Identifying Risk Factors from MSHA Accidents and Injury Data Using Logistic Regression

Abstract

The global mining industry has recorded significant declines in accident and injury rates attributed to the advancement in technology, increased enforcement, and safety consciousness. A goal of the mining industry is to achieve zero injury and occupational illness on all mine sites, prompting increased research into ways to further reduce mine accidents. A machine learning technique known as multiclass logistic regression is applied on a 10-year injury dataset from the Mine Safety and Health Administration (MSHA) to determine a miner’s susceptibility to a class of injury and to help identify significant risk factors associated with different classes of injury. The data is aggregated based on injury classification to provide statistically relevant results. The analysis identifies specific risk factors that influence a mine worker’s susceptibility to a given class of injury, i.e., non-fatal with no days lost or restricted activity, non-fatal with days lost and/or days of restricted work activity, and fatal and total permanent or partial permanent disability. These factors include miner’s age, mine type (coal vs. non-coal), experience on the current job (years), shift start time, employment type (operator vs. contractor), mining district, and type of accident. The results of the analysis indicate that a miner’s experience on the job, i.e., the number of years worked in a current job, is a significant risk to injury occurrence, even for those with decades of total mining experience. We further show the differences and similarities between the surface and underground mine incidents.