머신러닝 기반 노인의 거주 태도 유형 예측 모형

초록

This study examines how Aging in Place (AIP) support service utilization intentions affect residential attitudes among Korean elderly, applying machine learning-based prediction models. Using data from the 2023 National Survey of Older Koreans, 9,917 respondents aged 65 and over with valid responses were analyzed. The dependent variable comprised three categories: unconditional ‘AIP preference’, ‘health-conditional AIP preference’, and ‘unconditional non-AIP preference’. Random Forest and XGBoost models, combined with SHAP analysis, were applied to identify non-linear relationships and variable importance. Both ensemble models substantially outperformed the baseline multinomial logistic regression (XGBoost: Accuracy=0.647, Macro F1=0.564, AUC=0.742). SHAP analysis revealed that daily living support service utilization intentions and home visit medical service utilization were the strongest positive predictors of unconditional AIP preference, while willingness to pay for home renovation and daily living support showed strong negative associations with unconditional non-AIP preference. These findings suggest that AIP support services function not merely as welfare supplements but as structural determinants of residential stability, implying the need for policy differentiation by service type and residential preference group.

키워드