Data-Driven Safety Management in the Manufacturing Industry: A National Case Study of the Fabricated Metal Sector Using Machine Learning
Purpose: This study analyses OHS hazards and develops machine learning models to predict injury severity in PKD/NACE division 25 (manufacture of fabricated metal products). Using advanced predictive modelling, it provides practical OHS management recommendations while systematically addressing the challenges of highly unbalanced administrative datasets. Design/Methodology/Approach: The two-stage approach combines macro-level accident trend analysis with predictive modelling on a disaggregated ten-year dataset (28,082 records). Three Random Forest models were created and optimized using grid search and 10-fold cross-validation. Finally, Explainable AI (XAI) with a global surrogate tree extracted transparent decision rules. Findings: Macro-level analysis showed a persistent rise in incident severity despite declining overall accident frequencies. At the micro-level, standard classification algorithms suffered from the accuracy paradox due to unbalanced data. Combining the Random Forest algorithm with the SMOTEN method provided an optimal balance between sensitivity and specificity, with casualty age, material factor as injury source, and the activity performed at the time of the accident proving to be the dominant predictors of severe injuries. Practical Implications: The feature importance ranking and XAI surrogate tree provide safety decision-makers with clear, interpretable guidelines. Recommendations include shifting older, experienced workers from heavy, dynamic machinery operation to activities involving hand tools to reduce exposure to age-related injuries. Furthermore, targeted training and active safety measures should be intensified for younger workers operating dynamic machinery and moving structures. Originality/Value: This study bridges an essential gap in the literature by applying advanced ensemble learning and XAI techniques to a large, raw national dataset from the metal manufacturing sector. It represents an innovative, data-driven study designed to support proactive OHS management.