Prediction of Geopolymer Concrete Compressive Strength Using Novel Machine Learning Algorithms

Area of Science:

• Civil Engineering
• Materials Science
• Computer Science

Background:

• Geopolymer concrete (GPC) offers a sustainable alternative to traditional concrete, reducing environmental impact.
• Accurate prediction of GPC mechanical properties is crucial for its widespread adoption in civil engineering.
• Machine learning (ML) provides powerful tools for forecasting material performance.

Purpose of the Study:

• To predict the compressive strength (CS) of high calcium fly-ash-based GPC using ML algorithms.
• To compare the predictive performance of artificial neural network (ANN), boosting, and AdaBoost ML approaches.
• To identify the most effective ML technique for GPC CS prediction.

Main Methods:

• Utilized Python coding for implementing supervised ML algorithms: ANN, boosting, and AdaBoost.
• Trained and evaluated models on datasets to predict the compressive strength of GPC.
• Performed statistical analysis, including R-squared, MAE, MSE, RMSE, and k-fold cross-validation, to assess model accuracy.

Main Results:

• Boosting demonstrated the highest predictive accuracy with R² = 0.96.
• AdaBoost also showed strong performance with R² = 0.93, outperforming the individual ANN model (R² = 0.87).
• Boosting exhibited the lowest error values (MAE, MSE, RMSE), confirming its high precision.

Conclusions:

• Ensemble ML techniques (boosting and AdaBoost) are superior to individual ANN for predicting GPC compressive strength.
• Boosting is identified as the most accurate method among those tested.
• Further exploration of ensemble methods like bagging and gradient boosting could enhance prediction accuracy.