Scholarlink

<p>Reliable estimation of residual shear strength (<i>S</i><sub><i>ur</i></sub>) of liquefied soils is critical for the assessment of geotechnical stability in seismically active areas. This research proposes a new hybrid approach that combines empirical procedures with sophisticated machine learning (ML) techniques to improve prediction precision of <i>S</i><sub><i>ur</i></sub>. A hybrid model is formulated, which includes important geotechnical parameters: corrected Standard Penetration Test blow count, void ratio, initial vertical effective stress, and mean effective stress. The coefficients of the equation are optimized by weighted averaging of known models, and ML algorithms such as XGBoost, LightGBM, and Random Forest are utilized for verification. A normalized and expanded dataset from varied Indian soils adds robustness to the model’s validity. Interpretability is increased through SHAP and LIME analysis, which identifies as being key drivers. Findings show that the hybrid model, with the aid of ensemble ML techniques, performs better than conventional empirical methods, providing a solid and flexible tool for geotechnical risk assessment.</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

Integrating Empirical Models and ML Techniques to Enhance Residual Shear Strength Estimation

  • 期刊论文

  • 发表日期：

    2025年7月11日

查看全文
  • Shubhendu Vikram Singh,
  • Sufyan Ghani,
  • Faizanul Haque

摘要

Reliable estimation of residual shear strength (Sur) of liquefied soils is critical for the assessment of geotechnical stability in seismically active areas. This research proposes a new hybrid approach that combines empirical procedures with sophisticated machine learning (ML) techniques to improve prediction precision of Sur. A hybrid model is formulated, which includes important geotechnical parameters: corrected Standard Penetration Test blow count, void ratio, initial vertical effective stress, and mean effective stress. The coefficients of the equation are optimized by weighted averaging of known models, and ML algorithms such as XGBoost, LightGBM, and Random Forest are utilized for verification. A normalized and expanded dataset from varied Indian soils adds robustness to the model’s validity. Interpretability is increased through SHAP and LIME analysis, which identifies as being key drivers. Findings show that the hybrid model, with the aid of ensemble ML techniques, performs better than conventional empirical methods, providing a solid and flexible tool for geotechnical risk assessment.