Scholarlink

<p>This study investigates the stabilization of black cotton soil (BCS) through a combination of experimental investigations and machine learning (ML) techniques, using unconfined compressive strength (UCS) as the principal performance indicator. Five sustainable stabilizing additives namely, biochar, rice husk ash, glass powder, fly ash, and wood ash, were incorporated to enhance the UCS of BCS. A total of 90 laboratory-generated datasets were developed using seven input parameters namely, additive percentage, BCS percentage, specific gravity, liquid limit, plastic limit, maximum dry density, and optimum moisture content. Three ML models, namely deep neural network (DNN), support vector regression (SVR), and extreme gradient boosting (XGBoost), were employed to predict UCS. Model performance was assessed using several statistical indicators, including the coefficient of determination (R²), a20-index, explained variance score, root mean square error (RMSE), mean absolute error, and index of scattering. Among the developed models, DNN exhibited superior predictive capability, achieving the highest R² values (0.990 for training and 0.925 for testing) and the lowest RMSE values (0.002 for training and 0.064 for testing), followed by SVR and XGBoost. Additional validation techniques, including rank analysis, reliability index, regression analysis, objective function criteria, curve fitting, and external validation, further confirmed the robustness of the models.</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

Experimental and Machine Learning Approaches for Predicting the Unconfined Compressive Strength of Stabilized Black Cotton Soil

  • 期刊论文

  • 发表日期：

    2026年6月26日

查看全文
  • Rashid Mustafa,
  • Md Nisar Ahmed

摘要

This study investigates the stabilization of black cotton soil (BCS) through a combination of experimental investigations and machine learning (ML) techniques, using unconfined compressive strength (UCS) as the principal performance indicator. Five sustainable stabilizing additives namely, biochar, rice husk ash, glass powder, fly ash, and wood ash, were incorporated to enhance the UCS of BCS. A total of 90 laboratory-generated datasets were developed using seven input parameters namely, additive percentage, BCS percentage, specific gravity, liquid limit, plastic limit, maximum dry density, and optimum moisture content. Three ML models, namely deep neural network (DNN), support vector regression (SVR), and extreme gradient boosting (XGBoost), were employed to predict UCS. Model performance was assessed using several statistical indicators, including the coefficient of determination (R²), a20-index, explained variance score, root mean square error (RMSE), mean absolute error, and index of scattering. Among the developed models, DNN exhibited superior predictive capability, achieving the highest R² values (0.990 for training and 0.925 for testing) and the lowest RMSE values (0.002 for training and 0.064 for testing), followed by SVR and XGBoost. Additional validation techniques, including rank analysis, reliability index, regression analysis, objective function criteria, curve fitting, and external validation, further confirmed the robustness of the models.