Scholarlink

<p>This study investigates the bio-reinforcement effects of <i>Nandina domestica</i> roots on clayey silt and evaluates statistical versus machine learning predictive models. Laboratory unconfined compression tests were conducted on soil composites with varying root diameters (1–3&#xa0;mm), root weight densities (1–4&#xa0;g/100&#xa0;cm³), and moisture contents (20–35%). Multiple Linear Regression (MLR) was applied to predict ultimate UCS (peak stress), while Random Forest (RF) and LightGBM models were developed to predict the complete nonlinear stress–strain trajectories. Experimental results confirmed that roots consistently improved soil strength, with finer roots (≈ 1&#xa0;mm) providing the most substantial reinforcement, increasing UCS by an average of 62%. Regarding predictive performance, MLR provided a reliable baseline for peak strength (R<sup>2</sup> = 0.85). However, for stress–strain trajectories, both machine learning models demonstrated superior accuracy, with LightGBM achieving the highest performance (R<sup>2</sup> = 0.948, RMSE = 4.52&#xa0;kPa) compared to RF (R<sup>2</sup> = 0.939). Feature importance analysis revealed that axial strain and moisture content were the dominant predictors, collectively explaining over 80% of the variation in axial stress, while root density and diameter served as significant secondary contributors. These findings establish LightGBM as a robust tool for simulating the complex mechanical behavior of root–soil composites.</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

Machine Learning Prediction of Stress–strain Trajectories and Unconfined Compressive Strength in Nandina Domestica Root–reinforced Clayey Silt

  • 期刊论文

  • 发表日期：

    2026年1月10日

查看全文
  • Masoud Ebrahimi Derakhshan,
  • Mehrab Ramzani,
  • Jaber Mamaghanian,
  • Hamid Reza Razeghi

摘要

This study investigates the bio-reinforcement effects of Nandina domestica roots on clayey silt and evaluates statistical versus machine learning predictive models. Laboratory unconfined compression tests were conducted on soil composites with varying root diameters (1–3 mm), root weight densities (1–4 g/100 cm³), and moisture contents (20–35%). Multiple Linear Regression (MLR) was applied to predict ultimate UCS (peak stress), while Random Forest (RF) and LightGBM models were developed to predict the complete nonlinear stress–strain trajectories. Experimental results confirmed that roots consistently improved soil strength, with finer roots (≈ 1 mm) providing the most substantial reinforcement, increasing UCS by an average of 62%. Regarding predictive performance, MLR provided a reliable baseline for peak strength (R2 = 0.85). However, for stress–strain trajectories, both machine learning models demonstrated superior accuracy, with LightGBM achieving the highest performance (R2 = 0.948, RMSE = 4.52 kPa) compared to RF (R2 = 0.939). Feature importance analysis revealed that axial strain and moisture content were the dominant predictors, collectively explaining over 80% of the variation in axial stress, while root density and diameter served as significant secondary contributors. These findings establish LightGBM as a robust tool for simulating the complex mechanical behavior of root–soil composites.