Using Machine Learning to Evaluate Coal Geochemical Data with Respect to Dynamic Failures.

Area of Science:

• Geochemistry
• Mining Engineering
• Data Science

Background:

• Dynamic failure events pose significant risks in underground coal mining.
• Previous research identified geochemical markers correlating with these events, but causality is unclear.

Purpose of the Study:

• To develop a machine learning model for assessing dynamic failure probability using geochemical and petrographic data.
• To identify key geochemical parameters influencing dynamic failure.

Main Methods:

• Applied machine learning techniques including linear regression, random forest, dimensionality reduction, and cluster analysis.
• Utilized data from the Pennsylvania Coal Sample Databank and Mine Safety and Health Administration (MSHA) accident data.
• Performed hierarchical clustering after dimensionality reduction.

Main Results:

• Identified 7 out of 18 geochemical parameters as most impactful for model performance.
• Achieved high classification precision: 85.7% with logistic regression and 96.7% with random forest.
• Discovered four distinct clusters, with one predominantly representing dynamic failure events.

Conclusions:

• Machine learning models can effectively predict dynamic failure events in coal mines.
• Geochemical composition is a significant predictor of dynamic failure.
• Further research can refine predictive models for enhanced mine safety.