Prediction method for rock shear strength parameters based on data-driven and interpretability analysis.

Area of Science:

• Geotechnical Engineering
• Computational Intelligence
• Machine Learning

Background:

• Predicting rock shear strength is crucial for geotechnical engineering safety.
• Existing models face limitations in handling complex nonlinearities and hyperparameter optimization.
• Accurate estimation of rock shear strength parameters (cohesion 'c' and friction angle 'φ') is essential for infrastructure design and stability analysis.

Purpose of the Study:

• To develop a novel, robust framework for predicting rock shear strength parameters.
• To address the challenge of random hyperparameter selection in ensemble models.
• To enhance prediction accuracy and interpretability in geotechnical engineering applications.

Main Methods:

• A stacking ensemble model was constructed using LightGBM, XGBoost, CatBoost, and Random Forest as base learners, with XGBoost as the meta-learner.
• The Chaos-Improved Sparrow Search Algorithm (CISSA), incorporating tent chaotic mapping and hybrid mutation, was developed to optimize the stacking model's hyperparameters.
• The model was trained and validated on 199 rock datasets using fivefold cross-validation, with performance evaluated using R², RMSE, and MAE. SHAP analysis was used for interpretability.

Main Results:

• The CISSA-Stacking model achieved high accuracy, with R² values of 0.9936 for cohesion (c) and 0.9744 for friction angle (φ).
• Performance metrics (RMSE, MAE) demonstrated the model's superior accuracy over benchmark methods.
• SHAP analysis identified key predictive factors: Vp, UCS, BTS, and ρ for cohesion, and ρ, UCS, Vp, and BTS for friction angle.

Conclusions:

• The developed CISSA-Stacking model offers a significant advancement in predicting rock shear strength parameters.
• The optimized ensemble approach provides superior accuracy and reliability compared to conventional methods.
• An intelligent prediction software was created, enabling rapid and accurate estimation of rock shear strength parameters for practical engineering applications.