Prediction of Subjective Wellbeing and Eudaimonia of Korean Youths and Identification of Their Primary Features Using Tree-Based Machine Learning Algorithms

Background and Purpose

Prior studies have demonstrated that happiness is the main cause and results of certain life domains. Especially for youth, happiness is the protective factor against negative environmental effects and the positive developmental indicator. South Korea, however, scored the lowest in youth happiness among 28 OECD countries. This study aims to find primary domains and features affecting youth happiness, and to compare important factors between the two concepts of happiness: subjective well-being and Eudaimonia, and discuss their implications.

Methods

A nationally representative sample, Youth Happiness Community’s Index (YHCI), was used as input data, collected in 2019 by the National Youth Policy Institute (NYPI) in South Korea. It assessed a wide range of domains regarding the happiness of youths and included measures of Eudaimonia guided by the OECD in 2013. A total of 8,426 youths in grades 4 to 12 participated in the survey, and 7,980 samples were used after deleting missing data.

YHCI 2019 consisted of eight domains: subjective wellbeing, relationships, health, education, safety, participation, activity/leisure, and environments. From the subjective wellbeing domain, two target variables were constructed: subjective wellbeing and Eudaimonia. Subjective wellbeing consisted of life satisfaction, three positive emotions, and three negative emotions. Eudaimonia had six items measuring eudaimonic happiness. 57 features were used to train and predict subjective wellbeing and Eudaimonia ranging from 1 to 5 from the other eight domains. Five tree-based machine learning models including decision tree, random forest, adaptive boosting, gradient boosting, and extreme gradient boosting (XGBoost) were employed to train and predict the two target variables, respectively. Mean squared error was used to measure model performances and feature importance was computed.

Results

Exploratory statistics for subjective wellbeing and Eudaimonia showed that they had very similar distributions and correlations with their feature sets. XGBoost and gradient boosting performed best among all models. Common importance features in both subjective wellbeing and Eudaimonia were School Life Satisfaction, Family Relationship Satisfaction, Health Satisfaction, and Stress Level. Activity Satisfaction, Leisure Time Satisfaction, and Leisure Space Satisfaction were highly ranked in subjective wellbeing. In contrast, Child Rights Recognition, Neighbor Relationship Satisfaction, Academic Achievement, and Neighbor Reliability were highly ranked in Eudaimonia.

Conclusions and Implications

We were able to predict subjective wellbeing and Eudaimonia using 57 features from eight specific domains and discussed how various features affected youth happiness in Korea. We also identified two concepts of happiness correlated and shared important features while finding unique features that determined subjective wellbeing and Eudaimonia respectively. Our results suggest that the policies on youth happiness should explore wide range and diverse dynamics of related domains and should focus on increasing awareness about new factors such as youth activities, leisure, and the rights of children and youth.