Scholarlink

<p>The accurate estimation of unsaturated soil properties is highly crucial for vast geotechnical engineering projects especially for foundations design. Among different soil properties, shear strength and settlement characteristics are two imperative parameters. This study determines the effective stress parameter (χ) of unsaturated soil which is a key parameter to compute effective stress through advanced machine learning methods, as the laboratory task for finding c is complicated. In this, Genetic Programming (GP) and Multivariate Adaptive Regression Spline (MARS) machine learning models was adopted to forecast χ by adopting various unsaturated soil parameters (net confining stress, h<sub>b</sub>, q<sub>r</sub>, and l) and water parameters (qs &amp; suction). Among the adopted methods, the GP model executed better when compared with other MARS model with greater precision of R<sup>2</sup> = 0.964 and less RMSE and MAE errors. To further validate the potential of the model, along with the statistical indices, some of the visual representation like Taylor diagram, Cook’s D chart was developed. The comparative analysis was carried out which technically depicts both the models have provided better results, however GP model outperformed the MARS model. The sensitivity analysis also carried out to find which input parameter was more influential in estimating the c value. In addition, the GP &amp; MARS model provided the direct equation for estimating the value of χ.</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

Comparative analysis of machine learning models for predicting effective stress parameters in unsaturated soils

  • 期刊论文

  • 发表日期：

    2025年7月16日

查看全文
  • J. Jagan,
  • B. R. Vinod,
  • S. Gobinath,
  • Pijush Samui,
  • Gnana Jeba Das

摘要

The accurate estimation of unsaturated soil properties is highly crucial for vast geotechnical engineering projects especially for foundations design. Among different soil properties, shear strength and settlement characteristics are two imperative parameters. This study determines the effective stress parameter (χ) of unsaturated soil which is a key parameter to compute effective stress through advanced machine learning methods, as the laboratory task for finding c is complicated. In this, Genetic Programming (GP) and Multivariate Adaptive Regression Spline (MARS) machine learning models was adopted to forecast χ by adopting various unsaturated soil parameters (net confining stress, hb, qr, and l) and water parameters (qs & suction). Among the adopted methods, the GP model executed better when compared with other MARS model with greater precision of R2 = 0.964 and less RMSE and MAE errors. To further validate the potential of the model, along with the statistical indices, some of the visual representation like Taylor diagram, Cook’s D chart was developed. The comparative analysis was carried out which technically depicts both the models have provided better results, however GP model outperformed the MARS model. The sensitivity analysis also carried out to find which input parameter was more influential in estimating the c value. In addition, the GP & MARS model provided the direct equation for estimating the value of χ.