Scholarlink

<p>Accurately predicting the unconfined compressive strength (UCS) of cement-stabilized soils is essential for designing safe and long-lasting geotechnical structures. Traditional empirical models often fall short in capturing the nonlinear interactions among geotechnical variables, cement content, and curing duration. This study introduces an explainable artificial intelligence (AI) framework for UCS prediction, utilizing seven advanced machine learning (ML) models: artificial neural network (ANN), support vector regression (SVR), decision tree regression (DTR), random forest regression (RFR), gradient boosting machine (GBM), extreme gradient boosting (XGBoost), and CatBoost. A dataset of 500 samples with eight key input features—cement content, curing time, liquid limit, plasticity index, maximum dry density, optimum moisture content, fines content, and specific gravity—was used for model development. All models were evaluated using 10-fold cross-validation and multiple performance metrics, including R², RMSE, IOA, a20 accuracy, and prediction intervals. XGBoost achieved the best performance (R² = 0.923, RMSE = 0.269&#xa0;MPa, IOA = 0.961, a20 = 94.8%). SHAP-based interpretability and OAT sensitivity analysis identified cement content and curing time as the most influential features. A field-representative case study and SHAP force plots further validated model robustness. The proposed framework offers accurate, transparent, and field-ready UCS prediction, supporting data-driven soil stabilization and geotechnical decision-making.</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 framework for predicting unconfined compressive strength of cement stabilized soils using SHAP and field validation

  • 期刊论文

  • 发表日期：

    2026年5月5日

查看全文
  • Debashish Chandra,
  • Jasvir Singh

摘要

Accurately predicting the unconfined compressive strength (UCS) of cement-stabilized soils is essential for designing safe and long-lasting geotechnical structures. Traditional empirical models often fall short in capturing the nonlinear interactions among geotechnical variables, cement content, and curing duration. This study introduces an explainable artificial intelligence (AI) framework for UCS prediction, utilizing seven advanced machine learning (ML) models: artificial neural network (ANN), support vector regression (SVR), decision tree regression (DTR), random forest regression (RFR), gradient boosting machine (GBM), extreme gradient boosting (XGBoost), and CatBoost. A dataset of 500 samples with eight key input features—cement content, curing time, liquid limit, plasticity index, maximum dry density, optimum moisture content, fines content, and specific gravity—was used for model development. All models were evaluated using 10-fold cross-validation and multiple performance metrics, including R², RMSE, IOA, a20 accuracy, and prediction intervals. XGBoost achieved the best performance (R² = 0.923, RMSE = 0.269 MPa, IOA = 0.961, a20 = 94.8%). SHAP-based interpretability and OAT sensitivity analysis identified cement content and curing time as the most influential features. A field-representative case study and SHAP force plots further validated model robustness. The proposed framework offers accurate, transparent, and field-ready UCS prediction, supporting data-driven soil stabilization and geotechnical decision-making.