Depression Detection from Social Media Comments Using Deep Learning with Interpretability

Depression is one of the most prevalent and serious psychological disorders. If left untreated for a long time, it can lead to serious consequences. Therefore, early intervention can be significantly helpful in this regard. Nowadays, people often express themselves freely on social media, which makes these platforms a very good means for detecting depression among users. Various deep learning models can be used to automate the process of detecting depression from text data as they have proven very effective in text classification tasks. In this work, we have analyzed the performances of deep learning models having RNN layers and that of BERT Base Uncased, which is a pre-trained transformer model, on the Reddit depression dataset available on Kaggle. Results show that the BERT model outperforms the RNN-based deep learning models with accuracy, precision, recall, and f1 score values of 98.02%, 98.06%, 98.02%, and 98.02% respectively. Finally, we analyze the predictions made by the classifier models using the LIME XAI model to gain further insight into the classification mechanism. Implementation of XAI models alongside the deep learning models can assert faith in the predictions made.