Scholarlink

<p>The stabilization of weak subgrade soils using lime is widely practiced in geotechnical engineering; however, accurate prediction of strength parameters such as the California bearing ratio (CBR) remains challenging due to complex soil–lime interactions. This study develops a hybrid experimental–machine learning framework for predicting the CBR of hydrated lime-treated low-plasticity clay (CL) soils by integrating laboratory investigation with explainable artificial intelligence (XAI). Laboratory results showed that CBR increased from 4.2% in untreated soil to 21.5% at 10% lime content, confirming strength enhancement through cation exchange, flocculation, and pozzolanic reactions. To improve model robustness, the dataset was expanded to 300 samples using controlled Gaussian-based statistical sampling while preserving statistical characteristics. Ten input variables were used to train seven machine learning models, including MLR, SVR, ANN, DTR, RFR, XGBoost, and CatBoost. Among these, CatBoost achieved the highest predictive accuracy (R<sup>2</sup> = 0.98, RMSE = 1.08, MAE = 0.87). SHAP analysis and partial dependence plots identified lime content, maximum dry density, and curing period as dominant factors governing CBR. The proposed framework offers a reliable and interpretable tool for subgrade strength prediction, reducing reliance on extensive laboratory testing and supporting data-driven pavement design.</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

Hybrid experimental–explainable machine learning framework for predicting subgrade strength of lime-treated clay soils

  • 期刊论文

  • 发表日期：

    2026年5月22日

查看全文
  • Prashant Tiwari,
  • Smita Tung

摘要

The stabilization of weak subgrade soils using lime is widely practiced in geotechnical engineering; however, accurate prediction of strength parameters such as the California bearing ratio (CBR) remains challenging due to complex soil–lime interactions. This study develops a hybrid experimental–machine learning framework for predicting the CBR of hydrated lime-treated low-plasticity clay (CL) soils by integrating laboratory investigation with explainable artificial intelligence (XAI). Laboratory results showed that CBR increased from 4.2% in untreated soil to 21.5% at 10% lime content, confirming strength enhancement through cation exchange, flocculation, and pozzolanic reactions. To improve model robustness, the dataset was expanded to 300 samples using controlled Gaussian-based statistical sampling while preserving statistical characteristics. Ten input variables were used to train seven machine learning models, including MLR, SVR, ANN, DTR, RFR, XGBoost, and CatBoost. Among these, CatBoost achieved the highest predictive accuracy (R2 = 0.98, RMSE = 1.08, MAE = 0.87). SHAP analysis and partial dependence plots identified lime content, maximum dry density, and curing period as dominant factors governing CBR. The proposed framework offers a reliable and interpretable tool for subgrade strength prediction, reducing reliance on extensive laboratory testing and supporting data-driven pavement design.