Scholarlink

<p>The marble industry holds immense significance within the mining sector, gradually evolving to meet the growing demand for marble dimensional stone and continuously expanding its manufacturing capacity. As development and production escalate, the quantity of marble waste powder (MWP) generated has become a pressing concern for this industry. Due to the inert nature of MWP, it cannot serve as a complete substitute for cement; hence, a partial replacement of up to 20% is preferred. This research employs various supervised data-driven machine learning (ML) algorithms to predict the compressive strength of MWP-incorporated concrete composites. The study utilizes a dataset consisting of 647 mix designs and eleven features. Apart from MWP, the concrete mixes include cementitious materials such as silica fume, fly ash, and granite powder. The dataset is split into 80% for training and 20% for testing. Ten supervised data-driven ML algorithms, namely multivariate linear regression, support vector machine, artificial neural networks, decision tree regressor, random forest regressor, adaptive boosting regressor, light gradient boosting machine, gradient boosting regressor, extreme gradient boosting, and Categorical Boosting (CatBoost), are employed to forecast the compressive strength. The performance of each model is evaluated using statistical metrics: coefficient of determination, root mean square error, mean absolute error and mean absolute percentage error. Hyperparameter tuning and a <i>k-fold</i> cross-validation technique enhance each model’s efficiency. Also, the feature importance for the best models is determined using Shapely additive explanation dependency plots. Following extensive research, it has been determined that among the data-driven models employed in the study, the CatBoost model stands out as the most proficient in predicting the compressive strength of concrete incorporating MWP. Additionally, XGBoost and LGBM emerged as strong contenders, demonstrating exceptional predictive capabilities, thus enhancing understanding of the top-performing models in this context. Moreover, the Shapley additive explanation (SHAP) analysis results revealed that features such as curing days, silica fume, superplasticizer, and MWP hold higher feature importance compared to others, shedding light on the key factors contributing to the predictive performance of the CatBoost model.</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

Predicting compressive strength of concrete composites with marble waste powder as a supplementary cementitious material using supervised data-driven models

  • 期刊论文

  • 发表日期：

    2025年5月30日

查看全文
  • Pala Ravikanth,
  • T. Jothi Saravanan,
  • K. I. Syed Ahmed Kabeer

摘要

The marble industry holds immense significance within the mining sector, gradually evolving to meet the growing demand for marble dimensional stone and continuously expanding its manufacturing capacity. As development and production escalate, the quantity of marble waste powder (MWP) generated has become a pressing concern for this industry. Due to the inert nature of MWP, it cannot serve as a complete substitute for cement; hence, a partial replacement of up to 20% is preferred. This research employs various supervised data-driven machine learning (ML) algorithms to predict the compressive strength of MWP-incorporated concrete composites. The study utilizes a dataset consisting of 647 mix designs and eleven features. Apart from MWP, the concrete mixes include cementitious materials such as silica fume, fly ash, and granite powder. The dataset is split into 80% for training and 20% for testing. Ten supervised data-driven ML algorithms, namely multivariate linear regression, support vector machine, artificial neural networks, decision tree regressor, random forest regressor, adaptive boosting regressor, light gradient boosting machine, gradient boosting regressor, extreme gradient boosting, and Categorical Boosting (CatBoost), are employed to forecast the compressive strength. The performance of each model is evaluated using statistical metrics: coefficient of determination, root mean square error, mean absolute error and mean absolute percentage error. Hyperparameter tuning and a k-fold cross-validation technique enhance each model’s efficiency. Also, the feature importance for the best models is determined using Shapely additive explanation dependency plots. Following extensive research, it has been determined that among the data-driven models employed in the study, the CatBoost model stands out as the most proficient in predicting the compressive strength of concrete incorporating MWP. Additionally, XGBoost and LGBM emerged as strong contenders, demonstrating exceptional predictive capabilities, thus enhancing understanding of the top-performing models in this context. Moreover, the Shapley additive explanation (SHAP) analysis results revealed that features such as curing days, silica fume, superplasticizer, and MWP hold higher feature importance compared to others, shedding light on the key factors contributing to the predictive performance of the CatBoost model.