Predicting Teacher Turnover in Private Universities: A Machine Learning Approach Based on Ten Years of Data and Satisfaction Factors

Jingwen Wang^1*yi liu¹xiaohong yang²

• ¹School of Physical Education, Xi'an Fanyi University, Xi'an, China
• ²Northwest Normal University, Lanzhou, China

ABSTRACT Teacher turnover poses a significant challenge to the sustainable development of private universities in China. While machine learning (ML) has been increasingly applied to turnover prediction, existing studies often overlook psychological factors and lack longitudinal analysis. This study aims to bridge these gaps by integrating a ten-year longitudinal dataset with satisfaction surveys from a private university in Western China. Exploratory Factor Analysis (EFA) was employed to extract four key dimensions influencing turnover: Compensation, Benefits, and Development; Professional Efficacy; School Management; and Work Environment. Three ML models— K-Nearest Neighbors (KNN), Naive Bayes (NB), and Backpropagation Neural Network (BPNN)—were constructed and evaluated using accuracy, F1-score, and AUC. Results indicate that the KNN model achieved the highest predictive performance (accuracy = 83.64%, F1 = 84.16%, AUC = 0.901), while NB exhibited the highest recall (92.50%), suitable for preliminary risk screening. The "Compensation, Benefits, and Development" dimension was identified as the most influential factor, accounting for 25.41% of the variance. Qualitative analysis of open-ended responses further validated the EFA framework. This study proposes an "EFA + ML" hybrid approach that enhances feature interpretability and prediction robustness, offering practical insights for human resource management in private higher education institutions.