HELIOS-Stack: A Novel Hybrid Ensemble Learning Approach for Precise Joint Roughness Coefficient Prediction in Rock...

Area of Science:

• Geotechnical Engineering
• Rock Mechanics
• Machine Learning Applications

Background:

• Accurate joint roughness coefficient (JRC) estimation is vital for rock mass mechanical behavior analysis.
• Existing predictive models struggle with complex geological surface morphology.
• Limitations in current methods necessitate advanced JRC prediction techniques.

Purpose of the Study:

• To develop an advanced hybrid ensemble learning methodology (HELIOS-Stack) for enhanced JRC prediction.
• To integrate multiple machine learning models and statistical analysis for superior accuracy.
• To establish a new benchmark for geological surface characterization.

Main Methods:

• Implemented a hybrid ensemble approach combining Random Forest, XGBoost, LightGBM, Support Vector Regression, and Multilayer Perceptron.
• Utilized a LightGBM meta-learner as the final estimator.
• Analyzed 112 rock samples using eight statistical parameters and evaluated against 12 empirical models.

Main Results:

• HELIOS-Stack achieved high accuracy with R² values of 0.9884 (training) and 0.9769 (testing).
• Demonstrated superior performance across metrics like Mean Absolute Error and Concordance Index compared to other models.
• Identified three distinct roughness clusters: high (JRC 16-20), moderate (JRC 8-15), and smooth (JRC 0.4-7).

Conclusions:

• The HELIOS-Stack methodology significantly advances rock discontinuity characterization.
• This approach offers unprecedented precision in JRC prediction for geological applications.
• Transformed applications in geotechnical engineering, rock mass stability, and geological modeling are anticipated.