Machine Learning Prediction Models to Evaluate the Strength of Recycled Aggregate Concrete

Area of Science:

• Materials Science
• Civil Engineering
• Computational Science

Background:

• Recycled aggregate concrete (RAC) exhibits lower compressive and flexural strength compared to natural aggregate concrete.
• Numerous factors influence RAC strength, but their combined effects are complex to study experimentally.
• Experimental evaluation of RAC properties is time-consuming and expensive.

Purpose of the Study:

• To predict the compressive and flexural strengths of recycled aggregate concrete (RAC).
• To evaluate the effectiveness of ensemble machine learning methods for predicting RAC strength.
• To identify the most accurate machine learning model for RAC strength prediction.

Main Methods:

• Utilized ensemble machine learning techniques, specifically gradient boosting and random forest.
• Analyzed the influence of twelve input factors on RAC strength.
• Validated models using correlation coefficients (R²), variance analysis, statistical tests, and k-fold cross-validation.

Main Results:

• Random forest models achieved higher accuracy (R² of 0.91 for compressive strength, 0.86 for flexural strength) compared to gradient boosting.
• Random forest models demonstrated lower error metrics (MAE: 4.19 MPa compressive, 0.56 MPa flexural) than gradient boosting.
• The study confirmed the superior predictive capability of random forest for RAC strength.

Conclusions:

• Ensemble machine learning, particularly random forest, provides a rapid and cost-effective method for predicting RAC strength.
• Machine learning applications can significantly benefit the construction sector by streamlining material property evaluation.
• This approach overcomes the limitations of traditional experimental methods for complex material analysis.