Scholarlink

<p>Sustainable stabilization of problematic clayey soils with industrial by-products has recently become a topic of considerable interest because of the environmental and economic issues relating to traditional binders. This paper examines the interaction of fly ash (FA) and waste paper sludge (WPS) in enhancing the unconfined compressive strength (UCS) of soil that contains Clay and trains relevant machine learning (ML) models to precisely forecast the strength of clayey soil. An overall experimental data set of 265 samples was obtained by changing the contents of FA and WPS (0–20%) under conditions of 3–90 days of curing. Three ensemble ML models (XGBoost, Random Forest and Stacking Regressor) were created to measure the nonlinear relationships between soil properties, compaction parameters and curing effects. Results indicate that the predictive performance of the models is really great with the R<sup>2</sup> of 0.9958 (XGBoost), 0.9932 (Random Forest), and 0.9960 (Stacking); the Stacking model had the lowest error (RMSE = 4.0145&#xa0;kPa, MAE = 3.2904&#xa0;kPa). The residual analysis operated to reveal that there was little bias and controlled variance, which indicated a high level of generalization when used in a laboratory environment. In order to make the models easier to understand, SHAP analysis was used, which showed that the determining factors that control the strength development are the curing period and the aspects related to binder, whereas WPS predicts the physical modification, but with a comparatively smaller power. Sensitivity analysis and monotonicity analysis also supported the idea that strength behavior is controlled by coherent chemical and mechanical processes with curing-driven pozzolanic reactions having a significant role. On the whole, the paper provides a solid and explainable ML model of predicting UCS and proves the fact that the FA-WPS stabilization can be implemented as an effective alternative to geotechnical applications.</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

Prediction of unconfined compressive strength of fly ash and waste paper sludge stabilized clayey soil using explainable machine learning

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月6日

查看全文
  • Vandana Aggarwal,
  • Bharpur Singh Abhi,
  • Seema Gupta,
  • Rippandeep Kaur,
  • Ranjeeta Saini,
  • Priya Sharma,
  • Prince Sharma,
  • Amenjor Senagah,
  • Ankush Mehta,
  • Rupesh Gupta

摘要

Sustainable stabilization of problematic clayey soils with industrial by-products has recently become a topic of considerable interest because of the environmental and economic issues relating to traditional binders. This paper examines the interaction of fly ash (FA) and waste paper sludge (WPS) in enhancing the unconfined compressive strength (UCS) of soil that contains Clay and trains relevant machine learning (ML) models to precisely forecast the strength of clayey soil. An overall experimental data set of 265 samples was obtained by changing the contents of FA and WPS (0–20%) under conditions of 3–90 days of curing. Three ensemble ML models (XGBoost, Random Forest and Stacking Regressor) were created to measure the nonlinear relationships between soil properties, compaction parameters and curing effects. Results indicate that the predictive performance of the models is really great with the R2 of 0.9958 (XGBoost), 0.9932 (Random Forest), and 0.9960 (Stacking); the Stacking model had the lowest error (RMSE = 4.0145 kPa, MAE = 3.2904 kPa). The residual analysis operated to reveal that there was little bias and controlled variance, which indicated a high level of generalization when used in a laboratory environment. In order to make the models easier to understand, SHAP analysis was used, which showed that the determining factors that control the strength development are the curing period and the aspects related to binder, whereas WPS predicts the physical modification, but with a comparatively smaller power. Sensitivity analysis and monotonicity analysis also supported the idea that strength behavior is controlled by coherent chemical and mechanical processes with curing-driven pozzolanic reactions having a significant role. On the whole, the paper provides a solid and explainable ML model of predicting UCS and proves the fact that the FA-WPS stabilization can be implemented as an effective alternative to geotechnical applications.