1
24-009
An Integrated Method to Classify Ground-Fall Accidents and
to Estimate Ground-Fall Trends in U.S. Mines Using Machine
Learning Algorithms
Gamal Rashed
CDC/NIOSH/PMRD
Yuting Xue
CDC/NIOSH/PMRD
Connor Brown
CDC/NIOSH/PMRD
Zoheir Khademian
CDC/NIOSH/PMRD
Khaled Mohamed
CDC/NIOSH/PMRD
ABSTRACT
Ground falls in U.S. underground coal mines can lead to
significant consequences, including loss of life, injuries,
damaged equipment, and production stoppage. Improving
the safety of the workplace is of utmost importance for
mine workers and the U.S. economy. The Mine Safety and
Health Administration (MSHA) accident/injury/illness
dataset provides short narratives for reported incidents,
including ground-falls. The main objective of this study is
to develop a framework that includes: 1) utilizing machine
learning algorithms to categorize ground-fall incidents
from narratives based on the main cause of the occurrence
and 2) demonstrating an example of a user-friendly visual-
ization to display injury/fatality trends from narratives in
U.S. coal mines between 1983 and 2021. The developed
framework was tested on a subset of the data and achieved
an average F1-score of 96% in categorizing the incidents.
The outcome will help identify areas requiring additional
research and innovative solutions to reduce severe occupa-
tional hazards.
INTRODUCTION
Accidents due to ground-fall failures in coal mines can
potentially have severe consequences, including both fatal
and non-fatal injuries, damage of equipment, impaired
ventilation, and production delay/stoppage. Improving
the safety of the workplace in U.S. coal mines is of utmost
importance for mine workers, mine operators, and the
U.S. economy. Between 2010 and 2019, the ground-fall
incidents in U.S. mines resulted in 46 fatalities, 33 per-
manent disabilities, 3,082 injuries, 119,520 non-fatal days
lost, and 12,433 days of restricted work activities (Rashed
et al. 2022). In MSHA dataset, the accident/injury/illness
datafile combines five different types of ground fall acci-
dents (roof fall, rib fall, face fall, rock outburst, and high-
wall failure) into two categories called “fall of roof, back,
or brow from in-place” and “fall of face, rib, pillar, side,
or highwall.” The MSHA dataset provides short narratives
for ground-fall incidents however, it does not classify them
based on the root cause of the incident. MSHA fatality
reports include the root cause of incidents. However, they
are not considered in this study. From a prevention perspec-
tive, ascertaining root causes of incidents, often provided
in the narratives, is necessary for the research to address
mitigation strategies and identifying potential gaps in sci-
entific research. Note that accidents and incidents are used
interchangeably in this study.
Rashed et al. 2022 manually classified thousands of
ground-fall narratives into five categories using a subset of
the MSHA dataset. However, this process was time-con-
suming and tedious, which is why in this study the authors
raised a question about the capability of utilizing machine
24-009
An Integrated Method to Classify Ground-Fall Accidents and
to Estimate Ground-Fall Trends in U.S. Mines Using Machine
Learning Algorithms
Gamal Rashed
CDC/NIOSH/PMRD
Yuting Xue
CDC/NIOSH/PMRD
Connor Brown
CDC/NIOSH/PMRD
Zoheir Khademian
CDC/NIOSH/PMRD
Khaled Mohamed
CDC/NIOSH/PMRD
ABSTRACT
Ground falls in U.S. underground coal mines can lead to
significant consequences, including loss of life, injuries,
damaged equipment, and production stoppage. Improving
the safety of the workplace is of utmost importance for
mine workers and the U.S. economy. The Mine Safety and
Health Administration (MSHA) accident/injury/illness
dataset provides short narratives for reported incidents,
including ground-falls. The main objective of this study is
to develop a framework that includes: 1) utilizing machine
learning algorithms to categorize ground-fall incidents
from narratives based on the main cause of the occurrence
and 2) demonstrating an example of a user-friendly visual-
ization to display injury/fatality trends from narratives in
U.S. coal mines between 1983 and 2021. The developed
framework was tested on a subset of the data and achieved
an average F1-score of 96% in categorizing the incidents.
The outcome will help identify areas requiring additional
research and innovative solutions to reduce severe occupa-
tional hazards.
INTRODUCTION
Accidents due to ground-fall failures in coal mines can
potentially have severe consequences, including both fatal
and non-fatal injuries, damage of equipment, impaired
ventilation, and production delay/stoppage. Improving
the safety of the workplace in U.S. coal mines is of utmost
importance for mine workers, mine operators, and the
U.S. economy. Between 2010 and 2019, the ground-fall
incidents in U.S. mines resulted in 46 fatalities, 33 per-
manent disabilities, 3,082 injuries, 119,520 non-fatal days
lost, and 12,433 days of restricted work activities (Rashed
et al. 2022). In MSHA dataset, the accident/injury/illness
datafile combines five different types of ground fall acci-
dents (roof fall, rib fall, face fall, rock outburst, and high-
wall failure) into two categories called “fall of roof, back,
or brow from in-place” and “fall of face, rib, pillar, side,
or highwall.” The MSHA dataset provides short narratives
for ground-fall incidents however, it does not classify them
based on the root cause of the incident. MSHA fatality
reports include the root cause of incidents. However, they
are not considered in this study. From a prevention perspec-
tive, ascertaining root causes of incidents, often provided
in the narratives, is necessary for the research to address
mitigation strategies and identifying potential gaps in sci-
entific research. Note that accidents and incidents are used
interchangeably in this study.
Rashed et al. 2022 manually classified thousands of
ground-fall narratives into five categories using a subset of
the MSHA dataset. However, this process was time-con-
suming and tedious, which is why in this study the authors
raised a question about the capability of utilizing machine