Compressive Strength of Steel Fiber-Reinforced Concrete Employing Supervised Machine Learning Techniques.

Area of Science:

• Materials Science
• Civil Engineering
• Computational Mechanics

Background:

• Experimental determination of material properties is time-consuming and costly.
• Supervised machine learning offers a potential alternative for predicting material characteristics.
• Steel fiber-reinforced concrete (SFRC) requires accurate strength prediction for structural applications.

Purpose of the Study:

• To evaluate individual and ensemble machine learning models for predicting the 28-day compressive strength of SFRC.
• To compare the predictive accuracy and efficiency of Support Vector Regression (SVR), SVR AdaBoost, and SVR bagging techniques.
• To assess the influence of input parameters on prediction outcomes using a sensitivity analysis.

Main Methods:

• Utilized individual (SVR) and ensemble (SVR AdaBoost, SVR bagging) machine learning models.
• Employed coefficient of determination (R²), statistical assessment, and k-fold cross-validation for performance evaluation.
• Applied a sensitivity analysis technique to determine parameter influence on predictions.

Main Results:

• All evaluated machine learning approaches demonstrated effective prediction capabilities for SFRC compressive strength.
• The SVR AdaBoost model achieved the highest accuracy with R² = 0.96.
• SVR bagging (R² = 0.87) and SVR (R² = 0.81) also showed satisfactory predictive performance, with SVR AdaBoost exhibiting the lowest error metrics (MAE = 4.4 MPa, RMSE = 8 MPa).

Conclusions:

• Ensemble machine learning, particularly SVR AdaBoost, provides a highly accurate and efficient method for predicting SFRC mechanical properties.
• These computational approaches can significantly reduce the need for extensive experimental testing.
• The developed models hold potential for application in predicting the mechanical characteristics of various other construction materials.