Journal of Occupational

Background: While traditional statistical methods remain indispensable tools in exposure assessment, they are often limited in capturing multifactorial interactions which characterize modern workplace hazards. This study aimed to integrate classical statistical inference with data-driven machine learning approaches to ascertain and prioritize occupational health hazards in a large industrial setting. This study compares classical inferential statistics with data mining techniques to stratify Health, Safety, and Environment (HSE) risks.
Materials and Methods: In this cross-sectional comparative study (2025), a stratified random sampling approach was applied to extract 647 validated HSE measurements across 50+ job roles within a major steel manufacturing plant in Iran. In order to enable multi-hazard comparisons, all exposure data were mathematically standardized into a dimensionless Exposure Severity Ratio (ESR = Measured/OEL). Classical analyses comprised one-sample T-tests, one-way ANOVA, and Scheffé post-hoc tests (α=0.05; SPSS v26.0). The data mining approach utilized a CHAID decision tree algorithm (parent node n=60, child node n=30, 10-fold cross-validation).
Results: Significant threshold exceedances (p<0.001) were noted for cadmium, chromium, iron, lead, noise, lighting/radiation, thermal stress, and ergonomic factors (overall mean ESR=1.083, SD=0.965). ANOVA confirmed significant inter-occupational variances (F(13,634)=24.309, p<0.001). The CHAID algorithm successfully stratified the risk landscape, identifying critical non-linear interactions between specific hazard domains and job roles (p<0.001), yielding actionable multidimensional risk profiles.
Conclusion: Compared to the isolated assessment of classical methods, the CHAID decision tree provides superior interpretability for uncovering non-linear HSE hazard interactions. This integrated methodological approach offers a highly effective means for prioritizing targeted, evidence-based risk mitigation strategies in heavy industries.