Scholarlink

<p>Accurate prediction and classification of concrete compressive strength (CS) are essential for optimizing material selection and ensuring structural integrity. Traditional experimental methods are time-consuming and resource-intensive, necessitating advanced computational approaches. This study developed a hybrid machine learning framework for the simultaneous prediction and classification of concrete CS into three classes: Low-strength concrete (LSC), Normal-strength concrete (NSC), and High-strength concrete (HSC). Five Gradient Boosting Models (GBMs): Stochastic Gradient Boosting (SGB)f, Extreme Gradient Boosting (XGB), Light Gradient Boosting (LGB), Categorical Boosting (CGB), and Natural Gradient Boosting (NGB) were conducted, with Bayesian Optimization (BO) and k-fold cross-validation (CV) employed for hyperparameter tuning. Model interpretability was ensured using Shapley Additive Explanations (SHAP) to analyze feature importance. A comprehensive evaluation framework integrating both visual and quantitative assessments was conducted, including regression performance metrics, classification accuracy, and uncertainty analysis to ensure robust model comparison. Results showed that the SGB demonstrated the highest accuracy (0.925 for LSC and 0.902 for HSC) in CS classification, followed by LGB, while XGB ranked third, and CGB and NGB exhibited lower classification accuracy. For prediction, XGB achieved the highest accuracy (R<sup>2</sup> = 0.967, RMSE = 3.072 MPa), followed by LGB (R<sup>2</sup> = 0.966, RMSE = 3.052 MPa), confirming their superior generalization ability. SHAP analysis revealed that cement content and curing duration were the most influential factors affecting CS. To enhance practical usability, a Graphical User Interface (GUI) was developed, integrating the best-performing model to provide real-time CS predictions and classifications. This user-friendly tool facilitates AI-driven decision-making in structural engineering and material selection. The findings confirm that the proposed hybrid ML framework effectively optimizes predictive accuracy, model interpretability, and real-world applicability, contributing to advancements in AI-driven concrete strength assessment and construction technology.</p> Graphical abstract <p></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

Hybrid gradient boosting models for concrete compressive strength classification and prediction

  • 期刊论文

  • 发表日期：

    2025年8月25日

查看全文
  • Mohamed Kamel Elshaarawy,
  • Abdelrahman Kamal Hamed,
  • Mostafa M. Alsaadawi

摘要

Accurate prediction and classification of concrete compressive strength (CS) are essential for optimizing material selection and ensuring structural integrity. Traditional experimental methods are time-consuming and resource-intensive, necessitating advanced computational approaches. This study developed a hybrid machine learning framework for the simultaneous prediction and classification of concrete CS into three classes: Low-strength concrete (LSC), Normal-strength concrete (NSC), and High-strength concrete (HSC). Five Gradient Boosting Models (GBMs): Stochastic Gradient Boosting (SGB)f, Extreme Gradient Boosting (XGB), Light Gradient Boosting (LGB), Categorical Boosting (CGB), and Natural Gradient Boosting (NGB) were conducted, with Bayesian Optimization (BO) and k-fold cross-validation (CV) employed for hyperparameter tuning. Model interpretability was ensured using Shapley Additive Explanations (SHAP) to analyze feature importance. A comprehensive evaluation framework integrating both visual and quantitative assessments was conducted, including regression performance metrics, classification accuracy, and uncertainty analysis to ensure robust model comparison. Results showed that the SGB demonstrated the highest accuracy (0.925 for LSC and 0.902 for HSC) in CS classification, followed by LGB, while XGB ranked third, and CGB and NGB exhibited lower classification accuracy. For prediction, XGB achieved the highest accuracy (R2 = 0.967, RMSE = 3.072 MPa), followed by LGB (R2 = 0.966, RMSE = 3.052 MPa), confirming their superior generalization ability. SHAP analysis revealed that cement content and curing duration were the most influential factors affecting CS. To enhance practical usability, a Graphical User Interface (GUI) was developed, integrating the best-performing model to provide real-time CS predictions and classifications. This user-friendly tool facilitates AI-driven decision-making in structural engineering and material selection. The findings confirm that the proposed hybrid ML framework effectively optimizes predictive accuracy, model interpretability, and real-world applicability, contributing to advancements in AI-driven concrete strength assessment and construction technology.

Graphical abstract