Predicting mental health problems in adolescence using machine learning techniques.

Area of Science:

• Child and Adolescent Psychiatry
• Psychometrics
• Computational Psychiatry

Background:

• Early identification of adolescent mental health issues is vital for timely intervention.
• No current model screens the general population for mental health risk.
• Machine learning offers potential improvements over traditional prediction models like logistic regression.

Purpose of the Study:

• Develop a predictive model for mid-adolescence mental health problems.
• Compare the efficacy of machine learning techniques against logistic regression.

Main Methods:

• Utilized data from 7,638 twins in the Child and Adolescent Twin Study in Sweden.
• Employed 474 predictors from parental reports and register data.
• Assessed mental health problems using the Strengths and Difficulties Questionnaire; model performance measured by AUC.

Main Results:

• Random forest (AUC=0.739) and support vector machines (AUC=0.735) showed comparable performance.
• Confidence intervals overlapped, indicating no significant superiority of machine learning models over logistic regression.
• Top models were not suitable for immediate clinical application.

Conclusions:

• The study provides foundational work for future mental health prediction models.
• Parent-rated assessments are recommended for future research.
• Complex machine learning methods may not offer significant advantages over logistic regression for this task.