COMPARATIVE ANALYSIS OF LSTM, BILSTM, GRU, CNN, AND RNN FOR DEPRESSION DETECTION IN SOCIAL MEDIA

The prevalence of mental health issues and the increasing use of social media provide an opportunity to leverage technology for early detection of depression. This study evaluates and compares five deep learning models, LSTM, BiLSTM, GRU, CNN, and RNN for detecting depressive tendencies from over 10,000 annotated social media messages. These models were trained on preprocessed data using standard techniques, including cleansing, tokenization, and padding. Evaluation metrics such as accuracy, precision, recall, and F1-score were utilized. BiLSTM emerged as the best-performing model with an accuracy of 98.45% and an F1-score of 96.37%, attributed to its bidirectional architecture for contextual analysis. In contrast, CNN achieved high precision (98.55%) but struggled with recall (15.14%), while RNN and GRU exhibited limitations in capturing complex patterns, with GRU showing no measurable performance. These findings establish BiLSTM as a robust tool for mental health monitoring. Future research could explore transformer-based models such as BERT or multilingual datasets for enhanced applicability.