Crossing the digital threshold: An interpretable machine learning model of life satisfaction for older adults in...

Under the dual influences of aging and digitization, it is of great significance to reveal the key factors that affect the life satisfaction of older adults in a digital society. This study investigates the determinants of life satisfaction by analyzing survey data from 1,102 older adults in China. Four machine learning algorithms (Regularized Logistic Regression, Random Forest, XGBoost, and Support Vector Machine) were systematically evaluated to identify the most effective predictive model. The XGBoost model demonstrated superior performance. An Interpretable Machine Learning (IML) framework (XGBoost-SHAP-PDP) was then employed to move beyond prediction and explain how these factors operate. The results showed that the feature importance analysis included economic situation, digital competence, digital self-efficacy, and use intensity as the main predictors. The SHAP analysis revealed a significant asymmetry: only digital competence is positively rewarding, while use intensity functions primarily as a one-way penalty for low use, offering no significant positive impact. Furthermore, the PDP analysis identified a non-linear zero-to-activation dynamic, showing that the benefits of digital competence do not appear until a specific proficiency threshold is attained. These findings provide quantitative refinements to digital divide theory, demonstrating that the quality of digital engagement, not the quantity, is the key determinant of life satisfaction. It is necessary to shift from promoting basic access toward intensive training programs designed to help older adults cross this functional competence threshold.