Enhancing Mental Health Detection via Machine Learning: A Sentiment Analysis Technique Utilizing Social Media Data

The rise in the use of social media allows for new avenues for the monitoring of mental health, the automation of analysis requires methods due to the large quantum of data. This paper will review the application of sentiment analysis to detect and classify mental health conditions using data from social media. Three training machine learning models include Bernoulli Naive Bayes, Logistic Regression, and XGBoost (XGB) on wholesome data over seven mental health statuses: normal, depression, suicidal, anxiety, stress, bipolar, and personality disorder. The recent one seems to be the most performing, the XGB Classifier, with an accuracy of about 80%, way more performant than the other tested models when it comes to critical conditions like suicidal thoughts and depression. Then logistic regression significantly follows with an accuracy of 76.54%, and Bernoulli Naive Bayes is the last one, with an accuracy of 64.06%. This paper finally points out that machine learning has the potential to contribute much to sentiment-based mental health analysis and offer a scalable, non-intrusive method for monitoring emotional well-being via social media. Future works may be related to improving feature extraction and dealing with cultural variations in the expression of emotions.