Multi-Label Classification of Suicidal Thoughts

Abstract

Teenage suicide rates continue to rise and given teenagers' increased communication via digital platforms, traditional identification techniques can overlook subtle indicators of distress. This study uses artificial intelligence to examine social media text data for early identification of teenage suicidal thoughts. Using multi-label classification, we optimized the BERT (Bidirectional Encoder Representations from Transformers) model to handle complex cases where people may display several risk variables concurrently. Filtered for relevance to student life, the dataset consisted in student-generated content from Reddit's SuicideWatch and Humor subreddits, tagged across academic performance, bereavement, psychological problems, family issues, and past suicide attempts. Particularly in mental illnesses (F1 = 0.95) and family issues (F1 = 0.91), the model demonstrated great detection ability, high precision, recall, and F1-scores. This method presents a rich analysis that could improve early intervention and support for teenage mental health problems by emphasizing the possibilities of integrating machine learning into real-time monitoring systems.