Enhancing shear strength predictions of rocks using a hierarchical ensemble model

Area of Science:

• Geotechnical Engineering
• Rock Mechanics
• Material Science

Background:

• Shear strength (SS) parameters are critical for material behavior analysis.
• Traditional models have limitations in predicting SS parameters accurately.

Purpose of the Study:

• To propose a novel hierarchical ensemble model (HEM) for predicting cohesion (c) and angle of internal friction (φ).
• To address limitations of existing machine learning models for SS parameter prediction.

Main Methods:

• Developed and validated a hierarchical ensemble model (HEM).
• Employed leave-one-out cross-validation (LOOCV) and out-of-bag (OOB) evaluation.
• Assessed model accuracy using R-squared (R²), AAREP, Taylor diagrams, and Q-Q plots.

Main Results:

• HEM achieved R² values of 0.93 for cohesion and 0.979 for friction angle.
• AAREP values were 4.7% for cohesion and 1.96% for friction angle.
• The model demonstrated superior performance compared to previous studies on the same dataset.

Conclusions:

• The HEM significantly enhances the prediction accuracy and generalization capability for cohesion and friction angle.
• Maximum principal stress (σ_3max) was identified as the most impactful parameter.
• LOOCV and OOB methods showed the widest uncertainty bands for cohesion and friction angle, respectively.