Scholarlink

<p>This study presents a comprehensive evaluation of advanced machine learning models for predicting the end-bearing capacity of rock-socketed piles. Eight algorithms were examined, including multivariate expression programming, artificial neural networks, decision trees, random forests, extreme gradient boosting, convolutional neural networks, long short-term memory networks, and categorical boosting. Five key parameters, uniaxial compressive strength of rock, geological strength index, pile diameter, pile length in soil, and pile length in rock, were used as input variables. Among the models, categorical boosting and extreme gradient boosting achieved the highest accuracy, with coefficients of determination of 0.979 and 0.978, respectively. A simplified mathematical equation was also developed using the multivariate expression programming model to provide a practical and interpretable prediction tool. Feature importance and partial dependence analyses revealed that rock strength and geological strength index are the dominant factors influencing end-bearing capacity, while pile geometry has a secondary effect. The results indicate that machine learning models outperform traditional empirical equations, offering significantly improved prediction accuracy and generalization. This research highlights the potential of data-driven approaches for more reliable design of pile foundations and provides a foundation for future studies incorporating larger datasets and diverse geological conditions to enhance model robustness and applicability.</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

An Interpretable Machine Learning Approach for Predicting the End-Bearing Capacity of Rock-Socketed Piles

  • 期刊论文

  • 发表日期：

    2025年12月5日

查看全文
  • Abdul Aziz,
  • Laiba Gulaly,
  • Majid Khan

摘要

This study presents a comprehensive evaluation of advanced machine learning models for predicting the end-bearing capacity of rock-socketed piles. Eight algorithms were examined, including multivariate expression programming, artificial neural networks, decision trees, random forests, extreme gradient boosting, convolutional neural networks, long short-term memory networks, and categorical boosting. Five key parameters, uniaxial compressive strength of rock, geological strength index, pile diameter, pile length in soil, and pile length in rock, were used as input variables. Among the models, categorical boosting and extreme gradient boosting achieved the highest accuracy, with coefficients of determination of 0.979 and 0.978, respectively. A simplified mathematical equation was also developed using the multivariate expression programming model to provide a practical and interpretable prediction tool. Feature importance and partial dependence analyses revealed that rock strength and geological strength index are the dominant factors influencing end-bearing capacity, while pile geometry has a secondary effect. The results indicate that machine learning models outperform traditional empirical equations, offering significantly improved prediction accuracy and generalization. This research highlights the potential of data-driven approaches for more reliable design of pile foundations and provides a foundation for future studies incorporating larger datasets and diverse geological conditions to enhance model robustness and applicability.