Construction of a Big Data-Driven Predictive Analysis Platform for Hospital Talent Attrition

Abstract

Background: Hospital talent attrition poses a critical challenge to healthcare systems, particularly in high-stress environments where workforce stability is directly tied to service quality and patient outcomes. Existing turnover prediction models often neglect the psychological and cultural dimensions that shape employee behavior, relying solely on demographic and job-related data.
Objective: This study aims to construct a big data-driven predictive analysis platform for hospital staff attrition that integrates machine learning with psychological constructs. Specifically, we introduce Negotiable Fate, a culturally rooted belief system, as a key explanatory variable influencing turnover behavior via psychological capital and organizational citizenship.
Methods: We utilized structured human resource data from a tertiary public hospital, including 20+ features across 400+ employees. Due to the imbalance in attrition cases (approx. 5%), we employed the SMOTE technique to generate a balanced dataset. Four machine learning classifiers—Logistic Regression, Decision Tree, Random Forest, and XG Boost—were trained and evaluated using accuracy, precision, recall, and F1-score. Additionally, we conducted hierarchical regression, mediated moderation modeling, and confirmatory factor analysis on psychological survey data, involving variables such as negotiable fate, psychological capital, perceived organizational support, job performance, and OCB.
Results: Machine learning models achieved high performance, with Random Forest and XG Boost showing superior recall for the minority class. Feature importance rankings consistently identified working hours, income, job type, and satisfaction as core predictive features. Theoretical model testing confirmed that negotiable fate significantly predicts job performance (β=0.30, p<.001) and OCB (β=0.36, p<.001) through the mediating role of psychological capital (β=0.33, p<.001). Moderation effects of perceived organizational support on the NF→PC pathway were also significant (β=0.16, p<.001), confirming a mediated moderation mechanism.
Conclusion: By combining machine learning with psychological theory, our platform not only enhances the accuracy of hospital attrition prediction but also provides deeper behavioral explanations. This integrative approach enables more targeted and culturally sensitive HR interventions in medical institutions.