Exploring Deep Learning Architectures for Mental Health Assessment: A Comparative Analysis

The major consequences of the covid-19 pandemic have influenced the question of the ability of anger - especially depression and anxiety - to dominate public discourse. The aim of this research is to find advanced methods of understanding, illness and situation, using deep learning methods in the research of depression and stress. Five different models of deep learning are defined and classified in a motivational and anxiety framework based on three different data sets: dataset 1 (data set from student surveys that do not focus), dataset 2 (self-isolation and depression through separation) and dataset 3 (depression and anxiety of significant partners and childbirth). These models are called Neural Networks (NN), Long Short-Term Memory (LSTM), Recurrent Neural Networks (RNN), TabNet, and Convolutional Neural Networks (CNN). The interest is growing in exploring how deep learning is used for analyzing more complex data. Many studies have been done on the analysis of depression and anxiety using machine learning models, mainly for the dual classification. For the case of datasets 1 and 3, predictions with the highest accuracies for anxiety and depression were made by LSTM. Such results for dataset 2 were obtained using CNN. The work opens wide avenues in demonstrating deep learning applied to multiclass classification that could improve diagnosis and provide personalized mental health treatment approaches.