Scholarlink

<p>This study applied machine learning methods to predict the durability performance (specifically shrinkage and freeze-thaw resistance) of solid waste-activated cementitious materials. It also offered insights for optimizing material formulations through feature impact analysis. The study collected a total of 130 sets of shrinkage data and 106 sets of freeze-thaw data, establishing various models, including BP, GA-BP, SVM, RF, RBF, and LSTM. The results revealed that the SVM model performed the best on the test dataset. It achieved an <i>R</i><sup>2</sup> of 0.935 8 for shrinkage prediction, with MAE and RMSE values of 0.464 4 and 0.625 4, respectively. Regarding freeze-thaw quality loss prediction, the <i>R</i><sup>2</sup> was 0.917 8, with MAE and RMSE values of 0.313 9 and 0.532 8, respectively. The study analyzed the impact of different features on the outcomes using the SHAP method, highlighting that the alkaline activator dosage, Al<sub>2</sub>O<sub>3</sub>, SiO<sub>2</sub>, and water glass modulus were critical factors influencing shrinkage, while CaO, water-cement ratio, water, and Al<sub>2</sub>O<sub>3</sub> were crucial for freeze-thaw resistance. By investigating feature interactions through single-factor and two-factor analysis, the study proposed recommendations for optimizing material formulations. This research validated the efficacy of machine learning in predicting the durability of solid waste cementitious materials and offered insights for material optimization through feature impact analysis, thereby laying the groundwork for the development of related materials.</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 Prediction and Feature Impact Analysis of Durability Performance of Solid Waste-Alkali Activated Cementitious Materials

  • 期刊论文

  • 发表日期：

    2025年10月29日

查看全文
  • Wei Wei,
  • Yongjie Ding,
  • Yongxiang Zhou,
  • Jiaojiao Wang,
  • Yanghui Wang

摘要

This study applied machine learning methods to predict the durability performance (specifically shrinkage and freeze-thaw resistance) of solid waste-activated cementitious materials. It also offered insights for optimizing material formulations through feature impact analysis. The study collected a total of 130 sets of shrinkage data and 106 sets of freeze-thaw data, establishing various models, including BP, GA-BP, SVM, RF, RBF, and LSTM. The results revealed that the SVM model performed the best on the test dataset. It achieved an R2 of 0.935 8 for shrinkage prediction, with MAE and RMSE values of 0.464 4 and 0.625 4, respectively. Regarding freeze-thaw quality loss prediction, the R2 was 0.917 8, with MAE and RMSE values of 0.313 9 and 0.532 8, respectively. The study analyzed the impact of different features on the outcomes using the SHAP method, highlighting that the alkaline activator dosage, Al2O3, SiO2, and water glass modulus were critical factors influencing shrinkage, while CaO, water-cement ratio, water, and Al2O3 were crucial for freeze-thaw resistance. By investigating feature interactions through single-factor and two-factor analysis, the study proposed recommendations for optimizing material formulations. This research validated the efficacy of machine learning in predicting the durability of solid waste cementitious materials and offered insights for material optimization through feature impact analysis, thereby laying the groundwork for the development of related materials.