Scholarlink

<p>Soil behavior, especially settlement, is a critical factor in the analysis and design of engineering projects. If not properly predicted, settlement can lead to serious issues for geotechnical engineers. Field testing is considered the most reliable way for this purpose; though, it is the last resort in the view of engineers because of two reasons: cost and the specific conditions each test may have. Artificial intelligence, a helpful tool for resolving engineering problems during the past few decades, can be an alternative to field testing. Consequently, this study employed it to estimate the settlement potential of soil more efficiently. To this end, six ensemble learning models, i.e., Bagging Regressor, Extra Trees Regressor, Random Forest Regressor, AdaBoost, Gradient Boosting Regressor, and Stacking Regressor, were adopted and trained by a dataset of 786 records. First, various characteristics of soil were examined and four of them were finalized as the inputs of the model: Silt fraction, moisture content, applied stress, and dry density. The findings showed that ensemble learning is highly effective in understanding the intricate behavior of collapsible soils, with minimal data processing. In addition, it has fewer parameters (inputs), which can be easily derived from conventional testing. The findings of this study could have contribution to traditional geotechnical engineering practices in problematic soils. </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

Prediction of the Wetting-Induced Compression of Collapsible Soils Using Ensemble Machine Learning

  • 期刊论文

  • 发表日期：

    2025年7月17日

查看全文
  • Ali Maroof,
  • Nader Karballaeezadeh,
  • Danial Mohammadzadeh S.,
  • Danial Jahed Armaghani,
  • Hamidreza Mirzaei

摘要

Soil behavior, especially settlement, is a critical factor in the analysis and design of engineering projects. If not properly predicted, settlement can lead to serious issues for geotechnical engineers. Field testing is considered the most reliable way for this purpose; though, it is the last resort in the view of engineers because of two reasons: cost and the specific conditions each test may have. Artificial intelligence, a helpful tool for resolving engineering problems during the past few decades, can be an alternative to field testing. Consequently, this study employed it to estimate the settlement potential of soil more efficiently. To this end, six ensemble learning models, i.e., Bagging Regressor, Extra Trees Regressor, Random Forest Regressor, AdaBoost, Gradient Boosting Regressor, and Stacking Regressor, were adopted and trained by a dataset of 786 records. First, various characteristics of soil were examined and four of them were finalized as the inputs of the model: Silt fraction, moisture content, applied stress, and dry density. The findings showed that ensemble learning is highly effective in understanding the intricate behavior of collapsible soils, with minimal data processing. In addition, it has fewer parameters (inputs), which can be easily derived from conventional testing. The findings of this study could have contribution to traditional geotechnical engineering practices in problematic soils.