What question did this study set out to answer?

The aim is to create a screening model using machine learning to evaluate depression risk among the elderly in China.

April 3, 2026Open Access

Development of a machine learning-based screening model for the risk of depression among the elderly in China

Key Points

The aim is to create a screening model using machine learning to evaluate depression risk among the elderly in China.
Extracted data from 11,672 records of the China Longitudinal Ageing Social Survey
Developed eight supervised machine learning models after data cleaning and preprocessing
Conducted feature selection based on health status, psychosocial needs, and activities of daily living
Used cross-validation for model training and evaluation with sensitivity and F2 score as priorities
The bagged tree model had the best performance in terms of F2 score and sensitivity
Key factors influencing depression included frequent internet use reducing risk, while poor physical health and social exclusion increased risk
Findings highlighted the inconclusive impact of radio use on depression risk

Abstract

Despite numerous factors being evaluated as risk factors for depression among elderly people, the precise relationship remains inconclusive, partly due to limited large-scale data that allows an advanced analysis of multiple associated factors within a single study. To fill the research gap, this study aimed to develop a machine learning (ML)-based screening model to assess the risk of depression among the elderly population and to identify health-related, psychosocial, and activities of daily living (ADL) factors deemed significant to the model. This retrospective study extracted 11,672 records from the China Longitudinal Ageing Social Survey (CLASS). Eight supervised machine learning models—bagged tree, regularised discriminant analysis (RDA), logistic regression (LR), multivariate adaptive regression splines (MARS), artificial neural network (ANN), random forest (RF), support vector machine (SVM), and extreme gradient boosting (XGBoost); were developed following data cleaning and data pre-processing. Feature selection analysis of predictors related to health status, psychosocial needs, and activities of daily living was conducted using a combination of normalised information gain, gain ratio, and symmetrical uncertainty. Models were trained and evaluated using cross-validation, with model selection prioritising sensitivity and F2 score for depression risk screening. After data cleaning and exclusion of incomplete records, 10,502 records were included in the final analysis. The bagged tree model performed best in terms of F2 score and sensitivity. The five most influential factors associated with depression among the elderly population in China included frequent use of the internet, which was associated with alleviation of depression. In contrast, higher perceived ill physical health, the perception that social changes were unfavourable to the elderly and feeling that one was socially excluded were associated with increased risk of depression. However, the frequency of radio use was inconclusive whether it alleviates or worsens the risk of depression among the elderly. The government may consider enhancing media use (internet), improving physical health, and promoting social involvement among older adults to mitigate the prevalence of depression and safeguard their mental health. However, the study’s findings warrant confirmation in a future longitudinal study.

AIに質問

Bookmark

View Full Paper