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

This study develops a big data-driven predictive platform for hospital staff attrition, integrating machine learning (ML) with psychological constructs. Negotiable Fate (NF), a culturally rooted belief system, is examined as a predictor of turnover via psychological capital (PC) and organizational citizenship.

Structured HR data from 400+ employees at a tertiary public hospital, covering 20+ features, were analyzed. Due to attrition imbalance (~5%), SMOTE was applied to balance the dataset. Four ML classifiers-logistic regression, decision tree, random forest, and XGBoost-were evaluated using accuracy, precision, recall, and F1-score. Statistical analyses assessed mediation, moderation, and construct validity using survey variables: NF, PC, perceived organizational support, job performance (JP), and organizational citizenship behavior.

Random Forest and XGBoost achieved superior recall for attrition cases. Feature importance consistently highlighted working hours, income, job type, and satisfaction as key predictors. NF significantly predicted JP (β = 0.30, p < 0.001) and organizational citizenship (β = 0.36, p < 0.001) through PC (β = 0.33, p < 0.001). Perceived organizational support moderated the NF → PC pathway (β = 0.16, p < 0.001), confirming mediated moderation.

Integrating ML with psychological theory enhances both the prediction and understanding of hospital staff attrition. The platform enables culturally sensitive, data-driven HR interventions, helping administrators identify high-risk employees and implement targeted strategies to reduce attrition, stabilize the workforce, and improve patient care.