A Hybrid Deep Learning Model for Predicting Depression Symptoms From Large-Scale Textual Dataset

A significant number of individuals are facing mental health issues due to a lack of timely treatment and support for detecting depression. This lack of early treatment is a primary factor contributing to conditions such as anxiety disorders, bipolar disorders, sleep disorders, depression, and, in severe cases, self-harm and suicide. Consequently, identifying individuals suffering from mental health disorders and offering prompt intervention is an extraordinarily challenging task. Therefore, this research introduced a novel hybrid deep-learning method for predicting depression at an early stage. In this study, we proposed a hybrid deep learning model for depression prediction, which mainly combines a Convolution Neural Network (CNN) and a Long Short-Term Memory (LSTM) model. An enhanced version of the LSTM approach, namely Two-State LSTM (TS-LSTM), is applied based on the feature attention mechanism. The proposed framework incorporates a feature attention mechanism into the TS-LSTM approach, which increases the ability to identify relationships and extract keywords for depression detection using the attention layer. This methodology is employed on a large dataset obtained from a publicly accessible online platform for young people. This dataset consists of text questions asked by young users on the platform. We extracted features through a one-hot encoding method from robust indicators of potential depression symptoms, which were predefined by medical and psychological experts. In comparative evaluations compared to conventional approaches, our system demonstrates superior performance. The experimental outcomes revealed that the proposed approach attained an accuracy of 97.23%, a precision of 98.57%, a recall of 97.13%, an F1-score of 97.84%, and a specificity of 97.93%, respectively. These results highlight the efficiency of the developed methodology that accurately predicts depression.