Detection of Mental Illness from Reddit Data using Deep Learning and Mitigation using Generative AI

Mental health issues are increasing worldwide, making early detection and timely intervention increasingly critical. This project introduces a two-step system that uses deep learning and generative AI to both detect and respond to signs of mental distress. In the first phase, a hybrid BERT-LSTM model classifies social media posts into mental health categories such as depression, anxiety, bipolar disorder, schizophrenia and BPD. To improve accuracy, the system uses a novel technique called Meaning Retention Pipelined Preprocessing, which retains important parts of the original message — like emojis, slang, and punctuation — to better understand the full meaning behind the text. In the second phase, a Generative AI model rewrites flagged posts, making the content more supportive and less emotionally triggering, while still keeping the original intent. Performance is measured using custom metrics such as Semantic Meaning Retention Accuracy and Sentiment Compound Score, both of which validate the advantages of the proposed preprocessing method. The framework provides a holistic solution that not only detects distress but also promotes a healthier digital environment and integrates actionable resources such as helpline numbers, giving users direct access to support.