Predicting internal occupational benzene exposure levels in landfills using machine learning models

Abstract

Landfills are significant sources of fugitive benzene release, posing a threat to the health of occupational populations. In this study, long-term monitoring of both external and internal benzene occupational exposure was conducted at a municipal solid waste (MSW) landfill site. The annual average concentration of benzene exposure in the landfill was 0.78 ± 1.08 μg/m³, with a notable increase in autumn (1.40 ± 5.29 μg/m³). The internal biomarker level (urinary t, t-muconic acid) significantly increased post-shift for the occupational population (6.65 ± 50.75 mg/g cr) compared to pre-shift (5.32 ± 31.62 mg/g cr, p < 0.05), exceeding the American Conference of Government Industrial Hygienists (ACGIH) limit by 13 times (0.5 mg/g cr). Machine learning models, particularly the Support Vector Regression (SVR) algorithm, outperformed traditional methods (e.g., Michaelis-Menten) in predicting internal exposure (R² = 0.989, root mean square error = 0.085). Using the SVR model, the predicted internal benzene level under the occupational exposure limit (1.7 mg/m³) was 1.65 mg/g cr, exceeding the ACGIH limit by three-fold. These findings provide a novel framework for benzene exposure risk assessment and inform targeted landfill management strategies.