Integrating Machine Learning and Sentiment Analysis: A Comparative Study on Mental Health Classification from Social Media Data

The present research paper categorises the different statuses of mental health, employing various techniques of machine learning. It also looks into the emergence of problems due to class imbalance and increases the feature overlapping in any multiclass dataset. The dataset contains 52,681 entries compiled painstakingly from Reddit, Twitter, and Kaggle. There are seven predefined statuses: Normal, Depression, Suicidal, Anxiety, Stress, Bipolar, and Personality Disorder. In the dataset preparation, comprehensive preprocessing techniques involved text cleaning, tokenization, stemming, and TF-IDF vectorization. The following step was to resample and address class imbalance using methods like random over-sampling. Finally, this paper developed, trained, and evaluated three machine learning models: Decision Tree, Logistic Regression, and XGBoost-XGB-with hyperparameter optimization through GridSearchCV. Performance metrics compared for the models are accuracy, precision, recall, F1-score, and confusion matrices. XGBoost performed the best with an accuracy of 82%, outperforming precision and recall in most classes. This was followed by Logistic Regression at 78% and Decision Tree at 67%, which didn't do too well in predicting the minority class. Again, the results show that the ensemble methods, like XGBoost, are powerful when handling these kinds of imbalanced data with complex multiclass problems. Valuable insights provided by this study will help develop intelligent mental health tools, thereby facilitating timely intervention and care. Future studies will employ advanced techniques, such as transformer-based models, multi-modal integrations of data, and domain-specific embeddings, to further improve the accuracy of classification and contextual understanding. This work establishes a robust framework for advancing mental health analysis by machine-learning methods.