Scholarlink

<p>To accurately predict the slope stability, this study develops a hybrid machine learning framework that integrates the Beluga Whale Optimization (BWO) algorithm with interpretability analysis. The dataset consists of 324 slope samples characterized by six features: slope height (<i>H</i>), slope angle (<i>β</i>), unit weight (<i>γ</i>), cohesion (<i>C</i>), internal friction angle (<i>φ</i>), and pore water pressure coefficient (<i>r</i><sub><i>u</i></sub>). The BWO algorithm is employed to optimize the hyperparameters of five machine learning models [i.e., random forest (RF), gradient boosting decision tree (GBDT), support vector machine (SVM), multi-layer perceptron (MLP), and logistic regression (LR)]. Subsequently, the performance of these BWO-optimized models is systematically evaluated based on five classification indices. The BWO-RF model exhibits the best performance among all selected models, achieving an accuracy of 0.94 and an AUC of 0.96 on the test set, respectively. The BWO algorithm outperforms the genetic algorithm and particle swarm optimization in optimization effectiveness. SHapley Additive exPlanations interpretability analysis shows that the influence of the six features on slope stability is in the order of <i>C</i>, <i>φ</i>, <i>β</i>, <i>γ</i>, <i>H</i>, and <i>r</i><sub><i>u</i></sub>. <i>C</i>, <i>φ</i>, and <i>γ</i> are positive contributing factors to stability, while <i>H</i>, <i>β</i>, and <i>r</i><sub><i>u</i></sub> are negative inhibiting factors. The study also identifies critical influence thresholds for <i>C</i>, <i>φ</i>, <i>β</i>, <i>γ</i>, and <i>H</i>, with values of 31.2&#xa0;kPa, 28°, 40°, 22.2&#xa0;kN/m³, and 35&#xa0;m, respectively. These findings provide theoretical support for the optimized design and stability evaluation of slope engineering.</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 Framework for Slope Stability Prediction and Interpretability: Integration of BWO-Optimized Machine Learning Models with SHapley Additive exPlanations

  • 期刊论文

  • 发表日期：

    2026年7月15日

查看全文
  • Haiqiang Jiang,
  • Xiaoqi Li,
  • Xin Zhou,
  • Changhong Song,
  • Jing Luo,
  • Enliang Wang,
  • Yadi Min,
  • Aimin Chen,
  • Yuwei Chen

摘要

To accurately predict the slope stability, this study develops a hybrid machine learning framework that integrates the Beluga Whale Optimization (BWO) algorithm with interpretability analysis. The dataset consists of 324 slope samples characterized by six features: slope height (H), slope angle (β), unit weight (γ), cohesion (C), internal friction angle (φ), and pore water pressure coefficient (ru). The BWO algorithm is employed to optimize the hyperparameters of five machine learning models [i.e., random forest (RF), gradient boosting decision tree (GBDT), support vector machine (SVM), multi-layer perceptron (MLP), and logistic regression (LR)]. Subsequently, the performance of these BWO-optimized models is systematically evaluated based on five classification indices. The BWO-RF model exhibits the best performance among all selected models, achieving an accuracy of 0.94 and an AUC of 0.96 on the test set, respectively. The BWO algorithm outperforms the genetic algorithm and particle swarm optimization in optimization effectiveness. SHapley Additive exPlanations interpretability analysis shows that the influence of the six features on slope stability is in the order of C, φ, β, γ, H, and ru. C, φ, and γ are positive contributing factors to stability, while H, β, and ru are negative inhibiting factors. The study also identifies critical influence thresholds for C, φ, β, γ, and H, with values of 31.2 kPa, 28°, 40°, 22.2 kN/m³, and 35 m, respectively. These findings provide theoretical support for the optimized design and stability evaluation of slope engineering.