Related Experiment Video
Updated: Aug 5, 2026
12:18
A Machine Learning Approach to Design an Efficient Selective Screening of Mild Cognitive Impairment
Published on: January 11, 2020
Predicting Care Needs in Community-Dwelling Older Adults Using Explainable Machine Learning and a Multidimensional
1AgeTech-Service Convergence Major, Department of Gerontology, Kyung Hee University, 1732, Deogyeong-daero, Giheung-gu, Yongin-si, Gyeonggi-do, 17104, Republic of Korea, 82 312012940.
JMIR Aging
|July 31, 2026
Summary
Machine learning models can accurately identify older adults needing care using non-functional factors like health and social determinants. This approach supports proactive interventions and integrated community care policies for aging populations.
Area of Science:
- Gerontology and Artificial Intelligence
- Public Health and Social Policy
Background:
- Addressing the growing need for elder care due to population aging and workforce shortages.
- Highlighting the limitations of existing research focusing on isolated factors rather than integrated care needs.
Purpose of the Study:
- To develop and interpret an explainable machine learning (ML) model for identifying care needs in community-dwelling Korean older adults.
- To integrate physical health, mental health, cognitive function, health behaviors, and socioenvironmental factors for a comprehensive care needs assessment.
Main Methods:
- Utilized data from the 2023 Korea Senior Survey (n=10,078).
- Compared seven ML algorithms using stratified 5-fold cross-validation.
- Assessed model discrimination with Area Under the Curve (AUC) and interpretability with Shapley Additive Explanations (SHAP).
Main Results:
- Model A (excluding Activities of Daily Living/Instrumental Activities of Daily Living) showed strong predictive accuracy (AUC 0.892) using upstream factors.
- Key predictors included age, nutritional risk, employment, depressive symptoms, self-rated health, cognitive function, income, and home modification.
- Model B (including ADL/IADL) showed higher AUC but was influenced by the definition of care needs; an objective outcome measure yielded similar discrimination to Model A.
Conclusions:
- Explainable ML models effectively identify older adults' care needs with high accuracy and interpretability.
- Multidimensional non-functional factors are sufficient for identifying care needs, supporting proactive interventions.
- Findings provide evidence for improving the Long-Term Care Insurance system and community care policies.
