Predicting Nurse Turnover for Highly Imbalanced Data Using the Synthetic Minority Over-Sampling Technique and Machine...

Area of Science:

• Healthcare Management
• Data Science
• Nursing Workforce Research

Background:

• Nurse turnover presents a significant challenge in healthcare, impacting service quality and the nursing profession.
• Addressing class imbalance is critical for accurate predictive modeling in nurse turnover studies.

Purpose of the Study:

• To predict nurse turnover using machine learning algorithms on an imbalanced dataset.
• To identify key factors contributing to nurse turnover.
• To compare the performance of different machine learning models in predicting nurse turnover.

Main Methods:

• Utilized the 2018 National Sample Survey of Registered Nurses dataset.
• Applied Synthetic Minority Over-sampling Technique (SMOTE) to handle class imbalance.
• Implemented and compared four machine learning algorithms: logistic regression, random forests, decision tree, and extreme gradient boosting.
• Selected 18 variables as predictive features and performed model evaluation using accuracy, precision, recall, F1-score, and AUC.

Main Results:

• SMOTE-enhanced random forests demonstrated the highest predictive power, both with all 18 variables and an optimized set of eight key variables.
• Age was identified as the most influential factor in nurse turnover.
• Other significant factors included working hours, electronic health record/electronic medical record usability, individual income, and region.

Conclusions:

• Machine learning, particularly SMOTE-enhanced random forests, offers a robust approach for predicting nurse turnover.
• Identifying influential factors like age and working conditions can inform targeted retention strategies.
• This study provides valuable insights for healthcare stakeholders in selecting appropriate predictive models and understanding turnover drivers.