Overview
Automated detection of emotional distress and suicide risk indicators in social media posts for early mental health intervention, leveraging TF-IDF vectorization and extensive hyperparameter tuning.
Technical Highlights
- GridSearchCV: Explored 1,760 combinations with 8,800 fits to locate the optimal decision boundaries.
- Final Accuracy: 90.0% using Decision Trees with
criterion='gini',min_samples_split=15, andmax_leaf_nodes=35. - Interactive web demo built with Streamlit.
Tech Stack: Python, Scikit-learn, TF-IDF, Streamlit, Pandas
Last updated on September 16, 2026 at 12:23 PM UTC+7. See Changelog