Machine learning-based prediction of depression levels: developing a model for male and female senior citizens

서지정보

· 발행기관 : 연세대학교 사회복지연구소
· 수록지 정보 : 한국사회복지조사연구 / 70권 / 145 ~ 172페이지
· 저자명 : 홍기혜

초록

This study aims to develop a predictive model for individuals’ depression levels without using depression assessment scales. It will instead use a gradient boosting machine learning algorithm. This study will also analyze the predictive factors of depression by gender and suggests directions of intervention for depression in senior citizens. Data from the ‘Korean National Survey on the Elderly’ were used in this study. The participants were 12,544 elderly males and 18,425 elderly females. This study set 56 factors as explanatory variables based on stress-coping theory for the variables verified in previous studies, estimated a predictive model, and analyzed predictors by gender. The model performance of elderly males and females was evaluated, respectively, using six classification performance metrics: Accuracy was 76.0% and 73.9%. Recall was 60.7% and 73.3%. Specificity was 85.3% and 74.4%. Precision was 71.5% and 73.3. F1-score was 65.7% and 73.6%. ROC-AUC-score was 82.7% and 82.1%. Day-to-day health and small social activities were important in developing a predictive model for individuals’ depression levels. The predictors that have shown notable gender differences were marital relationships and exercise. This study demonstrated that it could be possible to predict geriatric depression with the factors that were recognizable to people close to the elderly. This predictive model can be used to identifying at-risk elderly individuals in the social welfare sector. This study is also meaningful in that 56 factors contributing to the prediction of depression levels were analyzed by gender. This supports an integrated perspective and provides gender-specific and gender-common evidence to prevent or reduce depression among senior citizens.

참고자료

· 없음