Scholarlink

<p>This study investigated machine learning models for predicting soil compaction parameters. Traditional methods were labor-intensive and were influenced by soil type, plasticity, and compaction energy. Using 265 datasets, machine learning models—including random forest, gradient boosting (G-BR), support vector, and multilinear regression—were trained, with compaction energy incorporated as an input variable for the first time. Results showed that G-BR outperformed other models in predicting maximum dry density (MaxDD) and Optimum moisture content (OptMC), achieving coefficient of determination of 0.98 and 0.94, respectively. Performance metrics included a mean squared error of 0.32 and 2.28, mean absolute error of 0.31 and 1.11, root mean square error of 0.56 and 1.5, a20-index of 77.36 and 98.1, index of scatter of 0.07 and 0.26, and index of agreement of 1 and 0.98, respectively. Taylor’s diagram and lab experiments validated the robustness of the models. Sensitivity analysis revealed that fine content (Fc), plastic limit (PL), and liquid limit were the most influential factors for OptMC, while sand content, Fc, and PL were key for MaxDD, listed in descending order of importance. Curve-fitting and rank analysis confirmed that G-BR provided superior accuracy compared to traditional methods, offering a reliable tool for soil compaction assessment in construction.</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

Model Formation and Experimental Validation of Ensemble-Artificial Intelligence Approach for Estimating Soil Compaction Parameters

  • 期刊论文

  • 发表日期：

    2025年8月11日

查看全文
  • Yonas Tilahun,
  • Qinghua Xiao,
  • Argaw Asha Ashango,
  • Xiangyu Han,
  • Yuan Hao

摘要

This study investigated machine learning models for predicting soil compaction parameters. Traditional methods were labor-intensive and were influenced by soil type, plasticity, and compaction energy. Using 265 datasets, machine learning models—including random forest, gradient boosting (G-BR), support vector, and multilinear regression—were trained, with compaction energy incorporated as an input variable for the first time. Results showed that G-BR outperformed other models in predicting maximum dry density (MaxDD) and Optimum moisture content (OptMC), achieving coefficient of determination of 0.98 and 0.94, respectively. Performance metrics included a mean squared error of 0.32 and 2.28, mean absolute error of 0.31 and 1.11, root mean square error of 0.56 and 1.5, a20-index of 77.36 and 98.1, index of scatter of 0.07 and 0.26, and index of agreement of 1 and 0.98, respectively. Taylor’s diagram and lab experiments validated the robustness of the models. Sensitivity analysis revealed that fine content (Fc), plastic limit (PL), and liquid limit were the most influential factors for OptMC, while sand content, Fc, and PL were key for MaxDD, listed in descending order of importance. Curve-fitting and rank analysis confirmed that G-BR provided superior accuracy compared to traditional methods, offering a reliable tool for soil compaction assessment in construction.