In recent years, sentiment analysis has become more important for studying mental health since digital platforms give people a way to share feelings about their mental health. Weevaluate the effectiveness of machine learning (Naive Bayes, Logistic Regression) and deep learning models (LSTM, GRU, BERT) for sentiment analysis on mental health-related tweets from a publicly available Kaggle dataset. While acknowledging the dataset's potential limitations in representing diverse populations due to social media's younger, tech-savvy user base, we demonstrate BERT's superior performance (statistically validated) attributed to its contextual understanding capabilities, with comparative analysis revealing significant accuracy improvements over sequential models like LSTM. Our key contributions include a rigorous benchmark of model performance with statistical significance testing, insights into transformer architectures' advantages for mental health sentiment classification, and practical implications for developing more accurate AI-based mental health chatbots. The findings advance sentiment analysis methodologies while transparently addressing dataset constraints, providing a foundation for future research on more representative data collection and hybrid modeling approaches.

Sentiment Analysis, Mental Health Classification, Machine Learning Models, Deep Learning Models, Kaggle Dataset.

The global surge in mental health challenges demands innovative digital solutions. Social media platforms have become unexpected windows into psychological well- being, where expressions like"I can't get out of bed today #depression"or"My anxiety is paralyzing—heart won't stop racing" reveal unfiltered emotional states. These digital traces enable sentiment analysis to decode mental health conditions through natural language patterns. The rise in mental health issues among people today demands new and creative digital tools to help address these concerns. The extensive usage of social media sites and online discussion boards presents a chance to get vital information about people's emotions. These digital spaces serve as valuable resources for analyzing sentiments, which means understanding the emotions and mental states expressed by users. By using a Kaggle dataset that is specially labeled with different mental health conditions, researchers can identify various states of mental well-being. This dataset includes categories such as normal, depression, anxiety, stress, suicidal thoughts, bipolar disorder, and personality disorder. By analyzing the content shared in these online spaces, we can gain insight into the mental health challenges that many individuals face, leading to better support and resources for those in need. The dataset applied in the present study blends data from different platforms: Reddit, Twitter, and theme-based mental health forums. This thus captures varying aspects of the discussions of mental health. This comprehensive data annotation gives a strong baseline for models to train and distinguish mental health states in a more efficient manner. With this dataset, we want to conduct an investigation of several models in ML and DL: Long Short-Term Memory (LSTM), Naive Bayes, Bidirectional Encoder Representations from Transformers (BERT), Gated Recurrent Unit (GRU), and Logistic Regression models used for the classification of mental health sentiments. Since they can understand text context and patterns, these models, which include a broad spectrum of deep neural networks in the area of natural language processing, are particularly well known for their extensive use in text classification. Despite the fact that numerous studies have utilized NLP approaches for sentiment analysis, none have attempted multi-class mental health classification using such a complex dataset that integrates the majority of mental health conditions with different levels of severity. In addition, with the increased proliferation of mental health support chatbots, it becomes increasingly important to certify the classification of different conditions of mental health for the enhancement of the chatbot’s empathy and relevance in various mental health dialogues. This study's comparative analysis results yield useful information on how effective or useful each model can be in identifying and classifying the patients’ mental health conditions, providing a basis towards building more effective mental health AI-enabled support systems. In the course of this study, the goal will be to advance the area of digital mental health by showing the advantages and weaknesses of existing, widely used NLP models aiming for better mental health interpretation and on- time assistance.