Machine Learning Based Suicide Ideation Prediction for Military Personnel

Area of Science:

• Psychiatry and Mental Health
• Computational Science
• Military Psychology

Background:

• Military personnel experience elevated psychological stress and suicide attempt risk compared to the general population.
• High mental stress is a significant factor contributing to suicide ideation, a precursor to suicide attempts.
• Traditional statistical methods show only moderate correlations between psychological stress and suicide ideation in non-psychiatric individuals.

Purpose of the Study:

• To apply machine learning techniques to predict suicide ideation in military personnel based on psychological stress domains.
• To compare the predictive performance of various machine learning algorithms against conventional criteria.

Main Methods:

• Utilized logistic regression, decision tree, random forest, gradient boosting regression tree, support vector machine, and multilayer perceptron.
• Trained and evaluated models using six key psychological stress domains in military males and females.
• Compared machine learning model performance against the Beck Depression Inventory-II (BDI-II) criterion (BSRS-5 score ≥7).

Main Results:

• All six machine learning methods achieved prediction accuracies exceeding 98%.
• Multilayer perceptron and support vector machine demonstrated near-perfect prediction of suicide ideation.
• Proposed algorithms significantly improved accuracy, sensitivity, specificity, precision, AUC of ROC curve, and AUC of PR curve compared to BSRS-5.

Conclusions:

• Machine learning offers a highly accurate approach for predicting suicide ideation in military populations.
• Advanced algorithms like multilayer perceptron and support vector machine show exceptional promise for early detection and intervention.
• These findings suggest a paradigm shift from traditional statistical methods to AI-driven solutions for mental health risk assessment in military service members.