Scholarlink

<p>This paper develops an explainable, data-driven framework to predict the compressive strength of concrete mixes containing Ground Granulated Blast-Furnace Slag (GGBS). Seven supervised regressors were trained, tested, and validated using a diverse dataset of 542 mixes, each covering a wide range of variations in quantities of cement, water, GGBS, aggregates, superplasticizer, and curing ages. Their performances were evaluated on the training, validation, and independent test sets using MAE, MSE, RMSE, and R². Visual plots (actual vs. predicted values) and SHAP global explanations were used to probe the learned feature influences. Gradient Boosting yielded the best generalization on unseen data (for test set, R<sup>2</sup> = 0.977, RMSE = 2.658, MAE = 1.571), followed by CatBoost (for test set, R<sup>2</sup> = 0.970, RMSE = 2.998, MAE = 1.610) and LightGBM (for test set, R<sup>2</sup> = 0.968, RMSE = 3.136, MAE = 2.245). SHAP analysis identified the major influencing parameters of strength, including curing age, water content, cement content, and GGBS proportions, in a consistent manner across models, thereby aligning model behavior with concrete performance. The novel elements in the present workflow are the development of a multi-model, test-centric benchmark with SHAP-driven interpretability focused on GGBS-rich mixtures that allow reliable, transparent predictions and directly offer guidelines toward sustainable mix optimization.</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

An enhanced prediction of compressive strength of GGBS incorporated concrete using explainable gradient-boosting algorithms

  • 期刊论文

  • 发表日期：

    2025年12月16日

查看全文
  • Swamy Charan Dass Zakkam,
  • Rama Rao Panugalla

摘要

This paper develops an explainable, data-driven framework to predict the compressive strength of concrete mixes containing Ground Granulated Blast-Furnace Slag (GGBS). Seven supervised regressors were trained, tested, and validated using a diverse dataset of 542 mixes, each covering a wide range of variations in quantities of cement, water, GGBS, aggregates, superplasticizer, and curing ages. Their performances were evaluated on the training, validation, and independent test sets using MAE, MSE, RMSE, and R². Visual plots (actual vs. predicted values) and SHAP global explanations were used to probe the learned feature influences. Gradient Boosting yielded the best generalization on unseen data (for test set, R2 = 0.977, RMSE = 2.658, MAE = 1.571), followed by CatBoost (for test set, R2 = 0.970, RMSE = 2.998, MAE = 1.610) and LightGBM (for test set, R2 = 0.968, RMSE = 3.136, MAE = 2.245). SHAP analysis identified the major influencing parameters of strength, including curing age, water content, cement content, and GGBS proportions, in a consistent manner across models, thereby aligning model behavior with concrete performance. The novel elements in the present workflow are the development of a multi-model, test-centric benchmark with SHAP-driven interpretability focused on GGBS-rich mixtures that allow reliable, transparent predictions and directly offer guidelines toward sustainable mix optimization.