Scholarlink

<p>This study utilizes a hybrid ensemble machine learning (ML) model to predict the compressive strength (CS) of metakaolin-based geopolymers. Nine hybrid models, such as Random Forest (RF), AdaBoost (ADB), and Extreme Gradient Boosting (XGB) optimized by bio-inspired metaheuristic algorithms such as Salp Swarm Optimizer (SSO), Aquila Optimizer (AOA), and Walrus Optimization Algorithm (WOA) were adopted. It utilized a dataset of 235 samples, which comprised significant mix variables, and the model was validated with a five-fold split-validation. The XGB-WOA hybrid ML model demonstrated better performance among the studied algorithms, with a coefficient of determination (<i>R</i><sup>2</sup> = 0.976) and Root Mean Squared Error (RMSE) of 2.153&#xa0;MPa on the test set. The analysis of SHAP values identified the alkali/binder ratio and curing temperature as the most significant input parameters among several others. Model robustness was confirmed by composite ranking and the Taylor diagram. Additionally, the GUI was created in Python to enable users to make practical predictions of CS based on their input. This research provides a scalable, interpretable, and highly accurate framework for optimization of the mix design of metakaolin-based geopolymer, promoting the adoption of low-carbon binder systems in sustainable construction.</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

Explainable machine learning integrated with bio-inspired optimization for predicting and interpreting the compressive strength of metakaolin-based geopolymers

  • 期刊论文

  • 发表日期：

    2026年2月24日

查看全文
  • Satish Paudel,
  • Ajad Shrestha,
  • Sanjog Chhetri Sapkota,
  • Hakas Prayuda

摘要

This study utilizes a hybrid ensemble machine learning (ML) model to predict the compressive strength (CS) of metakaolin-based geopolymers. Nine hybrid models, such as Random Forest (RF), AdaBoost (ADB), and Extreme Gradient Boosting (XGB) optimized by bio-inspired metaheuristic algorithms such as Salp Swarm Optimizer (SSO), Aquila Optimizer (AOA), and Walrus Optimization Algorithm (WOA) were adopted. It utilized a dataset of 235 samples, which comprised significant mix variables, and the model was validated with a five-fold split-validation. The XGB-WOA hybrid ML model demonstrated better performance among the studied algorithms, with a coefficient of determination (R2 = 0.976) and Root Mean Squared Error (RMSE) of 2.153 MPa on the test set. The analysis of SHAP values identified the alkali/binder ratio and curing temperature as the most significant input parameters among several others. Model robustness was confirmed by composite ranking and the Taylor diagram. Additionally, the GUI was created in Python to enable users to make practical predictions of CS based on their input. This research provides a scalable, interpretable, and highly accurate framework for optimization of the mix design of metakaolin-based geopolymer, promoting the adoption of low-carbon binder systems in sustainable construction.