Scholarlink

<p>Accurate prediction of the unconfined compressive strength (UCS) of stabilized expansive soils is vital for reliable geotechnical design. This study proposes a Slime Mould Algorithm (SMA), an optimized machine learning framework to predict the UCS of expansive soils treated with hydrated-lime-activated rice husk ash (HARHA). A database of 121 laboratory-tested samples was developed using seven input parameters: HARHA content, liquid limit, plastic limit, plasticity index, optimum moisture content, clay activity, and maximum dry density. Several machine learning models were evaluated in both conventional and SMA-optimized forms using R<sup>2</sup>, RMSE, MAE, and MSE. SMA optimization significantly improved model performance, increasing R<sup>2</sup> by 3–7% and reducing RMSE by up to 25%. The optimized models achieved R<sup>2</sup> values above 0.98, indicating excellent predictive accuracy. Compared with existing empirical and standalone models, the proposed framework showed substantial performance gains, up to 83% over SVR, 53% over ANN, and 15% over RF. Sensitivity analysis revealed that HARHA content and maximum dry density positively influence UCS, while clay activity and plasticity-related parameters have negative effects. Overall, the proposed approach offers a robust and sustainable alternative to conventional empirical methods, reducing reliance on extensive laboratory testing.</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-based prediction of the unconfined compressive strength of expansive soils stabilized with hydrated lime-activated rice husk ash

  • 期刊论文

  • 发表日期：

    2026年4月22日

查看全文
  • Bayram Ateş,
  • Mohammad Azim Eirgash,
  • Yang Zhao

摘要

Accurate prediction of the unconfined compressive strength (UCS) of stabilized expansive soils is vital for reliable geotechnical design. This study proposes a Slime Mould Algorithm (SMA), an optimized machine learning framework to predict the UCS of expansive soils treated with hydrated-lime-activated rice husk ash (HARHA). A database of 121 laboratory-tested samples was developed using seven input parameters: HARHA content, liquid limit, plastic limit, plasticity index, optimum moisture content, clay activity, and maximum dry density. Several machine learning models were evaluated in both conventional and SMA-optimized forms using R2, RMSE, MAE, and MSE. SMA optimization significantly improved model performance, increasing R2 by 3–7% and reducing RMSE by up to 25%. The optimized models achieved R2 values above 0.98, indicating excellent predictive accuracy. Compared with existing empirical and standalone models, the proposed framework showed substantial performance gains, up to 83% over SVR, 53% over ANN, and 15% over RF. Sensitivity analysis revealed that HARHA content and maximum dry density positively influence UCS, while clay activity and plasticity-related parameters have negative effects. Overall, the proposed approach offers a robust and sustainable alternative to conventional empirical methods, reducing reliance on extensive laboratory testing.