Prediction of Compressive Strength of Fly Ash Based Concrete Using Individual and Ensemble Algorithm.

Area of Science:

• Civil Engineering
• Materials Science
• Computational Science

Background:

• Machine learning algorithms are increasingly utilized for predicting concrete's mechanical properties.
• Comparing individual algorithms with ensemble approaches, such as bagging, is crucial for optimizing predictive accuracy.

Purpose of the Study:

• To compare the performance of individual machine learning algorithms against ensemble methods for predicting concrete mechanical properties.
• To optimize the bagging ensemble approach by creating 20 sub-models for enhanced accuracy.

Main Methods:

• Utilized variables including cement content, aggregate size and type, water content, binder-to-water ratio, fly ash, and superplasticizer for model development.
• Employed statistical indicators such as Mean Absolute Error (MAE), Mean Square Error (MSE), and Root Mean Square Error (RMSE) for performance evaluation.
• Applied K-fold cross-validation to confirm model robustness and accuracy.

Main Results:

• The ensemble model achieved a superior R-squared value of 0.911, outperforming individual Decision Tree (DT) and Gene Expression Programming (GEP) algorithms.
• Statistical analysis indicated significant enhancements in error reduction: 25% for MAE, 121% for MSE, and 49% for RMSE when using the ensemble Decision Tree approach.
• Individual algorithms demonstrated moderate bias, whereas the ensemble model provided more reliable predictions.

Conclusions:

• Ensemble machine learning models, particularly bagging with optimized sub-models, offer superior accuracy in predicting concrete mechanical properties compared to individual algorithms.
• The findings highlight the potential of ensemble methods to enhance the reliability and precision of material property predictions in civil engineering applications.
• K-fold cross-validation confirmed the robust performance and accuracy of the developed ensemble model.