1Student, School of Computer Science and Engineering, Lovely Professional University, Phagwara, Punjab, India. Email: vvnssrnath2002@gmail.com
2Assistant Professor, School of Computer Science and Engineering, Lovely Professional University, Phagwara, Punjab, India. Email: tejinder.15312@lpu.co.in
3Assistant Professor, School of Computer Applications, Lovely Professional University, Phagwara, Punjab, India. Email: uminder.23377@lpu.co.in
Background: The rapid increase in the social media platforms has promoted the need to have smart systems that could identify suicidal thoughts in real-time. The study is aimed at automated suicide risk detection on the dataset Suicide Post Detection on Social Media Articles published by Google. Our machine learning and deep learning techniques varied and included Logistic Regression, Support Vector Machine (SVM), and ensemble techniques, including Stacking and Voting Classifiers. Deep learning architectures such as Convolutional Neural Network (CNN), Bidirectional Long Short-Term Memory (Bi-LSTM), and CNN-LSTM hybrid, and finely tuned transformer-based models such as RoBERTa and ELECTRA were also used.
Results: The ELECTRA model performed far better compared to the rest with an accuracy of 95.52 and a global F1-score of 95.53. To simplify the interpretation of the results, explainable AI systems such as LIME and SHAP were applied in order to highlight specific characteristics of the language that influenced the selection of models.
Conclusion: This practice will combine transparency and predictive intelligence, thus helping to identify high-risk posts effectively and useful assistance in the early intervention of suicide by analyzing text in real-time.
Index Terms - suicide detection, social media analysis, deep learning, transformer models, ELECTRA, RoBERTa, explainable AI, mental health monitoring.
How to cite this article: Nath VVNSSR, Thind T, Kaur U. AI Against Suicide: Real – Time Risk Detection from Social Media Texts. Int J Drug Deliv Technol. 2026;16(13s): 137-144. DOI: 10.25258/ijddt.16.13s.14
Source of support: Nil.
Conflict of interest: None