Harnessing Machine Learning to Predict Nurse Turnover Intention and Uncover Key Predictors: A Multinational Investigation

Aims: To predict nurses' turnover intention using machine learning techniques and identify the most influential psychosocial, organisational and demographic predictors across three countries. Design: A cross-sectional, multinational survey design. Methods: Data were collected from 1625 nurses in the United States, Türkiye and Malta between June and September 2023 via an online survey. Twenty variables were assessed, including job satisfaction, psychological safety, depression, presenteeism, person-group fit and work engagement. Turnover intention was transformed into a binary variable using unsupervised machine learning (k-means clustering). Six supervised algorithms—logistic regression, random forest, XGBoost, decision tree, support vector machine and artificial neural networks—were employed. Model performance was evaluated using accuracy, precision, recall, F1 score and Area Under the Curve (AUC). Feature importance was examined using logistic regression (coefficients), XGBoost (gain) and random forest (mean decrease accuracy). Results: Logistic regression achieved the best predictive performance (accuracy = 0.829, f1 = 0.851, AUC = 0.890) followed closely by support vector machine (polynomial kernel) (accuracy = 0.805, f1 0.830, AUC = 0.864) and random forest (accuracy = 0.791, f1 = 0.820, AUC = 0.859). In the feature importance analysis, job satisfaction consistently emerged as the most influential predictor across all models. Other key predictors identified in the logistic regression model included country (USA), work experience (6–10 years), depression and psychological safety. XGBoost and random forest additionally emphasised the roles of work engagement, group-level authenticity and person–group fit. Job-stress-related presenteeism was uniquely significant in XGBoost, while depression ranked among the top predictors in both logistic regression and random forest models. Conclusion: Machine learning can effectively predict turnover intention using multidimensional predictors. This methodology can support data-driven decision-making in clinical retention strategies. Impact: This study provides a data-driven framework to identify nurses at risk of turnover. By integrating machine learning into workforce planning, healthcare leaders can develop targeted, evidence-based strategies to enhance retention and improve organisational stability. Reporting Method: This study adhered to STROBE reporting guideline. Patient and Public Contribution: This study did not include patient or public involvement in its design, conduct or reporting.