Predictive Modeling for Occupational Safety Outcomes and Days Away from Work Analysis in Mining Operations

Area of Science:

• Occupational Safety and Health
• Data Science
• Mining Engineering

Background:

• Mining is a high-risk occupation with persistent safety concerns.
• Traditional safety measures are insufficient to prevent significant accidents.
• Machine learning (ML) offers potential for proactive safety improvements in hazardous industries.

Purpose of the Study:

• To evaluate the efficacy of ML techniques in predicting mining accident outcomes and days away from work.
• To compare the predictive performance of ML models trained on narrative versus tabular data.
• To investigate the impact of data augmentation on imbalanced mining safety datasets.

Main Methods:

• Utilized decision tree, random forest, and artificial neural networks for accident outcome prediction.
• Employed logistic regression as a baseline for performance comparison.
• Trained models on structured (tabular) and unstructured (narrative) accident data from the Mine Safety and Health Administration.
• Applied synthetic data augmentation with word embedding to address data imbalance.

Main Results:

• Models trained on narrative data demonstrated superior predictive power for accident outcomes compared to tabular data.
• Tabular data models achieved lower mean squared error for predicting days away from work.
• Key predictors for days away from work include shift start time, accident time, and mining experience.
• Data augmentation improved F1 scores for underrepresented classes, enhancing model robustness.

Conclusions:

• Mining accident narratives contain valuable information for predicting injury severity, complementing structured data.
• ML models, particularly those using narrative data, can significantly enhance mining safety decision-making.
• Predictive analytics, informed by diverse data sources and augmentation techniques, are crucial for mitigating mining risks.