An innovative machine learning approach for slope stability prediction by combining shap interpretability and...

Area of Science:

• Geotechnical Engineering
• Data Science
• Machine Learning

Background:

• Conventional slope stability prediction methods are complex and data-intensive.
• Machine learning (ML) offers a promising alternative for overcoming these limitations.

Purpose of the Study:

• To develop and evaluate an innovative ML approach, combining SHapley Additive exPlanations (SHAP) and stacking ensemble learning (EL) (SHAP-EL), for classifying and predicting slope stability.
• To identify the most influential base learning models using SHAP analysis and integrate them via stacking EL for a robust predictive model.

Main Methods:

• Utilized 627 slope case records with geological and geotechnical parameters.
• Applied SHAP analysis to rank ten ML algorithms (XGBoost, GBM, AdaBoost, RF, SVM, ANN, ELM, GLM, GLMnet, CART).
• Developed a final predictive model by stacking the optimal ML models identified by SHAP analysis.

Main Results:

• The SHAP-EL model achieved high predictive performance with 96.24% accuracy, 91.89% Kappa, 96.74% Precision, 95.70% Recall, and 96.22% F1-Score.
• SHAP-EL outperformed individual ML models, including XGBoost (94.64%), RF (94.09%), AdaBoost (93.55%), and GBM (82.80%).
• SHAP analysis effectively identified influential base learning models for ensemble integration.

Conclusions:

• The SHAP-based stacking EL model demonstrates superior accuracy and reliability for soil slope stability prediction.
• This approach can enhance geotechnical risk assessment and mitigate damages from slope failures.
• The findings contribute to advancing data-driven methods in geotechnical engineering.