Scholarlink

<p>The interpretability of machine learning (ML) models has garnered significant attention in recent years. This is particularly due to the promising performance of ML techniques in uniaxial compressive strength (UCS) prediction. Uncovering precise relationships between predictors and UCS is crucial for understanding their causal effects and for developing effective measures. However, there is currently no comprehensive study that reviews ML model interpretation methods on UCS or offers clear guidance for UCS researchers and practitioners. This research addresses that gap by critically evaluating and comparing ML interpretation methods for their applicability in UCS prediction modelling. The ML modelling methods used in this research include light gradient boosting machine (LightGBM), extreme gradient boosting machine (XGBoost), categorical boosting (CatBoost), and the interpretability methods examined include local interpretable model-agnostic explanations (LIME), local sensitivity analysis (LSA), partial dependence plots (PDP), global sensitivity analysis (GSA), and Shapley additive explanations (SHAP). The results reveal that XGBoost performed better than LightGBM and CatBoost, especially in&#xa0;the testing phase. Furthermore, XGBoost with LIME, LSA, PDP, GSA, and SHAP reveal that Schmidt Hammer rebound number (SRn) is the most significant factor influencing UCS prediction, This research demonstrates the potential and benefits of interpretability methods to understand the prediction mechanism of ML models in rock UCS modelling, making up for the deficiency that ML models cannot directly differentiate causation.</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

Exploring the Interpretability of Machine Learning Approaches in Modelling the Uniaxial Compressive Strength of Rocks

  • 期刊论文

  • 发表日期：

    2025年7月15日

查看全文
  • Chukwuemeka Daniel,
  • Feng Gao,
  • Xin Yin,
  • Zakaria M. Barrie,
  • Leonardo Z. Wongbae,
  • Peitao Li,
  • Yucong Pan

摘要

The interpretability of machine learning (ML) models has garnered significant attention in recent years. This is particularly due to the promising performance of ML techniques in uniaxial compressive strength (UCS) prediction. Uncovering precise relationships between predictors and UCS is crucial for understanding their causal effects and for developing effective measures. However, there is currently no comprehensive study that reviews ML model interpretation methods on UCS or offers clear guidance for UCS researchers and practitioners. This research addresses that gap by critically evaluating and comparing ML interpretation methods for their applicability in UCS prediction modelling. The ML modelling methods used in this research include light gradient boosting machine (LightGBM), extreme gradient boosting machine (XGBoost), categorical boosting (CatBoost), and the interpretability methods examined include local interpretable model-agnostic explanations (LIME), local sensitivity analysis (LSA), partial dependence plots (PDP), global sensitivity analysis (GSA), and Shapley additive explanations (SHAP). The results reveal that XGBoost performed better than LightGBM and CatBoost, especially in the testing phase. Furthermore, XGBoost with LIME, LSA, PDP, GSA, and SHAP reveal that Schmidt Hammer rebound number (SRn) is the most significant factor influencing UCS prediction, This research demonstrates the potential and benefits of interpretability methods to understand the prediction mechanism of ML models in rock UCS modelling, making up for the deficiency that ML models cannot directly differentiate causation.