Machine learning-based predictive modeling of depressive symptoms in Chinese adolescents.

Area of Science:

• Adolescent mental health research
• Public health and epidemiology
• Machine learning in healthcare

Background:

• Adolescent depressive symptoms pose a significant public health challenge.
• Identifying reliable predictors for early detection is crucial for timely intervention.
• Socioeconomic status and lifestyle indicators are potential factors influencing adolescent mental well-being.

Purpose of the Study:

• To develop and validate prediction models for adolescent depressive symptoms using lifestyle indicators and socioeconomic status.
• To identify and rank the most influential predictors of depressive symptoms in adolescents.
• To explain the relationship between key predictors and the risk of depressive symptoms.

Main Methods:

• A large-scale cross-sectional study involving 32,389 school students (grades 4-12).
• Depressive symptoms identified using the Center for Epidemiologic Studies Depression Scale (CES-D score ≥ 16).
• Boruta-RF algorithm for feature selection and variable importance ranking, followed by Random Forest model construction and Partial Dependence Plots (PDP) for outcome explanation.

Main Results:

• The Boruta-RF algorithm identified self-rated health, sleep duration, parental support for physical exercise, breakfast intake, screen time, and skipping physical education classes as top predictors.
• The Random Forest model achieved a high predictive accuracy with an Area Under the Curve (AUC) of 0.829 (95% CI: 0.820 - 0.837).
• Partial Dependence Plots revealed nonlinear relationships between predictors and the risk of depressive symptoms, offering nuanced insights.

Conclusions:

• A prediction model utilizing routinely collected school lifestyle data can effectively screen adolescents at high risk for depressive symptoms.
• Early detection and mental health evaluation can be facilitated through this accessible screening tool.
• Future research could enhance accuracy by incorporating longitudinal designs, clinical diagnoses, and neuroimaging biomarkers.