Apply Machine Learning to Predict Risk for Adolescent Depression in a Cohort of Kenyan Adolescents

Area of Science:

• Mental Health Research
• Computational Psychiatry
• Global Health

Background:

• Adolescent depression is a significant public health issue in low- and middle-income countries (LMICs).
• Effective prevention strategies require identifying key risk factors for depression.
• Machine learning (ML) offers advanced analytical capabilities for complex health data.

Purpose of the Study:

• To apply machine learning (ML) techniques to identify risk factors for adolescent depression.
• To enhance the design of adolescent depression prevention programs in LMICs.
• To evaluate the predictive accuracy of ML models using multidomain risk factors.

Main Methods:

• Six ML approaches, including random forests, were employed.
• Data from 269 adolescents in Kenya (2024-2025) were analyzed.
• Childhood adversity, home adversity, adolescent stress, and school adversity experiences were used to predict depression (PHQ9-A scores).

Main Results:

• ML proved effective for early identification of adolescents at risk for depression.
• The random forest model demonstrated superior performance, especially with multidomain risk data.
• Predictive accuracy reached 75.1% using the top 20 risk factors, comparable to using all 49 factors (78.3%).
• Childhood or home adversity alone were weak predictors; adding adolescent and community stressors improved prediction significantly.

Conclusions:

• ML and predictive modeling can revolutionize preventive mental health care by leveraging multidomain data for early risk identification.
• ML findings can inform the development of tailored interventions and optimize resource allocation in low-resource settings.
• This study demonstrates a promising approach for LMICs to address adolescent depression prevention.