Scholarlink

<p>Accurate and explainable slope stability assessment is critical in geotechnical engineering, where uncertainties in soil properties and site conditions often limit conventional analytical methods. Machine learning models have shown strong predictive capabilities for slope stability analysis, but their black-box nature hinders adoption in engineering decision-making. To address this gap, this paper presents a hybrid framework that integrates extreme gradient boosting (XGBoost) with Bayesian optimization (BO) for slope stability assessment and incorporates explainable techniques to enhance both accuracy and interpretability. The Bayesian algorithm optimizes hyperparameters and identifies key input features within the XGBoost model. The framework is tested on a dataset of 168 samples with six input parameters such as slope angle, friction angle, slope height, unit weight, cohesion, pore pressure ratio and stability status as the target. For model transparency, SHAP and LIME methods quantify feature contributions and reveal input–output relationships. Results show that the proposed BO-XGB model achieves 98.04% accuracy, precision, recall, and F1-score. It outperforms standard XGBoost and gradient boosting (96.08%), random forest and k-nearest neighbors (92.16%), support vector machine (88.24%), AdaBoost (80%), and logistic regression (60.78%). The proposed BO-XGB framework significantly improves prediction accuracy while maintaining interpretability, establishing it as a practical tool for smart geotechnical infrastructure applications.</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 Explainable and Optimized Machine Learning Framework for Slope Stability Assessment

  • 期刊论文

  • 发表日期：

    2026年5月30日

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

摘要

Accurate and explainable slope stability assessment is critical in geotechnical engineering, where uncertainties in soil properties and site conditions often limit conventional analytical methods. Machine learning models have shown strong predictive capabilities for slope stability analysis, but their black-box nature hinders adoption in engineering decision-making. To address this gap, this paper presents a hybrid framework that integrates extreme gradient boosting (XGBoost) with Bayesian optimization (BO) for slope stability assessment and incorporates explainable techniques to enhance both accuracy and interpretability. The Bayesian algorithm optimizes hyperparameters and identifies key input features within the XGBoost model. The framework is tested on a dataset of 168 samples with six input parameters such as slope angle, friction angle, slope height, unit weight, cohesion, pore pressure ratio and stability status as the target. For model transparency, SHAP and LIME methods quantify feature contributions and reveal input–output relationships. Results show that the proposed BO-XGB model achieves 98.04% accuracy, precision, recall, and F1-score. It outperforms standard XGBoost and gradient boosting (96.08%), random forest and k-nearest neighbors (92.16%), support vector machine (88.24%), AdaBoost (80%), and logistic regression (60.78%). The proposed BO-XGB framework significantly improves prediction accuracy while maintaining interpretability, establishing it as a practical tool for smart geotechnical infrastructure applications.