Machine learning-based predictive factor analysis of depression among Chinese adolescents

Area of Science:

• Mental Health Research
• Computational Psychiatry
• Global Public Health

Background:

• Adolescent depression is a growing global concern, particularly in China, impacting development and social adaptation.
• Traditional methods struggle with complex factors influencing adolescent mental health.
• Machine learning (ML) offers advanced predictive capabilities for mental health disorders.

Purpose of the Study:

• Compare ML algorithms for classifying depression risk in Chinese adolescents.
• Identify key demographic, personality, and personal growth initiative (PGI) predictors.
• Explore non-linear relationships and interactions among predictive factors.

Main Methods:

• Trained and optimized seven ML algorithms using 5-fold cross-validation on data from 559 adolescents.
• Analyzed feature importance using SHAP values and tested interaction effects via permutation tests.
• Utilized Friedman and Nemenyi tests for model comparison and Youden's J statistic for threshold analysis.

Main Results:

• LightGBM achieved the highest performance (AUC 0.834), accurately classifying depression risk.
• Neuroticism was the strongest predictor, followed by proactive change, agreeableness, extraversion, and growth resilience.
• Significant interactions were found between neuroticism and proactive change, and proactive change and agreeableness.

Conclusions:

• ML, especially LightGBM, effectively identifies adolescent depression risk using personality and PGI factors.
• Findings support integrating multi-dimensional variables for early intervention in adolescent mental health.
• Reducing neuroticism and enhancing proactive growth behaviors may lower depression risk in Chinese adolescents.