FROM TEXT TO THERAPY: A STUDY ON SENTIMENT ANALYSIS IN MENTAL HEALTH

Mental health prediction using social media data has become a crucial research area due to the rising incidence of psychological diseases worldwide.This paper presents a comprehensive review of current methodologies for multiclass sentiment classification applied to mental health analysis, with a focus on ensemble and hybrid models.A diverse set of deep learning, machine learning, and transformer-based methods are analyzed across recent literature, highlighting their capabilities in detecting disorders such as depression, anxiety, and stress.Special emphasis is given to ensemble frameworks integrating classical classifiers and neural networks for robust predictions.Deep models enhanced with attention mechanisms, such as Bi-LSTM and GRU variants, demonstrate high accuracy, while novel hybrid architectures show potential in both high-resource and lowresource settings.This study also explores techniques for handling data imbalance and enhancing model interpretability through explainable AI.Findings suggest that while existing models achieve promising performance, challenges such as generalization across languages, real-time applicability, and privacy concerns persist.The review concludes with suggestions for future research directions to develop ethically sound, accurate, and scalable mental health prediction systems.