Early Detection of Mental Health over Social Media Using Machine Learning

Abstract

In this project, we have worked on the use of machine learning algorithms that may indicate the mental state or sentiment of the users of social media. The current linguistic research on tweets or statuses shows distinct language patterns talks much about the sentiments of the users. As such, we are encouraged to use the available social media dataset repository for performance analysis of the supervised learning models generally applied in the case of large datasets. We have captured the texts and preprocessed the text data using Python tools for feature selection that provides clues to sentiment and subsequently, the subjective sentences are classified as positive or negative. We built machine learning (ML) models out of the classified text features for random forest and LSTM algorithms. The accuracy of the classification in both cases is encouraging, particularly the LSTM algorithm. The whole work is carried out using rich libraries of Python machine language modules. A prototype web application is developed with a front end (input and output) and back end (ML models and classification) using a rapid prototype web application framework for testing the algorithms that enable us to check the mental status of a social media user to determine if they may require medical or emotional support. Social media emerges as the most critical source of information, reflecting people's expressions and communication through textual content. This work contributes to the application of machine learning to further enhance the early detection of the mental health of users.