Predicting Lifetime Suicide Attempts in a Community Sample of Adolescents Using Machine Learning Algorithms

Area of Science:

• Public Health
• Adolescent Psychology
• Computational Psychiatry

Background:

• Suicide is a critical global health issue, particularly impacting adolescents.
• Existing suicide prediction research primarily uses clinical or adult populations.
• Early identification of suicide risk factors in community-based adolescent samples is vital for prevention.

Purpose of the Study:

• To compare the predictive accuracy of logistic regression, elastic net regression, and gradient boosting machines for adolescent suicide attempts.
• To identify key variables for screening suicidal behavior in a large adolescent community sample.
• To evaluate the utility of advanced machine learning models in adolescent suicide prediction.

Main Methods:

• Utilized data from the Millennium Cohort Study (N = 7,347) comprising 17-year-olds.
• Combined diverse self- and other-reported variables across multiple categories.
• Employed logistic regression, elastic net regression, and gradient boosting machine algorithms for prediction.

Main Results:

• Machine learning algorithms (elastic net and gradient boosting) significantly outperformed logistic regression.
• Achieved high balanced accuracies: .76 (3 years prior) and .85 (same wave) for predicting suicide attempts.
• Identified essential variables critical for screening adolescent suicidal behavior.

Conclusions:

• Complex machine learning models demonstrate significant potential for accurate adolescent suicide attempt prediction.
• The findings highlight the importance of utilizing community samples and diverse data for effective suicide risk screening in adolescents.
• Identified key risk factors can inform targeted early intervention strategies.