Predicting Depression with Social Media Images

This study deliberated on the work of stress sight by means of analyzing pix related to social media customers. We created a database containing 485,121 images from 398 volunteer profiles that extended admittance to their data on the popular Russian-speaking social media platform Vkontakte. The results of the stress questionnaire were used to classify stress and control groups as well as to set the binary separation function. We have identified 3 types of user photos: profile pictures, photos from posts, and albums. We used object acquisition methods to retrieve object elements that determine the presence of 80 different object categories in user images. For the purpose of the task, different machine learning algorithms are trained in color objects and features. Our models reached a 65.5% F1-score with the task of exposing depressed users.