OECD 33개국 행복점수 결정요인의 머신러닝 기반 임계치 탐색 = Exploring Thresholds of Happiness Determinants in 33 OECD Countries Using Machine Learning

다국어 초록 (Multilingual Abstract)

This study empirically identifies key determinants of subjective well-being in 33 of 38 OECD countries and explores their nonlinear thresholds and interaction structures using machine learning and explainable AI (XAI) techniques. Departing from traditional linear regression approaches, it employs a triple ensemble model —XGBoost, Random Forest, and Elastic Net—to predict happiness scores (Ladder Score) using six variables: income (log GDP per capita), social support, healthy life expectancy, freedom of choice, generosity, and perceived corruption.
The analysis, based on SHAP values, partial dependence plots (PDPs), and variable interaction metrics, shows that GDP, social support, and life expectancy are the most influential factors, though their effects plateau beyond certain thresholds. Trust and social capital play a growing role in high-income contexts, and interaction effects like GDP × social support are confirmed. Surrogate trees and simulations help derive practical policy insights.
The study broadens the methodological base of happiness economics and offers evidence-based guidelines for setting quantitative thresholds and strategic goals to address the limits of economic growth on well-being.