Predicting and explaining life satisfaction among older adults using tree-based ensemble models and SHAP: Evidence...

As digital transformation continues to penetrate various sectors of society, the issue of the digital divide has become increasingly prominent. Against the backdrop of accelerating population aging, the barriers that older adults face in accessing and utilizing digital information have exerted a profound impact on their quality of life. This study employs tree-based ensemble learning algorithms to predict and identify the key factors of the digital divide that influence life satisfaction among older adults. It also evaluates the predictive performance of these models, thereby providing interpretive insights into the impact of the digital divide on subjective well-being. Using original data from the '2023 Report on Digital Information Divide Survey' conducted by the National Information Society Agency of South Korea, this study constructs an analytical framework that integrates both predictive capability and interpretability. First, the XGBoost model is employed to conduct feature importance analysis, identifying 15 key variables that are highly influential in predicting life satisfaction. These variables are further examined using the SHAP method to provide interpretive insights into their contributions. Subsequently, multiple tree-based ensemble learning algorithms-including Random Forest, XGBoost, LightGBM, and CatBoost-are applied to compare their predictive performance. The results indicate that variables related to technological self-efficacy, digital information literacy, social capital, experience and perception of AI services, and household monthly income are significant predictors of life satisfaction among older adults. Among the models tested, CatBoost demonstrates superior overall predictive accuracy, suggesting its effectiveness in forecasting life satisfaction in this demographic. This study expands the application of machine learning in areas such as aging research and the digital divide and proves the effectiveness of ensemble learning algorithms in predicting digital divide factors that affect the life satisfaction of older adults. This approach provides a novel and powerful methodological for addressing complex social problems. Moreover, the study uncovers the structural configuration of key digital information factors associated with life satisfaction, offering data-driven insights into the mechanisms through which the digital divide influences well-being. These results have practical implications for enhancing digital inclusion, improving adaptability among older adults, and fostering a stronger sense of participation and happiness in digital society.