Scholarlink

<p>The adoption of limestone calcined clay cement (LC<sup>3</sup>) as a low-carbon alternative to cement has drawn increasing interest due to its environmental benefits and promising mechanical performance. However, predicting the compressive strength of LC<sup>3</sup> mortar remains a challenge due to complex chemical and mix-design interdependencies. Initially, thirty input variables were considered to predict the compressive strength. A univariate filter-based feature-selection technique identified the top 13 most influential parameters. Five ensemble ML models, random forest, gradient boosting, extra trees, XGBoost, and CatBoost, were trained, evaluated, and compared. CatBoost achieved the best performance, with an MSE of 40.73, RMSE of 6.38, <i>R</i><sup>2</sup> of 0.863, and an EV of 0.86. Its enhanced performance was further confirmed by Radar and Taylor diagram analyses, indicating its predictive reliability compared to other models. Feature correlation and regression analysis revealed that higher CaO, SO<sub>3</sub>, and Na<sub>2</sub>O contents in OPC, binder content, and lower w/b ratio enhance compressive strength, while excessive Al<sub>2</sub>O<sub>3</sub> and LOI in CC, high calcination temperature, and overuse of gypsum reduce it. This study demonstrates the potential of ML in predicting and providing new insights into LC<sup>3</sup> mix design and oxide composition influences on mortar 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 of the Compressive Strength of Limestone Calcined Clay Cement (LC3) Mortar

  • 期刊论文

  • 发表日期：

    2026年2月23日

查看全文
  • Md Rakib Hossain,
  • Md Sajid,
  • Jamir Uddin Mamun,
  • Ebrahim Al-Wajih,
  • Jawad Khalil,
  • Amin Al-Fakih

摘要

The adoption of limestone calcined clay cement (LC3) as a low-carbon alternative to cement has drawn increasing interest due to its environmental benefits and promising mechanical performance. However, predicting the compressive strength of LC3 mortar remains a challenge due to complex chemical and mix-design interdependencies. Initially, thirty input variables were considered to predict the compressive strength. A univariate filter-based feature-selection technique identified the top 13 most influential parameters. Five ensemble ML models, random forest, gradient boosting, extra trees, XGBoost, and CatBoost, were trained, evaluated, and compared. CatBoost achieved the best performance, with an MSE of 40.73, RMSE of 6.38, R2 of 0.863, and an EV of 0.86. Its enhanced performance was further confirmed by Radar and Taylor diagram analyses, indicating its predictive reliability compared to other models. Feature correlation and regression analysis revealed that higher CaO, SO3, and Na2O contents in OPC, binder content, and lower w/b ratio enhance compressive strength, while excessive Al2O3 and LOI in CC, high calcination temperature, and overuse of gypsum reduce it. This study demonstrates the potential of ML in predicting and providing new insights into LC3 mix design and oxide composition influences on mortar strength.