Research on an identification model for mine water inrush sources based on the HBA-CatBoost algorithm

Area of Science:

• Geosciences
• Computer Science
• Mining Engineering

Background:

• Mine water inrush is a critical hazard in underground mining.
• Accurate identification of water inrush sources is essential for effective mine water management.
• Existing methods face challenges in pinpointing water inrush sources, particularly in complex geological settings.

Purpose of the Study:

• To introduce and validate a novel model, HBA-CatBoost, for the accurate and efficient identification of mine water inrush sources.
• To enhance the performance of the CatBoost model using the hybrid bat algorithm (HBA).
• To employ the SHAP method for interpreting the model's decision-making process.

Main Methods:

• Development of the HBA-CatBoost model by optimizing the CatBoost algorithm with the hybrid bat algorithm.
• Application of the SHAP (Shapley Additive Explanation) method to interpret model predictions.
• Comparative analysis of HBA-CatBoost against other models (HBA-RF, HBA-XGBoost, CatBoost, RF, XGBoost).
• Validation of the model's practicality at the Longfeng Coal Mine in northern Guizhou.

Main Results:

• The HBA-CatBoost model demonstrated superior classification performance compared to other evaluated models.
• Achieved high scores in accuracy (96.43%), recall (97.22%), precision (96.43%), and F1 (96.61%).
• The hybrid bat algorithm significantly improved CatBoost's convergence speed and classification accuracy.
• SHAP analysis provided insights into feature importance, enhancing model credibility.

Conclusions:

• The HBA-CatBoost model is a highly effective tool for identifying mine water inrush sources.
• The integration of HBA with CatBoost offers significant performance improvements for mine hazard prediction.
• The model's interpretability via SHAP increases its reliability and practical applicability in mining operations.