Development of targeted safety hazard management plans utilizing multidimensional association rule mining

Area of Science:

• Mining Engineering
• Data Science
• Safety Management

Background:

• Accident prevention in mining relies on hazard identification and management.
• Current methods often overlook temporal and spatial hazard data.
• Association rule mining (ARM) is underutilized for comprehensive mine safety plans.

Purpose of the Study:

• To integrate topic mining with ARM for analyzing multidimensional mine safety hazards.
• To uncover intrinsic association patterns among hazard attributes (Type, Level, Time, Location, Responsible Units).
• To develop a more specific and efficient hazard management framework for underground mines.

Main Methods:

• Applied Biterm Topic Model for dimensionality reduction and data aggregation.
• Developed multidimensional ARM model using standardized mining hazard attributes.
• Introduced evaluation indicators: Standard Lift and Difference Degree.
• Analyzed 1387 valid rules from an underground gold mine in China.

Main Results:

• Identified strong associations between hazard location and hazard type (1.934), responsible unit (1.412), and hazard level (1.240).
• Extracted 15 high-differentiation rules revealing complex attribute correlations.
• Developed targeted control measures for 19 specific locations.

Conclusions:

• Multidimensional ARM significantly enhances the specificity of safety training.
• The proposed approach improves the efficiency of safety hazard investigations.
• This method provides a robust framework for proactive mine safety management.