Explainable machine learning models predicting the risk of social isolation in older adults: a prospective cohort study
Abstract Introduction This study aimed to develop a machine learning system to predict social isolation risk in older adults. Methods Data from a sample of 6588 older adults in China were analyzed using information from China Health and Retirement Longitudinal Study from 2015 to 2018. We employed th...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
BMC
2025-05-01
|
| Series: | BMC Public Health |
| Subjects: | |
| Online Access: | https://doi.org/10.1186/s12889-025-23108-1 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Abstract Introduction This study aimed to develop a machine learning system to predict social isolation risk in older adults. Methods Data from a sample of 6588 older adults in China were analyzed using information from China Health and Retirement Longitudinal Study from 2015 to 2018. We employed the light gradient boosting machine (Lightgbm) algorithm to determine the most common predictors of social isolation among older adults. After identifying these predictors, we trained and optimized 7 models to predict the risk of social isolation among older adults: Lightgbm, logistic regression, decision tree, support vector machine, random forest, gradient boosting decision tree (Gbdt), and Xgboost. In addition, the Shapely additive explanation (SHAP) method was used to show the contribution of each social isolation predictor to the prediction. Statistical analysis was conducted from December 2023 to April 2024. Results The Gbdt model had the best performance with an accuracy of 0.7247, sensitivity of 0.9207, specificity of 0.6273, F1 score of 0.6894, and Area Under Curve of 0.84. In addition, the SHAP method demonstrated that intergeneration financial support, child visits, age, left-hand grip strength, and loneliness were the most important characteristics. Conclusions The combination of Gbdt and SHAP provides a clear explanation of the factors contributing to predicting the personalized risk of social isolation for individuals and an intuitive understanding of the impact of key features. |
|---|---|
| ISSN: | 1471-2458 |