Risk classification of hand-arm vibration exposure using machine learning for occupational health protection

This study evaluates the effectiveness of supervised machine learning models for hand-arm vibration (HAV) risk classification using field data collected from grass-cutting operators, selected as a controlled proxy group for high-vibration occupational tasks. The proposed framework is intended for future application in hazardous work environments, including mining, where early HAV risk identification is critical for occupational health protection.

A total of 292 field-based HAV exposure measurements were collected from six operators using standardized A(8) vibration assessments, complemented by Nordic Body Map (NBM) scores and personal attributes. Based on predictive performance and computational efficiency a total of 10 supervised machine learning models were compared by evaluating for multi-class HAV risk classification (Low Risk, Moderate Risk, High Risk). SHAP analysis was used for examining model interpretability.

Across risk categories, superior and balanced performance was demonstrated by ensemble-based models. The highest overall classification accuracy (98.31%) with very low inference time (0.002 s) was achieved by Gradient Boosting. Random Forest and K-Nearest Neighbours also showed high precision, recall, and F1-score values. Feature selection analysis reiterated vibration exposure magnitude as the dominant factor contributing to risk, with individual characteristics influencing risk differentiation.

The results showcased that early identification of High-Risk exposure conditions can be achieved through machine learning based HAV risk classification. This can enable timely preventive interventions such as exposure control, task modification, and targeted health surveillance. The study provides a robust methodological foundation for future field validation in risk areas like mining and other high-vibration occupational environments.