A Hybrid Machine Learning and Deep Learning Approach for Mental Health Detection using Text Analysis

This paper presents a hybrid framework combining machine learning, ensemble methods, and deep learning for mental health detection from text. A multi-source dataset of over 265,000 records was compiled from Reddit and public repositories. Models including Logistic Regression, SVM, XGBoost, and a CNN+LSTM architecture were trained and evaluated. The Voting Classifier achieved the highest accuracy of 83.68%, with the CNN+LSTM model achieving 78.34%.