<![CDATA[The rising prevalence of mental health discussions on social media platforms has created new opportunities for understanding and supporting individuals facing psychological challenges. This study examines the automated classification of mental health content on Reddit, focusing on five clinically significant conditions (ADHD, anxiety, bipolar disorder, depression, and PTSD) and non-clinical discussions. Reddit was selected as the primary data source due to its unique subreddit structure and rich user-generated content in mental health communities, where individuals actively seek support and share experiences. Using a Multi-layer Perceptron (MLP) architecture, the study conducted a comprehensive evaluation of three optimization algorithms (Adam, RMSProp, and SGD) in conjunction with L2 regularization (λ=0.01) for mental health text classification. The study incorporated Easy Data Augmentation (EDA) techniques to enhance model robustness, implementing paraphrase-based augmentation methods that improved classification performance by 3%. Through systematic evaluation across multiple metrics, the study found that the RMSProp optimizer without L2 regularization achieved optimal performance, demonstrating 83% precision and 82% recall across all diagnostic categories. Notably, the application of L2 regularization consistently resulted in decreased model performance across all optimizers, with performance degradation ranging from 3% to 52%. These findings contribute to the development of more accurate automated mental health monitoring systems while highlighting the critical role of optimizer selection in mental health-related Natural Language Processing (NLP) tasks.]]>
