Scholarlink

<p>Accurate prediction of unconfined compressive strength (UCS) of expansive soils is crucial in geotechnical engineering for evaluating soil stability and structural integrity. This study explores data-driven ensemble learning methods, including random forest, extreme gradient boosting (XGB), adaptive boosting regressor, and gradient boosting regressor, to develop reliable UCS prediction models. Bayesian hyperparameter optimization is utilized to fine-tune model parameters, improving predictive performance. Moreover, Shapley additive explanation analysis is applied for the first time to enhance the interpretability of the ensemble learning model and provide deeper insights into UCS prediction. The results indicate that maximum dry density, optimum moisture content, and sand are the most significant predictors of UCS, while liquid limit and plasticity index contribute minimally. Among the models tested, XGB demonstrates the highest accuracy and generalizability, making it the most effective for UCS estimation. This study highlights the benefits of combining ensemble learning with explainable artificial intelligence techniques to enhance predictive accuracy and interpretability in geotechnical applications.</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

Interpretable Ensemble Learning Approaches for Predicting Unconfined Compressive Strength of Expansive Soils

  • 期刊论文

  • 发表日期：

    2025年5月29日

查看全文
  • Thanh-Hai Do,
  • Minh-Vuong Pham,
  • Ngoc-Thi Huynh,
  • Thi-Anh-Thu Phan,
  • Quoc-Thang Dinh,
  • Hoai-Long Le,
  • Thanh-Nhan Nguyen,
  • Ho-Hong-Duy Nguyen

摘要

Accurate prediction of unconfined compressive strength (UCS) of expansive soils is crucial in geotechnical engineering for evaluating soil stability and structural integrity. This study explores data-driven ensemble learning methods, including random forest, extreme gradient boosting (XGB), adaptive boosting regressor, and gradient boosting regressor, to develop reliable UCS prediction models. Bayesian hyperparameter optimization is utilized to fine-tune model parameters, improving predictive performance. Moreover, Shapley additive explanation analysis is applied for the first time to enhance the interpretability of the ensemble learning model and provide deeper insights into UCS prediction. The results indicate that maximum dry density, optimum moisture content, and sand are the most significant predictors of UCS, while liquid limit and plasticity index contribute minimally. Among the models tested, XGB demonstrates the highest accuracy and generalizability, making it the most effective for UCS estimation. This study highlights the benefits of combining ensemble learning with explainable artificial intelligence techniques to enhance predictive accuracy and interpretability in geotechnical applications.