Effective Techniques for Depression Detection on Social Media: A Comprehensive Review

Currently depression has become one of the common and major public health problems in our society due to changes in lifestyle, high work pressure, persistent illness, loss of beloved ones in life, social and economic crisis due to pandemics, etc. Doctors have traditionally emphasized face-to-face interaction to diagnose depression using clinical depression criteria. Not every patient (less than 30^%) would seek medical care in the early stages of depression, and moreover many of them would not accept that they have a mental disorder, leading to delays in remedial measures, and deterioration of symptoms. Failure to detect depression in the early stages has led to suicide in many cases. To address this problem, and take advantage of the abundance of data, machine learning has been used significantly for depression detection in social networks. A variety of machine learning models, features, and media have been used in this area. After data pre-processing, Feature extraction techniques are used to extract psychological factors, Linguistic factors, and emotional factors from Text (LIWC, LDA, word embedding), Image properties, including saturation, color theme, brightness, and color clarity, and transcript text of Audio. Most of the classification algorithms used are supervised learning(SL) algorithms like Support Vector Machines(SVM), Decision Tree(DT), Naïve Bayes(NB), Random Forest(RF), and Neural Networks(NN). The popular metrics considered are ERDE, Precision, Recall, F1-Score, Accuracy, $\boldsymbol{\mathrm{F}_{\text{latency}}}$. This paper presents a study of research for the early identification of depression in social media using machine learning (ML) techniques.