Scholarlink

<p>This study involves investigates the relationships between input parameters and compressive strength of concrete using a comprehensive dataset and advanced machine learning based modeling techniques. Compressive strength showed significant increases with curing time, particularly between 14 and 28&#xa0;days, with optimal performance at 6–8% silica fume (SF) and moderate fly ash (FA) levels (30–50%). SVM-RBF, Random Forest, XGBoost based machine learning models along with non-parametric and linear regression models were developed in the current study. Among the models, XGBoost achieved the highest predictive performance (R<sup>2</sup>: 1.000 in training, 0.999 in testing), outperforming Random Forest and SVM-RBF in accuracy and robustness. Linear and non-parametric regressions exhibited higher errors, emphasizing the necessity of advanced approaches for complex data. Taylor diagrams for the models in training and testing phases also advocated the robustness of XGBoost model. Sensitivity analysis of the XGBoost model shows curing duration (76.844%) as the most critical factor Monotonicity analysis highlighted intricate nonlinear relationships, such as SF and coarse aggregate effects, which were overlooked by basic linear fittings. These findings demonstrate XGBoost’s capability to model complex dynamics, providing actionable insights into optimizing concrete mix design for enhanced compressive strength.</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

Machine learning based prediction models for the compressive strength of high-volume fly ash concrete reinforced with silica fume

  • 期刊论文

  • 发表日期：

    2025年3月14日

查看全文
  • Anish Kumar,
  • Sameer Sen,
  • Sanjeev Sinha

摘要

This study involves investigates the relationships between input parameters and compressive strength of concrete using a comprehensive dataset and advanced machine learning based modeling techniques. Compressive strength showed significant increases with curing time, particularly between 14 and 28 days, with optimal performance at 6–8% silica fume (SF) and moderate fly ash (FA) levels (30–50%). SVM-RBF, Random Forest, XGBoost based machine learning models along with non-parametric and linear regression models were developed in the current study. Among the models, XGBoost achieved the highest predictive performance (R2: 1.000 in training, 0.999 in testing), outperforming Random Forest and SVM-RBF in accuracy and robustness. Linear and non-parametric regressions exhibited higher errors, emphasizing the necessity of advanced approaches for complex data. Taylor diagrams for the models in training and testing phases also advocated the robustness of XGBoost model. Sensitivity analysis of the XGBoost model shows curing duration (76.844%) as the most critical factor Monotonicity analysis highlighted intricate nonlinear relationships, such as SF and coarse aggregate effects, which were overlooked by basic linear fittings. These findings demonstrate XGBoost’s capability to model complex dynamics, providing actionable insights into optimizing concrete mix design for enhanced compressive strength.