Scholarlink

<p>Accurate prediction of concrete compressive strength (CCS) is critical for ensuring structural safety and advancing sustainable infrastructure. Although numerous studies have applied machine learning to CCS prediction, most have been constrained by small datasets, limited algorithm comparisons, and inadequate interpretability. This study addresses these gaps by developing an integrated prediction–interpretation framework using a large dataset of 1134 high-performance concrete (HPC) mixes. Nine ML techniques were systematically evaluated, including support vector machine (SVM), Gaussian process regression (GPR), gene expression programming (GEP), random forest (RF), decision tree (DT), ensemble boosted tree (EBT), adaptive boosting (AB), gradient boosting (GBM), and artificial neural networks (ANN, Levenberg–Marquardt and scaled conjugate gradient). Among these, GPR achieved the highest predictive accuracy (R² = 0.87, RMSE = 5.07&#xa0;MPa, MAE = 2.9&#xa0;MPa), followed closely by SVM. To ensure interpretability, Shapley additive explanations (SHAP) were employed, revealing that testing age (28.5%), cement content (26.3%), and water-to-cement ratio (11%) were the most influential features, while fly ash exhibited minimal impact. The proposed framework thus combines predictive accuracy with transparent interpretation, offering a practical tool for mix design optimization. By leveraging a large dataset, a broad set of algorithms, and SHAP-based explainability, this study contributes a novel, multi-scale approach to advancing machine learning applications in HPC strength prediction.</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

Computational Intelligence for Predicting the Strength of High-Performance Concrete: Ensemble and Non-ensemble Machine Learning Framework

  • 期刊论文

  • 发表日期：

    2025年12月23日

查看全文
  • Munir Iqbal,
  • Ali Husnain,
  • Muhammad Ashraf

摘要

Accurate prediction of concrete compressive strength (CCS) is critical for ensuring structural safety and advancing sustainable infrastructure. Although numerous studies have applied machine learning to CCS prediction, most have been constrained by small datasets, limited algorithm comparisons, and inadequate interpretability. This study addresses these gaps by developing an integrated prediction–interpretation framework using a large dataset of 1134 high-performance concrete (HPC) mixes. Nine ML techniques were systematically evaluated, including support vector machine (SVM), Gaussian process regression (GPR), gene expression programming (GEP), random forest (RF), decision tree (DT), ensemble boosted tree (EBT), adaptive boosting (AB), gradient boosting (GBM), and artificial neural networks (ANN, Levenberg–Marquardt and scaled conjugate gradient). Among these, GPR achieved the highest predictive accuracy (R² = 0.87, RMSE = 5.07 MPa, MAE = 2.9 MPa), followed closely by SVM. To ensure interpretability, Shapley additive explanations (SHAP) were employed, revealing that testing age (28.5%), cement content (26.3%), and water-to-cement ratio (11%) were the most influential features, while fly ash exhibited minimal impact. The proposed framework thus combines predictive accuracy with transparent interpretation, offering a practical tool for mix design optimization. By leveraging a large dataset, a broad set of algorithms, and SHAP-based explainability, this study contributes a novel, multi-scale approach to advancing machine learning applications in HPC strength prediction.