Identifying Key Predictors and Developing a Machine Learning Model for Nurse Burnout in China

Increasing work pressure has elevated burnout risk among Chinese nurses; identifying associated factors and developing predictive models are essential for early intervention.

This study aims to develop a nurse burnout prediction model based on machine learning algorithms in order to assist nursing management in the future.

A cross-sectional online survey of 1391 Chinese nurses (burnout rate: 75.7%) was conducted from June to September 2025 using snowball sampling. Burnout was measured by MBI-GS, and 25 candidate variables were collected. Data were split 80/20 into training/validation sets. After variable selection (Boruta ∩ Group LASSO), seven algorithms were compared. Model evaluation used AUC, PR-AUC, calibration, DCA, and SHAP. The optimal threshold (Youden index) defined three risk tiers, and a Shiny web calculator was developed.

Random forest achieved the best test performance (AUC = 0.879, PR-AUC = 0.940, and sensitivity = 0.870). Key features included colleague relationship, work duration, exercise frequency, and night shift hours. SHAP analysis showed that harmonious colleague relationships, longer work duration, and regular exercise were associated with lower predicted risk, whereas longer night shifts and frequent training/exams were associated with higher predicted risk. Model calibration (HL p = 0.3505) and DCA confirmed clinical benefit. A Shiny web calculator was developed.

The random forest model effectively predicts nurse job burnout with good discrimination and calibration. The Shiny web calculator developed based on this model provides a practical tool for early identification and stratified intervention in nursing management. External validation is needed to further confirm generalizability.

The predictive model can be integrated into hospital human resource and nursing information systems to enable automatic assessment of burnout risk. This facilitates early identification of high-risk nurses, allowing managers to implement timely interventions such as adjusted schedules and psychological support. The deployed online calculator provides an accessible tool for dynamic monitoring and proactive management, supporting data-driven decision-making to improve nursing workforce stability and healthcare service quality.

No materials from other sources were used in this manuscript that require permission.

This study followed the TRIPOD + AI statement. The completed TRIPOD + AI checklist is provided as supporting information.