A Study on the Determinants of Social Isolation Among the Elderly Living Alone: Focused on Machine Learning-Based Importance Analysis

Key Points

• This study investigates the multidimensional factors influencing social isolation among the elderly living alone, aiming to derive intervention strategies.
• Analyzed data from 1,473 elderly care service users to identify factors affecting isolation.
• Utilized machine learning techniques, specifically XGBoost and SHAP analysis, for variable importance and interaction studies.
• Focused on defining an early warning system for isolation risk based on the quality of social relationships.
• The absence of community networks (friends and neighbors) is identified as the strongest predictor of social isolation.
• Physical visits and availability of help during emergencies are found to significantly contribute to reducing isolation from familial relationships.
• Nonlinear patterns reveal a sharp drop in willingness to engage in social activities as isolation worsens, highlighting structural limitations.

Abstract