A hybrid self-supervised model predicting life satisfaction in South Korea

Area of Science:

• Computational Social Science
• Psychological Wellbeing Research
• Artificial Intelligence in Social Sciences

Background:

• Life satisfaction is a key cognitive indicator of subjective wellbeing and overall life quality.
• Accurate prediction of life satisfaction is crucial for understanding societal and individual welfare.
• Existing models may lack interpretability, hindering practical application in social sciences.

Purpose of the Study:

• To develop a hybrid self-supervised model for predicting life satisfaction in South Korea.
• To enhance the interpretability of AI models used in social science research.
• To provide a transparent AI tool for professionals without data analytics expertise.

Main Methods:

• Utilized the 2021 Busan Metropolitan City Social Survey Data (32,390 individuals, 51 variables).
• Developed a self-supervised pre-training TabNet model.
• Integrated TabNet with the Local Interpretable Model-agnostic Explanation (LIME) technique for enhanced interpretability.

Main Results:

• The hybrid model achieved an AUC of 0.7778 (training) and 0.7757 (test), outperforming conventional tree-based ML models.
• The integration with LIME significantly improved the interpretability of local model behavior.
• The model offers a clear understanding of AI decision-making processes.

Conclusions:

• The proposed hybrid model provides a transparent and accurate method for predicting life satisfaction.
• This approach democratizes the use of AI in social sciences and psychology by simplifying model interpretation.
• The findings support the utility of interpretable AI in understanding complex human wellbeing indicators.