[Analysis of associated factors for occupational hearing loss and construction of a machine learning prediction model]

Objective: To investigate the associated factors of hearing loss among occupational populations and to develop a machine learning predictive model to serve as a reference for risk screening of occupational hearing loss. Methods: In March 2025, based on health examination data from 7969 occupational workers in Zhongshan City in 2020, single-factor analysis, the Boruta algorithm, LASSO regression, and multivariate logistic regression were employed to identify and analyze factors associated with hearing loss. Six predictive models-logistic regression, naive Bayes, decision tree, random forest, Bayesian network, and extreme gradient boosting (XGBoost) -were constructed. The optimal model was selected through comprehensive evaluation of discrimination and calibration performance, and the (Shapley additive explanations, SHAP) method was utilized to interpret feature importance. Results: Among the 7969 occupational workers individuals, hearing loss was detected in 39.0% of cases. Multivariate logistic regression analysis revealed that the risk of hearing loss was lower in females than in males (OR=0.40, 95%CI: 0.33-0.48, P<0.001) ; compared to the <30-year-old group, the risk of hearing loss was elevated in the 40-49-year-old group (OR=2.49, 95%CI: 2.17-2.86, P<0.001) ; both noise exposure (OR=1.70, 95%CI: 1.36-2.12, P<0.001) and exposure to welding fumes (OR=1.68, 95%CI: 1.46-1.93, P<0.001) were identified as significant exposure factors for hearing loss. Among the six models, the XGBoost model demonstrated the best performance in terms of accuracy (66.23%), precision (57.53%), and area under the curve (AUC) =0.701. Conclusion: Age, noise exposure, welding fume exposure, and sex were significantly associated with hearing loss. The XGBoost model showed relatively good discriminative ability for predicting hearing loss risk, indicating its preliminary potential for screening high-risk occupational groups. However, due to the cross-sectional design, causal relationships need to be verified in future studies.