Suicidal behaviors among high school graduates with preexisting mental health problems: a machine learning and GIS-based study

Background: Suicidal behavior among adolescents with mental health disorders, such as depression and anxiety, is a critical issue. This study explores the prevalence and predictors of past-year suicidal behaviors among Bangladeshi high school graduates, employing both traditional statistical and machine learning methods. Aims: To investigate the prevalence and predictors of past-year suicidal behaviors among high school graduates with mental health disorders, evaluate the effectiveness of various machine learning models in predicting these behaviors, and identify geographical disparities. Methods: A cross-sectional survey was conducted with 1,242 high school graduates (54.1% female) in June 2023, collecting data on sociodemographic characteristics, mental health status, sleep patterns, and digital addiction. Statistical analyses were performed using SPSS, while machine learning and GIS analyses were conducted with Python and ArcMap 10.8, respectively. Results: Among the participants, 29.9% reported suicidal ideation, 15.3% had suicide plans, and 5.4% attempted suicide in the past year. Significant predictors included rural residence, sleep duration, comorbid depression and anxiety, and digital addiction. Machine learning analyses revealed that permanent residence was the most significant predictor of suicidal behavior, while digital addiction had the least impact. Among the models used, the CatBoost model achieved the highest accuracy (69.42% for ideation, 87.05% for planning, and 94.77% for attempts) and demonstrated superior predictive performance. Geographical analysis showed higher rates of suicidal behaviors in specific districts, though overall disparities were not statistically significant. Conclusion: Enhancing mental health services in rural areas, addressing sleep issues, and implementing digital health and community awareness programs are crucial for reducing suicidal behavior. Future longitudinal research is needed to better understand these factors and develop more effective prevention strategies.