Predicting work–life balance insights: A data-driven approach using unsupervised learning

The evolving nature of work has altered how individuals manage the boundaries between their professional and personal lives, raising significant concerns about work–life balance and overall wellbeing. While work–life balance has been the subject of extensive research, contemporary methodologies often rely on predefined categories or self-reported metrics, limiting their ability to capture the intricate and diverse patterns that define daily work experiences. This study employs a data-driven approach to reveal hidden work–life balance profiles by applying unsupervised learning to comprehensive datasets on lifestyle and wellbeing. Using UMAP for dimensionality reduction alongside DBSCAN clustering enabled the identification of distinct groups characterized by different levels of productivity, stress, health behaviors, financial stability, and personal achievements. These profiles, derived from empirical data, are then evaluated using supervised learning models, including XGBoost, LightGBM, KNN, and other classification methods, to assess their stability and distinguishability. The findings indicate that the recognized work–life balance states are consistently distinguishable across various modeling techniques, suggesting a robust underlying structure. These results offer valuable insights for policymakers and organizations aiming to foster sustainable work practices and enhance wellbeing.