Machine Learning Approaches for Predicting Suicidal Behaviors among University Students in Bangladesh during the COVID-19 Pandemic

Abstract Background Psychological and behavioral stresses increased enormously during the global COVID-19 pandemic. This study intends to identify the best machine learning model to forecast suicide risk among university students in Bangladesh. Methodology: An anonymous online survey utilizing DASS-21 and Insomnia Severity Index (ISI) to assess depression, anxiety, and stress levels; Suicidal Behaviors Questionnaire-Revised (SBQ-R) to identify suicidal risk was conducted from 1 to 30 June 2022. We compared six popular machine learning models (MLM), including Logistic Regression (LR), Support Vector Machine (SVM), Naïve Bayes (NB), Classification Tree (CT), and Random Forest (RF), to identify the most efficient predictive model for suicidal behavior through several metrics such as accuracy, Kappa, and receiver operating characteristic curve (ROC). Result Determinants predicting suicidal behavior include depression, insomnia, anxiety, and stress. Besides, sex, relationship status, family income, loss of jobs, and death within the family from COVID-19 are crucial suicidal risk predictors. The performance evaluation and comparison of MLM show that all models behaved consistently and were comparable in predicting suicidal risk determinants since the ranges were for accuracy (0.76 to 0.79); Kappa (0.52 to 0.59); ROC (0.76 to 0.89); sensitivity (0.76 to 0.81), and specificity (0.72 to 0.82). SVM was the best and most consistent performing model among all MLM in terms of accuracy (79%), Kappa (0.59), ROC (0.89), sensitivity (0.81), and specificity (0.81). Conclusion SVM is the best MLM in identifying predictors of suicidal risk among university students to develop a screening tool that can guide policymakers and universities in designing appropriate, timely suicide prevention interventions.