Source discrimination of mine water based on the random forest method.

Area of Science:

• Hydrogeology
• Data Science
• Environmental Engineering

Background:

• Machine learning is increasingly used for predictive modeling in engineering.
• Accurate estimation of mine water sources is crucial for preventing engineering disasters.
• Random forest classification is a robust machine learning technique for pattern recognition.

Purpose of the Study:

• To estimate and classify mine water sources using machine learning tools.
• To develop a predictive model for mine water discrimination in the Pingdingshan coalfield.
• To leverage hydrochemical data for water source classification.

Main Methods:

• Utilized random forest classification, a machine learning algorithm.
• Trained the model using hydrochemical data from the Pingdingshan coalfield.
• Employed fourfold cross-validation (CV) to ensure model accuracy and robustness.

Main Results:

• The random forests model successfully classified five types of mine water (surface, Quaternary pore, Carboniferous limestone karst, Permian sandstone, Cambrian limestone karst).
• The model demonstrated significant predictive accuracy and robustness in discriminating between water types.
• Hydrochemical data processing formed the basis for the successful model training.

Conclusions:

• The developed random forests model provides effective guidance for mine water discrimination.
• Machine learning offers an accessible approach for predictive model development in preventing engineering disasters.
• This research highlights the utility of random forest classification in hydrogeological studies.