A MACHINE LEARNING FRAMEWORK FOR SUICIDAL THOUGHTS PREDICTION USING LOGISTIC REGRESSION AND SMOTE ALGORITHM

Abstract

Suicide, a global health challenge identified in Goal 3 of the global agenda for enhancing worldwide well-being, demands urgent attention. This study focused on predicting suicidal thoughts using machine learning, leveraging the 2021 National Women's Life Experience Survey (SPHPN) involving women aged 15 to 64. Analyzing 11,305 ever-married women, 504 (4.5%) reported experiencing suicidal thoughts. The outcome variable was binary (1 for suicidal thoughts, 0 for none). The study used seven predictors: age, education level, residence type, physical and sexual violence, smoking frequency, alcohol consumption, and depression. Ordinary logistic regression and SMOTE-based logistic regression were applied. The former identified physical violence, depression, and sexual violence as crucial factors, while the latter emphasized physical violence, sexual violence, and age. In cases of class imbalance, the SMOTE-enhanced model exhibited improved performance in terms of sensitivity, false positive rate, balanced accuracy, and Kappa statistic, with lower standard errors of parameter estimates. The findings highlight the importance of addressing violence and mental health in policies aimed at reducing suicidal thoughts among women. Policymakers can use these insights to develop targeted interventions, and healthcare providers can identify high-risk individuals for timely interventions. Community programs and public health campaigns should promote mental well-being and prevent suicidal behaviors using these findings. Future research should include more predictors, diverse populations, and longitudinal data to better understand causal relationships and timing. Interdisciplinary collaboration and advanced machine learning techniques can enhance predictive accuracy and model interpretability.