Predicting Mental Health Crises Using Social Media Sentiment Analysis: A Machine Learning Approach

Mental illnesses present lots of problems with regards to individuals and the general population hence require early identification and management. In light of the fact that more people are initiating accounts on these sites, social media is a useful tool for screening for signs of mental distress using word choice and attitudes. This research aims at examining how behavioural data analysed using Natural Language Processing and machine learning could be used to predict mental health crises. Mentally III or NOT? In this study, the authors aim to classify a user's textual content in the context of mental health representing more than 27,000 labeled social media messages. Some of the activities used in data preprocessing included text cleaning, stemming and tokenization were used prior to data analysis. The feature extraction step performed include using a Term Frequency-Inverse Document Frequency (TF-IDF) vectorizer. Three Machine learning algorithms namely Support Vector Classifier, Random Forest Classifier and Ada Boost Classifier were tested on the dataset. Of them, the largest accuracy of 92% was observed for the SVC, which indicates better performance of this method in identifying mental health-related issues. This research focuses on applying approaches of NLP integrated with supervised classification for the automated detection of mental health crisis. This study implies that technology should form a large part of the solution to the expanding problem of mental illness by drawing information about mental health from social media platforms.