Bi-LSTM-RF based Hybrid Deep Learning Approach to Identify Level of Depression from Bengali Facebook Status

Depression is a worldwide epidemic and is of special concern due to its impact on mental health and the need for comprehensive studies for better detection and treatment. As the severity of depression grows among the Bengali population, yet the identification is relatively uncommon in this wider community, it appears to be in high demand. In our research, an efficient deep hybrid learning is instituted to identify levels of depression in Bangla text. We gathered 3000 Bengali Facebook posts for our dataset labeled for three levels of depression, mild, moderate, and severe. A bidirectional Long Short-Term Memory (bi-LSTM) architecture and a Random Forest classifier have been applied. We followed some pre-processing steps for these texts, which involved cleaning, correcting mistakes, removing stop words, segmentation, and embedding the text using Word2Vec to turn the text into vectors. The bi-LSTM model performed at an accuracy of 90.5% but this was improved to 93.67% when combined with Random Forest by using Softmax generated class scores as features. The aspects of our proposed hybrid bi-LSTM-RF techniques suggest that it could be more precise and effective for depression detection in the less researched Bangla language. These results provide insight into the application of deep learning with classical classifiers for sentiment classification in the mental health field and enhancing the treatment of depression by early detection.