Identifying the most crucial factors associated with depression based on interpretable machine learning: a case study...

Area of Science:

• Gerontology
• Mental Health Research
• Computational Social Science

Background:

• Depression is a significant mental health issue in China's older adult population.
• Current methods for predicting depression risk and identifying its causes are limited.
• Accurate prediction and factor identification are crucial for effective depression control and risk reduction.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting depression risk in middle-aged and older adults in China.
• To identify key factors influencing depression using interpretable machine learning techniques.
• To understand the non-linear associations between identified factors and depression risk.

Main Methods:

• Utilized data from 25,586 samples from the harmonized China Health and Retirement Longitudinal Study (CHARLS).
• Compared five machine learning models (CatBoost, XGBoost, GBDT, RF, LightGBM) against Linear Regression.
• Employed SHapley Additive exPlanations (SHAP) for factor identification and Accumulated Local Effects (ALE) for non-linear relationship analysis.

Main Results:

• The CatBoost model demonstrated superior performance across multiple evaluation metrics (MAE, MSE, MedAE, R²).
• Top predictive factors identified were life satisfaction (r4satlife), self-reported memory (r4slfmem), and health status (r4shlta).
• SHAP and ALE analyses provided global and instance-level insights into depression risk factors.

Conclusions:

• The CatBoost model is suitable for predicting depression in the studied population.
• Interpretable methods (SHAP, ALE) successfully identified critical depression determinants.
• Findings highlight the importance of life satisfaction, memory, and health status in the context of late-life depression.