Assessing Employee Wellbeing and Retention Risk: An Integrated and Explainable Machine-Learning Decision-Support Framework Using Validated Psychometric Measures

Employee burnout, poor work-life balance, and voluntary turnover are costly problems, yet they are usually studied separately and inferred from administrative records rather than measured directly. This study models the three together, explains them transparently, and translates them into decision support within one framework. Guided by Job Demands-Resources and Conservation of Resources theory, we surveyed 304 full-time employees in India using the Copenhagen Burnout Inventory, the Hayman work-life balance scale, and the six-item Turnover Intention Scale, and established job demand and resource measures. All scales were reliable (Cronbach's alpha 0.70 to 0.95). For each outcome, five machine-learning algorithms were benchmarked under repeated five-fold cross-validation and interpreted with Shapley Additive Explanations computed on held-out observations. Turnover intention was the most learnable outcome (random forest R² = 0.52; area under the receiver operating characteristic curve (ROC-AUC) = 0.84); removing burnout and work-life balance from its inputs lowered these values to 0.39 and 0.81. Burnout and work-life balance were modeled moderately well (R² of 0.36 and 0.40). Job satisfaction, weekly working hours, overtime frequency, and organizational support were the leading controllable drivers. A partial least squares structural equation model supported H3, H5, H6, and H7 and partially supported H1, H2, and H4: weekly working hours, rather than perceived workload, carried the demand effect, and burnout and work-life balance mediated the effect of working conditions on turnover intention. A provisional Employee Wellbeing-Retention Index tiered the workforce by combined risk. Because the data are cross-sectional, the models identify employees at risk rather than forecasting turnover.