Predicting future suicidal behaviour in young adults, with different machine learning techniques: A population-based...

Area of Science:

• Psychiatry
• Data Science
• Public Health

Background:

• Predictive accuracy for suicidal behavior has stagnated for decades.
• Machine learning (ML) offers potential for improved prediction using longitudinal data.

Purpose of the Study:

• To explore ML's potential in predicting suicidal behavior.
• To compare ML algorithms against traditional methods using population-based data.

Main Methods:

• Utilized data from 3508 young adults (18-34 years) in the Scottish Wellbeing Study.
• Assessed psychological measures at baseline to predict suicide ideation and attempts at one-year follow-up.
• Compared regular logistic regression, K-nearest neighbors, classification tree, random forests, gradient boosting, and support vector machine algorithms.

Main Results:

• At follow-up, 336 (14%) reported suicide ideation and 50 (2%) reported suicide attempts.
• Random forest showed best prediction for ideation (AUC 0.83), gradient boosting for attempts (AUC 0.80).
• Performance metrics were similar across all tested ML algorithms.

Conclusions:

• ML techniques did not significantly improve suicidal behavior prediction accuracy with psychological data alone.
• Incorporating broader data (employment, education, healthcare) could enhance ML performance.
• Further research is needed to optimize ML for suicide risk prediction.