Predicting survival factor following suicide attempt in Iran: an ensemble machine learning technique

Area of Science:

• Public Health
• Mental Health Research
• Machine Learning Applications

Background:

• Suicide is a growing public health concern in Iran, necessitating improved predictive models.
• Traditional statistical methods may not suffice for individualized suicide risk assessment.
• Ensemble machine learning (ML) techniques offer enhanced accuracy for survival prediction.

Purpose of the Study:

• To apply ensemble ML techniques to predict survival after suicide attempts in Iran.
• To identify critical factors influencing survival post-suicide attempt.
• To improve the accuracy of suicide survival prediction models.

Main Methods:

• Utilized a longitudinal dataset (2017-2024) encompassing demographic, psychological, economic, and social factors.
• Applied ensemble ML algorithms: AdaBoostM1, J48 pruned tree, Bagging, LogitBoost, MultiBoostAB, J48, SVM, LibLINEAR, and Multilayer Perceptron.
• Determined critical survival factors through comparative analysis of ML model performance.

Main Results:

• LogitBoost ensemble models achieved the highest accuracy (94.3%), with J48 algorithm close behind (93.6%).
• Timing of hospital admission was identified as the most influential survival factor.
• The types of drugs used during the suicide attempt were the second most significant factor.

Conclusions:

• Ensemble ML techniques show significant promise for enhancing mental health research and suicide prediction.
• Findings provide crucial insights into factors influencing survival after suicide attempts in Iran.
• Results can inform personalized interventions to support high-risk populations.