Predicting wellbeing in empty-nesters: an ensemble machine learning approach with data-driven feature selection.

Machine learning (ML) methods promise to enrich psychological research, yet worked examples that translate the logic of predictive modeling into psychological questions remain scarce. This study demonstrates a predictive ML workflow with ensemble feature selection, using the empty nest as its substantive case. Three research questions were addressed: whether empty-nest parents differ from parents whose children still live at home in happiness and life satisfaction (RQ1); which of 77 candidate predictors carry robust predictive signal for each wellbeing outcome (RQ2); and whether the same predictors operate in both groups (RQ3). European Values Study data (N = 43, 670 parents aged ≥45) were analyzed through propensity score matching, consensus feature selection across five complementary methods with nested cross-validation, and a three-phase group comparison. The groups showed virtually no difference in either wellbeing outcome, and this null result persisted after demographic matching. Because the design is cross-sectional, it concerns differences associated with empty-nest status rather than adjustment during the transition itself. Health status and perceived freedom/control emerged as the strongest predictors of happiness, whereas life satisfaction drew on a broader set spanning personal resources, social connections, institutional confidence, and social capital, a pattern partly consistent with Self-Determination Theory, particularly its autonomy and relatedness needs. Predictors overlapped substantially between groups, with only a modest subset of robust group-specific interaction effects. The findings suggest that wellbeing among parents in midlife and later adulthood depends on factors operating across family configurations rather than on empty-nest status, and they illustrate how predictive modelling can complement hypothesis-driven psychological research.