Prediction of Behavioral Health Employee Turnover With HR Data-Based Machine Learning Combined With Job Well-Being Indicators.

OBJECTIVE: Identifying employees at risk for turnover is critical to supporting staff retention in community behavioral health organizations (CBHOs). The authors propose a method that uses machine-learning (ML) models trained with historical human resources (HR) data in combination with survey-based indicators of job well-being to predict employee turnover. METHODS: The ML models trained with historical HR data (starting in 2011) were applied to current employees' HR data to estimate turnover probability, labeled [Formula: see text]. Additionally, 12 job well-being indicators were collected from current employees via organization-wide surveys, and their actual turnover status was determined 12 months later. Logistic regressions were used to test the predictability of [Formula: see text] combined with the job well-being indicators for predicting future turnover. RESULTS: The study included 303 CBHO employees (mean±SD age=43.2±13.2 years); 72% were female, and 60% were White. Seventy-two (24%) left their CBHO voluntarily by the time turnover outcome data were extracted. The [Formula: see text] predicted actual turnover, and adding three critical job well-being indicators (career advancement opportunities, expectation alignment, and importance of supervision) significantly improved prediction accuracy (the area under the curve increased from 0.63 to 0.76). Particularly [Formula: see text] (OR=2.15) and career advancement opportunities (OR=0.61) significantly predicted employee turnover when the other well-being indicators were held constant. CONCLUSIONS: This study provides a viable and less labor-intensive method for identifying employees at high risk for turnover by applying ML models trained on historical HR data and identifies additional survey data that could augment predictions.