Machine Learning Models for Identifying Factors Associated With Workplace Violence Among Emergency Nurses: A Comparative Study.

OBJECTIVE: To develop and compare three machine learning models for identifying factors associated with workplace violence (WPV) among emergency department nurses. METHODS: A total of 1540 emergency department nurses from various regions of China were examined between December 2023 and January 2024. Data were collected using scales measuring WPV, work-family conflict, occupational stress, occupational burnout, nursing practice environment, and self-rated sleep quality. Three analytical models (logistic regression, decision tree, and random forest) were developed and compared to classify nurses with and without WPV experiences among emergency department nurses. Model performance was evaluated using sensitivity, specificity, PPV, NPV, F1-score, balanced accuracy, and area under the receiver operating characteristic curve (AUC). RESULTS: Among the 1540 nurses, 1309 individuals (85.0%) had experienced WPV in the past year. All three models indicated that work-family conflict, occupational burnout, and occupational stress were significantly associated with WPV experiences among emergency department nurses (p < 0.05). In the random forest and decision tree models, sleep disorders and the nursing practice environment were also identified as significant associated factors. The accuracy of the logistic regression, decision tree, and random forest models was 0.829, 0.851, and 0.859; the specificity was 0.667, 0.605, and 0.593; the sensitivity was 0.864, 0.903, and 0.916; and the F1 score was 0.893, 0.909, and 0.915, respectively, with AUC values of 0.832 (95% CI: 0.781-0.883), 0.768 (95% CI: 0.711-0.826), and 0.834 (95% CI: 0.783-0.885). CONCLUSIONS: Work-family conflict, burnout, occupational stress, the nursing practice environment, and sleep disorders were consistently associated with WPV experiences among emergency department nurses. Random forest achieved the highest sensitivity, F1-score, and overall accuracy, whereas logistic regression showed the highest specificity and balanced accuracy. Logistic regression and random forest demonstrated similar discrimination.