(A) review on machine learning applications for suicide monitoring, prevention and suicide note analysis techniques

Abstract

Suicide is a multifaceted phenomenon influenced by a range of social, psychological, and environmental factors. To comprehensively analyze these factors, initially suicide notes were utilized to uncover the psychosocial process in the decision to commit suicide. However recently, machine learning (ML) techniques have been increasingly employed for the identification and prediction of suicidal thoughts and behaviors (STBs). While ML holds significant promise, concerns have been raised regarding the potential overestimation of prediction performance in ML models and the limited empirical contributions of ML-focused studies to suicide research. Moreover, existing research predominantly emphasizes the predictive accuracy of ML algorithms, often neglecting the practical considerations necessary for real-world implementation. This review aims to critically examine notable ML studies in suicide research, identify key methodological strengths and limitations, and propose actionable pathways to enhance the practical utility of ML models. Our findings highlight the importance of data quality and type in shaping ML applications, address critical issues specific to various ML methodologies, and outline four key considerations for the effective integration of ML into suicide prevention efforts. Ultimately, we advocate for a paradigm shift from a prediction-centric approach to a feasibility-oriented framework, paving the way for the responsible and impactful use of ML techniques in suicide prediction and monitoring.