Predicting life satisfaction using machine learning and explainable AI

Machine learning can predict life satisfaction with high accuracy (93.80%), suggesting that computational methods may eventually complement or improve upon traditional survey-based approaches for measuring well-being, though the practical implications of this accuracy level for real-world interventions remain unclear.

Only 27 key questions were needed to achieve strong predictive performance, down from a larger original survey, indicating that life satisfaction assessments could potentially be simplified and made more accessible without sacrificing measurement quality—important for reducing respondent burden in mental health research.

Different age groups show varying relationships with life satisfaction determinants, as revealed through age-bracket analysis, suggesting that one-size-fits-all interventions for well-being may be less effective than age-tailored approaches informed by machine learning insights.

The study prioritized explainability and reproducibility by using feature learning techniques and identifying specific significant questions, demonstrating responsible AI practice by ensuring that predictions aren't "black boxes"—a critical consideration when algorithms inform mental health decisions affecting real people.

Biomedical language models outperformed clinical ones for this task (73.21% vs. lower clinical performance), highlighting that AI model selection matters significantly and that domain-specific training data influences which tools work best for sensitive applications like mental health assessment.