Predicting the Risk of Burnout Syndrome Using Korean Occupational Stress Scale (KOSS): A Machine Learning Approach.

Background: Changes in the workplace have increased occupational stress, leading to health issues such as burnout syndrome (BOS), which results from poorly managed chronic workplace stress. The Korean Burnout Syndrome Scale (KBOSS) has been used to diagnose these issues, but its stigma and decreased compliance with application have limited its widespread use. This study aimed to develop machine learning models to predict BOS risk from occupational stress factors and identify these influential factors. Methods: Using a dataset of 1,205 individuals across 40 companies, we evaluated the predictive performance of five machine learning algorithms. Each model was optimized via resampling and 5-fold grid search cross-validation and evaluated using an ROC-AUC, balanced accuracy, overall accuracy, and F1 score. SHAP was used to quantify the contribution of each feature to the prediction, identifying key occupational stress factors. Results: All five models demonstrated strong predictive performance, with random forest achieving the most balanced results across the evaluation metrics, including a ROC-AUC of 0.904. SHAP analysis identified "Job instability" and "Lack of reward" as the most substantial BOS risk factors; "Relationship conflict" and "Organizational system" also played important roles. Moreover, the relationship between the SHAP values and feature values revealed critical transition points between "Agree" and "Disagree" responses for each KOSS factor. Conclusion: This study demonstrated that machine learning can effectively predict BOS risk based on occupational stress factors. By enabling the early identification of at-risk employees, this approach improves cost efficiency and offers a scalable solution for BOS risk assessment and intervention.