Formulating AAMs typically involves labor-intensive trial-and-error methods.
• Industrial waste streams like blast furnace slag, fly ash, reducing slag, and waste glass are viable precursors for AAMs.

Purpose of the Study:

• To integrate machine learning (ML) for predicting and optimizing the compressive strength of AAMs.
• To streamline the AAM formulation process by reducing reliance on experimental trial-and-error.
• To identify key features influencing AAM compressive strength using ML.

Main Methods:

• Development and validation of a Random Forest (RF) ML model.
• Utilized a dataset comprising 42 samples from four industrial waste streams.
• Employed fivefold cross-validation for model reliability assessment.
• Performed meticulous data processing to identify influential features.

Main Results:

• The RF model achieved a significant improvement in prediction accuracy, increasing from 0.05 to 0.62.
• Experimental validation confirmed the ML model's efficacy, achieving desired strength thresholds.
• Achieved a 59.65% improvement in compressive strength compared to initial experiments.
• ML-recommended formulations required only approximately 5 minutes for design.

Conclusions:

• ML techniques, particularly RF, are highly effective for predicting and optimizing AAM compressive strength.
• ML integration substantially accelerates the AAM design and development process.
• The study demonstrates the transformative potential of ML in sustainable construction material development.