Scholarlink

<p>This research aimed to predict the unconfined compressive strength (UCS) of stabilized soil using hybrid models. Three hybrid models, namely K-Nearest Neighbors_Simulated Annealing (KNN_SA), Random Forest_Simulated Annealing (RF_SA), and Gradient Boosting_Simulated Annealing (GB_SA), and three state-of-the-art machine learning algorithms, including Extreme Gradient Boosting, Adaptive Boosting, and Support Vector Machine, were employed. The input soil parameters for modeling included organic content, sand content, silt content, gravel content, Atterberg limits, lime content, and cement content. The feature importance and sensitivity analysis were conducted to evaluate the effect of each input variable on the UCS of stabilized soil. The results show that all proposed hybrid models performed well with high prediction accuracy. The hybrid model of GB_SA achieved the highest accuracy with a coefficient of determination (R<sup>2</sup>) = 0.9520, Root Mean Square Error (RMSE) = 0.2052, and Mean Absolute Error (MAE) = 0.1564. The results of feature importance and sensitivity analysis indicate that cement content, lime content, and organic content of soil are the important variables, which strongly affect the UCS of stabilized soil. Among these three factors, cement content is the most dominant and crucial factor affecting as well as controlling the UCS of stabilized soil.</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

Estimation of Unconfined Compressive Strength of Stabilized Soils Containing Organic Matter Using Optimized Machine Learning Models

  • 期刊论文

  • 发表日期：

    2026年5月27日

查看全文
  • Van Quan Tran,
  • Viet Quoc Dang,
  • Ngoc Son Be,
  • Lanh Si Ho

摘要

This research aimed to predict the unconfined compressive strength (UCS) of stabilized soil using hybrid models. Three hybrid models, namely K-Nearest Neighbors_Simulated Annealing (KNN_SA), Random Forest_Simulated Annealing (RF_SA), and Gradient Boosting_Simulated Annealing (GB_SA), and three state-of-the-art machine learning algorithms, including Extreme Gradient Boosting, Adaptive Boosting, and Support Vector Machine, were employed. The input soil parameters for modeling included organic content, sand content, silt content, gravel content, Atterberg limits, lime content, and cement content. The feature importance and sensitivity analysis were conducted to evaluate the effect of each input variable on the UCS of stabilized soil. The results show that all proposed hybrid models performed well with high prediction accuracy. The hybrid model of GB_SA achieved the highest accuracy with a coefficient of determination (R2) = 0.9520, Root Mean Square Error (RMSE) = 0.2052, and Mean Absolute Error (MAE) = 0.1564. The results of feature importance and sensitivity analysis indicate that cement content, lime content, and organic content of soil are the important variables, which strongly affect the UCS of stabilized soil. Among these three factors, cement content is the most dominant and crucial factor affecting as well as controlling the UCS of stabilized soil.