Scholarlink

<p>This study explores the application of advanced machine learning techniques such as multivariate adaptive regression splines (MARS), artificial neural networks (ANNs), random forest (RF), and extreme gradient boosting (XGBoost) to develop robust predictive models for embedment depth estimation. A synthetic dataset of 500 samples, which includes the parameters unit weight of clay, unit weight of sand, cohesion, and friction angle of sand, was generated by integrating the coefficient of variation to represent soil heterogeneity. A performance evaluation using several performance indicators revealed that XGBoost achieved the highest accuracy during training (<i>R</i><sup>2</sup> = 0.994, RMSE = 0.017), whereas MARS demonstrated superior generalizability during testing (<i>R</i><sup>2</sup> = 0.991, RMSE = 0.022). Although the ANN delivered consistent predictions, the RF suffered from overfitting, limiting its effectiveness in the test phase. A reliability assessment using the first-order reliability method revealed that the RF had the highest reliability index (3.4423), albeit with reduced generalizability. Sensitivity analysis identified the unit weight and friction angle as the most dominant factors affecting the embedment depth. On the basis of the obtained results, it can be concluded that the proposed data-driven models, especially XGBoost and MARS, are reliable tools for enhancing the prediction and design of cantilever sheet piles.</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

Advanced Machine Learning Approaches for Predicting the Embedment Depth of Cantilever Sheet Piles

  • 期刊论文

  • 发表日期：

    2025年9月15日

查看全文
  • Chaiyathawat Boonyong,
  • Warit Wipulanusat,
  • Suraparb Keawsawasvong

摘要

This study explores the application of advanced machine learning techniques such as multivariate adaptive regression splines (MARS), artificial neural networks (ANNs), random forest (RF), and extreme gradient boosting (XGBoost) to develop robust predictive models for embedment depth estimation. A synthetic dataset of 500 samples, which includes the parameters unit weight of clay, unit weight of sand, cohesion, and friction angle of sand, was generated by integrating the coefficient of variation to represent soil heterogeneity. A performance evaluation using several performance indicators revealed that XGBoost achieved the highest accuracy during training (R2 = 0.994, RMSE = 0.017), whereas MARS demonstrated superior generalizability during testing (R2 = 0.991, RMSE = 0.022). Although the ANN delivered consistent predictions, the RF suffered from overfitting, limiting its effectiveness in the test phase. A reliability assessment using the first-order reliability method revealed that the RF had the highest reliability index (3.4423), albeit with reduced generalizability. Sensitivity analysis identified the unit weight and friction angle as the most dominant factors affecting the embedment depth. On the basis of the obtained results, it can be concluded that the proposed data-driven models, especially XGBoost and MARS, are reliable tools for enhancing the prediction and design of cantilever sheet piles.