Scholarlink

<p>The increasing demand for sustainable construction has driven the use of supplementary cementitious materials (SCMs) as partial replacements for cement in blended concrete. This study proposes a machine learning (ML)-based framework for predicting the compressive strength of binary and ternary blended concrete mixes, incorporating SCMs such as fly ash, silica fume, rice husk ash, Alccofine, and metakaolin. A unified predictive model was developed by employing categorical encoding (specifically one-hot encoding) to integrate diverse mix compositions into a single dataset, enabling efficient generalization across multiple SCM combinations. Four ML models—Gene Expression Programming (GEP), M5-Model Tree, Group Method of Data Handling Neural Network (GMDH-NN), and Feedforward Backpropagation Neural Network (FFBP-NN)—were trained and evaluated using a compiled dataset covering a compressive strength range of 8 to 72&#xa0;MPa. Among these, the FFBP-NN model demonstrated the highest performance, achieving R² values of 0.9887 for binary mixes and 0.9606 for ternary mixes. The M5-Model Tree performed best among interpretable models, effectively capturing the influence of curing time on strength development. These results underscore the value of integrating numerical and categorical features to improve prediction accuracy. This study provides a scalable approach for modeling blended concrete behavior and advocates for future research to incorporate larger datasets, more SCM types, and environmental parameters—such as location and exposure—as additional categorical inputs.</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 Compressive Strength of Blended Concrete Using Categorical Encoding of Cementitious Materials

  • 期刊论文

  • 发表日期：

    2025年9月16日

查看全文
  • Prasenjit Sanyal

摘要

The increasing demand for sustainable construction has driven the use of supplementary cementitious materials (SCMs) as partial replacements for cement in blended concrete. This study proposes a machine learning (ML)-based framework for predicting the compressive strength of binary and ternary blended concrete mixes, incorporating SCMs such as fly ash, silica fume, rice husk ash, Alccofine, and metakaolin. A unified predictive model was developed by employing categorical encoding (specifically one-hot encoding) to integrate diverse mix compositions into a single dataset, enabling efficient generalization across multiple SCM combinations. Four ML models—Gene Expression Programming (GEP), M5-Model Tree, Group Method of Data Handling Neural Network (GMDH-NN), and Feedforward Backpropagation Neural Network (FFBP-NN)—were trained and evaluated using a compiled dataset covering a compressive strength range of 8 to 72 MPa. Among these, the FFBP-NN model demonstrated the highest performance, achieving R² values of 0.9887 for binary mixes and 0.9606 for ternary mixes. The M5-Model Tree performed best among interpretable models, effectively capturing the influence of curing time on strength development. These results underscore the value of integrating numerical and categorical features to improve prediction accuracy. This study provides a scalable approach for modeling blended concrete behavior and advocates for future research to incorporate larger datasets, more SCM types, and environmental parameters—such as location and exposure—as additional categorical inputs.