Understanding Mental Health through Sentiment Analysis: using Machine Learning to Analyze Social Media for Emotional Insights

This paper examines the use of sentiment analysis to detect mental health conditions from social media posts using machine learning techniques. A dataset of over 53,000 labelled posts was processed, employing logistic regression for classification. Feature extraction was conducted through TF- IDF and sentiment polarity analysis. The model yields an overall accuracy of 87% in classifying conditions such as Anxiety, Bipolar, and Normal. On the other hand, this model has failed in the face of class imbalance and the threat of closely related categories such as Suicidal, Anxiety, and Depression since the sentiments are often too similar. Moreover, it was a big challenge for specification to mark those sarcastic and ambiguous statements in social network posts. The challenges in feature engineering, class imbalance, and the efficiency of the models used mean that future works should target overcoming these challenges by improving feature engineering, handling class imbalances, and using more sophisticated models such as transformers, with the ultimate aim of achieving real-time monitoring of mental health.