Scholarlink

<p>Fly ash-based geopolymer (FAG) concrete has gained attention as a sustainable alternative to conventional Portland cement (PC) concrete due to its eco-friendly composition and superior mechanical properties. However, existing studies on predicting its compressive strength (CS) using machine learning (ML) are often limited by small datasets, narrow CS ranges, and insufficient comparative analysis of ML algorithms. Addressing these gaps, this study leverages a robust dataset of 635 samples compiled from the literature, covering an extensive CS range of 4.5 to 118.9&#xa0;MPa, significantly broader than previous studies. Ten ML models, including random forest (RF), decision tree regression (DTR), support vector regression (SVR), k-nearest neighbours (KNN), extra trees regression (ETR), bagging (BG), multi-layer perceptron (MLP), and the less commonly used gradient boost regression (GBR), adaptive boosting (ADA), and extreme gradient boost (XGB), were developed and rigorously evaluated using an 80:20 training-test split. The models’ performances were assessed using eight evaluation metrics, with GBR achieving the highest accuracy (R<sup>2</sup> = 0.96) on the test set, while ADA excelled in K-fold cross-validation (R<sup>2</sup> = 0.95). However, the prediction R<sup>2</sup> of RF, DTR, SVR, KNN, ETR, BG, MLP, LR and XGB on test set predictions was recorded as 0.71, 0.73, 0.41, 0.69, 0.71, 0.77, 0.54, 0.31 and 0.76, respectively. Therefore, to evaluate these mixed performance trends, Friedman and post-hoc Conover tests were employed to categorize the models into best, good, moderate and poor performing groups at a 0.05 significance level. By examining relative performance of diverse ML models, this work offers insights pertinence of different ML algorithms for CS prediction of FAG concrete. Lastly, multi objective pareto fronts have been presented, further guiding towards sustainable and cost-effective design of FAG concrete. By and large, the study aims at promoting circularity in construction sector by guiding ML assisted development and optimisation FAG concrete for diverse applications.</p> Graphical Abstract <p></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

Predictive modeling on compressive strength of fly ash-based geopolymer concrete using machine learning approaches

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年10月28日

查看全文
  • Tejinderpal Singh,
  • Rajwinder Singh,
  • Paramveer Singh,
  • Ayush Rathore,
  • Kanish Kapoor,
  • Mahesh Patel

摘要

Fly ash-based geopolymer (FAG) concrete has gained attention as a sustainable alternative to conventional Portland cement (PC) concrete due to its eco-friendly composition and superior mechanical properties. However, existing studies on predicting its compressive strength (CS) using machine learning (ML) are often limited by small datasets, narrow CS ranges, and insufficient comparative analysis of ML algorithms. Addressing these gaps, this study leverages a robust dataset of 635 samples compiled from the literature, covering an extensive CS range of 4.5 to 118.9 MPa, significantly broader than previous studies. Ten ML models, including random forest (RF), decision tree regression (DTR), support vector regression (SVR), k-nearest neighbours (KNN), extra trees regression (ETR), bagging (BG), multi-layer perceptron (MLP), and the less commonly used gradient boost regression (GBR), adaptive boosting (ADA), and extreme gradient boost (XGB), were developed and rigorously evaluated using an 80:20 training-test split. The models’ performances were assessed using eight evaluation metrics, with GBR achieving the highest accuracy (R2 = 0.96) on the test set, while ADA excelled in K-fold cross-validation (R2 = 0.95). However, the prediction R2 of RF, DTR, SVR, KNN, ETR, BG, MLP, LR and XGB on test set predictions was recorded as 0.71, 0.73, 0.41, 0.69, 0.71, 0.77, 0.54, 0.31 and 0.76, respectively. Therefore, to evaluate these mixed performance trends, Friedman and post-hoc Conover tests were employed to categorize the models into best, good, moderate and poor performing groups at a 0.05 significance level. By examining relative performance of diverse ML models, this work offers insights pertinence of different ML algorithms for CS prediction of FAG concrete. Lastly, multi objective pareto fronts have been presented, further guiding towards sustainable and cost-effective design of FAG concrete. By and large, the study aims at promoting circularity in construction sector by guiding ML assisted development and optimisation FAG concrete for diverse applications.

Graphical Abstract