Scholarlink

<p>To explore how the carbonation treatment improve the quality of recycled concrete aggregate and to understand the relationship among key parameters such as mix proportions, duration of carbonation treatment on recycled aggregate, concentration of CO<sub>2</sub> maintained and relative humidity maintained during carbonation treatment, achieved compressive strength, a complete dataset of experimental findings gathered from the literature was built. Two conventional machine learning (CML) models, namely linear regression (LR), support vector machines (SVM), and two ensemble machine learning (EML) models, such as random forest (RF), and gradient boosting (GB) were adopted to forecast the compressive strength of carbonated recycled aggregate concrete (CRCA). The models’ performance was thoroughly evaluated by advanced visualization and interpretability techniques, specifically employing Taylor diagram for robust statistical validation against observed patterns, and SHAP (SHapley Additive exPlanations) analysis to elucidate feature contributions and model behavior. Results indicated that the coefficient of determination (R<sup>2</sup>) of LR, SVM, RF and GB were 0.5777, 0.5385, 0.7070 and 0.8561 respectively, showing the superior suitability and reliability of Ensemble Machine Learning (EML) models over Conventional Machine Learning (CML) models to foresee concrete compressive strength of carbonated recycled aggregate concrete. Feature importance analysis using SHAP analysis revealed cement content, w/c ratio and carbonated recycled aggregate content as significant characteristics that affect mechanical performance of CRCA, with SHAP values of 5.32, 3.62 and 2.75 respectively. Furthermore Taylor diagram revealed that, Gradient Boosting (GB) proved to be the utmost accurate and robust model, with the Random Forest (RF) model immediately following. Conversely, Logistic Regression and Support Vector Machine showed reduced performance. The machine learning models in this study show great promise for real-world use in various engineering applications.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Comparative study on the performance of carbonated recycled aggregate concrete using conventional and ensemble machine learning models

  • 期刊论文

  • 发表日期：

    2026年1月5日

查看全文
  • Murali Kannan Sundharam Paulpandian,
  • Kelita Samuel,
  • N. Arunachelam

摘要

To explore how the carbonation treatment improve the quality of recycled concrete aggregate and to understand the relationship among key parameters such as mix proportions, duration of carbonation treatment on recycled aggregate, concentration of CO2 maintained and relative humidity maintained during carbonation treatment, achieved compressive strength, a complete dataset of experimental findings gathered from the literature was built. Two conventional machine learning (CML) models, namely linear regression (LR), support vector machines (SVM), and two ensemble machine learning (EML) models, such as random forest (RF), and gradient boosting (GB) were adopted to forecast the compressive strength of carbonated recycled aggregate concrete (CRCA). The models’ performance was thoroughly evaluated by advanced visualization and interpretability techniques, specifically employing Taylor diagram for robust statistical validation against observed patterns, and SHAP (SHapley Additive exPlanations) analysis to elucidate feature contributions and model behavior. Results indicated that the coefficient of determination (R2) of LR, SVM, RF and GB were 0.5777, 0.5385, 0.7070 and 0.8561 respectively, showing the superior suitability and reliability of Ensemble Machine Learning (EML) models over Conventional Machine Learning (CML) models to foresee concrete compressive strength of carbonated recycled aggregate concrete. Feature importance analysis using SHAP analysis revealed cement content, w/c ratio and carbonated recycled aggregate content as significant characteristics that affect mechanical performance of CRCA, with SHAP values of 5.32, 3.62 and 2.75 respectively. Furthermore Taylor diagram revealed that, Gradient Boosting (GB) proved to be the utmost accurate and robust model, with the Random Forest (RF) model immediately following. Conversely, Logistic Regression and Support Vector Machine showed reduced performance. The machine learning models in this study show great promise for real-world use in various engineering applications.