Scholarlink

<p>Expansive soils pose severe problems to the pavement performance due to the presence of high swelling-shrinkage characteristics, low strength, and sensitivity to moisture. The paper assesses Rice husk ash (RHA) as a sustainable stabilizer of expansive subgrade soil by undertaking laboratory tests and machine learning. The experimental findings indicated that RHA decreased the differential free swell and plasticity and enhanced the unconfined compressive strength (UCS) and soaked California Bearing Ratio (CBR) and performed best at 10–15% RHA. The improved subgrade properties enabled a reduction of flexible pavement thickness by 26% as per IRC 37:2018. A dataset of 186 experimentally generated soil–stabilizer conditions was used to develop predictive models for UCS. XGBoost was found to be the best performing model with high predictive accuracy on both independent testing data (R<sup>2</sup> = 0.9895) and five-fold cross-validation (R<sup>2</sup> = 0.9924), and strong generalization and low risk of overfitting. The explainability analysis based on SHAP revealed the most significant factors affecting the development of strengths were curing period, intrinsic soil properties, RHA dosage, and swelling potential. Uncertainty quantification also showed that the prediction intervals were in line with the variability in the experiment. The results verify that RHA is a good eco-friendly stabilizer and that interpretable machine learning can assist in the reliable performance-based 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

Performance-based evaluation and UCS prediction of rice husk ash-stabilized expansive subgrade soil using experiments and explainable machine learning

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月17日

查看全文
  • Abhishek Sharma,
  • Kanwarpreet Singh,
  • Amenjor Senagah,
  • Neelam Sidhu,
  • Aditya Kumar Tiwary,
  • Gaurav Juneja,
  • Shubham Kumar Verma,
  • Rupesh Gupta,
  • Pardeep Singh Joia

摘要

Expansive soils pose severe problems to the pavement performance due to the presence of high swelling-shrinkage characteristics, low strength, and sensitivity to moisture. The paper assesses Rice husk ash (RHA) as a sustainable stabilizer of expansive subgrade soil by undertaking laboratory tests and machine learning. The experimental findings indicated that RHA decreased the differential free swell and plasticity and enhanced the unconfined compressive strength (UCS) and soaked California Bearing Ratio (CBR) and performed best at 10–15% RHA. The improved subgrade properties enabled a reduction of flexible pavement thickness by 26% as per IRC 37:2018. A dataset of 186 experimentally generated soil–stabilizer conditions was used to develop predictive models for UCS. XGBoost was found to be the best performing model with high predictive accuracy on both independent testing data (R2 = 0.9895) and five-fold cross-validation (R2 = 0.9924), and strong generalization and low risk of overfitting. The explainability analysis based on SHAP revealed the most significant factors affecting the development of strengths were curing period, intrinsic soil properties, RHA dosage, and swelling potential. Uncertainty quantification also showed that the prediction intervals were in line with the variability in the experiment. The results verify that RHA is a good eco-friendly stabilizer and that interpretable machine learning can assist in the reliable performance-based pavement design.