Scholarlink

<p>This study aims to enhance concrete compressive strength (CSC) estimation by emphasizing the advantages of ensemble models over single models. We integrated the Reduced Error Pruning Tree (REPT) method with Bagging and Rotation Forest ensemble learning and compared their performance with single Support Vector Machines (SVM). During the training phase, the ensemble Rotation Forest-REPT (ROF-REPT) model demonstrated the highest correlation coefficient (<i>R</i> = 0.972), the lowest mean absolute error (MAE = 2.985), and the lowest root mean square error (RMSE = 2.029&#xa0;MPa) among the models, indicating a high accuracy and predictive performance. The REPT model also performed well, with a high R (0.955) and a low MAE (3.739&#xa0;MPa), but a higher RMSE (2.662) than ROF-REPT. The ensemble Bagging-REPT (BGG-REPT) model fell between REPT and ROF-REPT in terms of performance metrics, with a high R (0.963), a low MAE (3.425&#xa0;MPa), and a low RMSE (2.422&#xa0;MPa). The SVM model exhibited comparatively weaker performance. In the validation phase, the ensemble BGG-REPT model achieved the best performance in R (0.879), indicating a strong linear relationship between predicted and observed values. SVM also showed a moderate MAE (6.02) and a low RMSE (4.649&#xa0;MPa), suggesting a reasonable average absolute difference and high overall predictive accuracy. The ensemble ROF-REPT model was similar to BGG-REPT in terms of R (0.876) and RMSE (5.151&#xa0;MPa), but had a slightly higher MAE (6.448&#xa0;MPa). The REPT model had a moderate performance in R (0.774) and RMSE (7.706&#xa0;MPa), but a higher MAE (5.779&#xa0;MPa) than BGG-REPT and ROF-REPT. The SVM model had the lowest performance in R (0.821) and the highest MAE (7.097&#xa0;MPa) among the models, but presented the lowest RMSE (5.49). This study contributes to the existing literature by providing a comprehensive evaluation of the performance of different methods in CSC estimation. The findings suggest that ensemble modeling approaches, can significantly enhance the accuracy of CSC estimations. This has important implications for the construction industry and opens up new avenues for future research in this area.</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

BGG-REPT and ROF-REPT: ensemble machine learning models for the prediction of compressive strength of concrete

  • 期刊论文

  • 发表日期：

    2025年1月28日

查看全文
  • Binh Thai Pham

摘要

This study aims to enhance concrete compressive strength (CSC) estimation by emphasizing the advantages of ensemble models over single models. We integrated the Reduced Error Pruning Tree (REPT) method with Bagging and Rotation Forest ensemble learning and compared their performance with single Support Vector Machines (SVM). During the training phase, the ensemble Rotation Forest-REPT (ROF-REPT) model demonstrated the highest correlation coefficient (R = 0.972), the lowest mean absolute error (MAE = 2.985), and the lowest root mean square error (RMSE = 2.029 MPa) among the models, indicating a high accuracy and predictive performance. The REPT model also performed well, with a high R (0.955) and a low MAE (3.739 MPa), but a higher RMSE (2.662) than ROF-REPT. The ensemble Bagging-REPT (BGG-REPT) model fell between REPT and ROF-REPT in terms of performance metrics, with a high R (0.963), a low MAE (3.425 MPa), and a low RMSE (2.422 MPa). The SVM model exhibited comparatively weaker performance. In the validation phase, the ensemble BGG-REPT model achieved the best performance in R (0.879), indicating a strong linear relationship between predicted and observed values. SVM also showed a moderate MAE (6.02) and a low RMSE (4.649 MPa), suggesting a reasonable average absolute difference and high overall predictive accuracy. The ensemble ROF-REPT model was similar to BGG-REPT in terms of R (0.876) and RMSE (5.151 MPa), but had a slightly higher MAE (6.448 MPa). The REPT model had a moderate performance in R (0.774) and RMSE (7.706 MPa), but a higher MAE (5.779 MPa) than BGG-REPT and ROF-REPT. The SVM model had the lowest performance in R (0.821) and the highest MAE (7.097 MPa) among the models, but presented the lowest RMSE (5.49). This study contributes to the existing literature by providing a comprehensive evaluation of the performance of different methods in CSC estimation. The findings suggest that ensemble modeling approaches, can significantly enhance the accuracy of CSC estimations. This has important implications for the construction industry and opens up new avenues for future research in this area.