Scholarlink

<p>This paper presents a hybrid explainable framework using a Gradient Boosting model and Bayesian Optimization for predicating the end-bearing capacity of rock-socketed piles in geotechnical applications. The proposed system is applied to a dataset of 138 field tests, involving several parameters like uniaxial compressive strength, Geological Strength Index, pile lengths in soil and rock, and pile diameter. The performance of the proposed methodology is compared with other models as Random Forest and Decision Tree, and empirical formulas. From the results, the suggested model accomplished a coefficient of determination (R<sup>2</sup>) of 0.985, a root-mean-square error (RMSE) of 8.08, and a mean absolute error (MAE) of 4.86, outperforming other models. In contrast, empirical methods yielded R<sup>2</sup> values below 0.2. SHAP and LIME investigations presented model interpretability, identifying the uniaxial compressive strength, pile diameter, and pile length in rock as the most influential parameters. In concise, the proposed framework provides accurate, explainable results, enabling more efficient and convincing geotechnical design.</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

A Hybrid XAI Framework Using Hyperparameter-Optimized Machine Learning for Predicting the End-Bearing Capacity of Rock-Socketed Piles

  • 期刊论文

  • 发表日期：

    2026年3月31日

查看全文
  • Ezz El-Din Hemdan,
  • M.E. Al-Atroush

摘要

This paper presents a hybrid explainable framework using a Gradient Boosting model and Bayesian Optimization for predicating the end-bearing capacity of rock-socketed piles in geotechnical applications. The proposed system is applied to a dataset of 138 field tests, involving several parameters like uniaxial compressive strength, Geological Strength Index, pile lengths in soil and rock, and pile diameter. The performance of the proposed methodology is compared with other models as Random Forest and Decision Tree, and empirical formulas. From the results, the suggested model accomplished a coefficient of determination (R2) of 0.985, a root-mean-square error (RMSE) of 8.08, and a mean absolute error (MAE) of 4.86, outperforming other models. In contrast, empirical methods yielded R2 values below 0.2. SHAP and LIME investigations presented model interpretability, identifying the uniaxial compressive strength, pile diameter, and pile length in rock as the most influential parameters. In concise, the proposed framework provides accurate, explainable results, enabling more efficient and convincing geotechnical design.