Scholarlink

<p>This study aims to develop precise and interpretable predictive models for estimating the compressive strength (CS) of Rice Husk Ash (RH Ash) based concrete through the application of symbolic machine learning techniques. Given the increasing emphasis on sustainable construction materials, it is essential that predictive models are both accurate and explainable. In this research, Gene Expression Programming (GEP) and Multi-Expression Programming (MEP) algorithms were employed using six critical input variables: cement, RH Ash, water, superplasticizer, age, and fine aggregate. The GEP model achieved R<sup>2</sup> values of 0.87 (training), 0.91 (testing), and 0.84 (validation), whereas the MEP model demonstrated superior performance with R<sup>2</sup> = 0.93, RMSE = 4.73&#xa0;MPa, and MAE = 3.88&#xa0;MPa. To enhance model transparency, SHAP (SHapley Additive exPlanations) analysis was conducted. Cement (mean SHAP value ≈ 0.60) and specimen age (≈ 0.52) were identified as the most influential predictors of CS. Water ( ≈ − 0.48) consistently exhibited a negative contribution, while RH Ash demonstrated an optimal non-linear influence (≈ 0.41), underscoring the importance of dosage control. Fine aggregate and superplasticizer exhibited lower contributions (≈ 0.28 and ≈ 0.21, respectively). The integration of symbolic machine learning and SHAP-based interpretation not only enhances predictive capability but also provides engineering insights for mix design optimization. This research will contribute to the development of performance-based design frameworks for sustainable concrete, offering a valuable tool for future research and construction industry applications involving industrial by-products such as Rice Husk Ash.</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

Enhancing compressive strength prediction of sustainable concrete using MEP and GEP models with SHAP-based interpretation

  • 期刊论文

  • 发表日期：

    2025年8月12日

查看全文
  • Sujin George,
  • Syed Aamir Hussain,
  • Ahmed A. Alamiery,
  • Mohammad Gulfam Pathan,
  • Syed Sabihuddin,
  • Nisha Thakur,
  • Ali Majdi,
  • Aseel Smerat

摘要

This study aims to develop precise and interpretable predictive models for estimating the compressive strength (CS) of Rice Husk Ash (RH Ash) based concrete through the application of symbolic machine learning techniques. Given the increasing emphasis on sustainable construction materials, it is essential that predictive models are both accurate and explainable. In this research, Gene Expression Programming (GEP) and Multi-Expression Programming (MEP) algorithms were employed using six critical input variables: cement, RH Ash, water, superplasticizer, age, and fine aggregate. The GEP model achieved R2 values of 0.87 (training), 0.91 (testing), and 0.84 (validation), whereas the MEP model demonstrated superior performance with R2 = 0.93, RMSE = 4.73 MPa, and MAE = 3.88 MPa. To enhance model transparency, SHAP (SHapley Additive exPlanations) analysis was conducted. Cement (mean SHAP value ≈ 0.60) and specimen age (≈ 0.52) were identified as the most influential predictors of CS. Water ( ≈ − 0.48) consistently exhibited a negative contribution, while RH Ash demonstrated an optimal non-linear influence (≈ 0.41), underscoring the importance of dosage control. Fine aggregate and superplasticizer exhibited lower contributions (≈ 0.28 and ≈ 0.21, respectively). The integration of symbolic machine learning and SHAP-based interpretation not only enhances predictive capability but also provides engineering insights for mix design optimization. This research will contribute to the development of performance-based design frameworks for sustainable concrete, offering a valuable tool for future research and construction industry applications involving industrial by-products such as Rice Husk Ash.