Machine Learning Applications in Suicide Prediction and Prevention: A Narrative Review

Abstract

Background: Suicide is a complex and preventable public health issue where traditional statistical techniques have shown limited effectiveness in predicting future suicide deaths. Machine learning offers promising approaches to identify complex patterns and improve prediction accuracy.

Methods: This narrative review examined the application of machine learning in suicide prediction by searching academic databases (PubMed, CINAHL Plus, IEEE Xplore) using MeSH terms 'Machine Learning' and 'Suicide.' English-language articles published within the last five years focusing on suicide, suicide deaths, and prevention were included. The final selection comprised 18 articles after removing duplicates.

Results: Key risk factors identified included mental health conditions (particularly depression), socioeconomic factors (unemployment and financial difficulties), family-related issues, and demographic characteristics (age, gender). Various machine learning approaches demonstrated effectiveness in predicting suicide risk. K-Nearest Neighbors and ensemble models (combining Random Forest and XGBoost) showed particularly strong performance. Time series models like ARIMA variants excelled at temporal predictions, while ensemble methods demonstrated versatility with multiple data sources.

Conclusion: Machine learning techniques offer substantial improvements over traditional approaches for suicide prediction, with model selection dependent on data availability, geographical scale, and temporal requirements. Ensemble methods perform best with multiple data sources, while time series models excel with temporal data.