DEEP LEARNING-BASED MENTAL HEALTH DETECTION USING FINE-TUNED BERT: A MULTICLASS TEXT CLASSIFICATION APPROACH
Urvashi1, Syed Wajahat Abbas Rizvi2, P.K. Dwivedi3, Ashish Kumar Pandey4, Sandhya5
1,2Amity University, Uttar Pradesh, India
3,4Dr. Rammanohar Lohia Awadh University, Uttar Pradesh, India
5Pathfinder, Lucknow, Uttar Pradesh, India
Abstract: Mental health disorders, including anxiety, bipolar disorder, and suicidal tendencies, significantly affect individual well-being and necessitate timely detection for effective intervention. Traditional assessment methods, such as clinical evaluations and self-reported surveys, are often time-consuming and subjective. This paper introduces a deep learning-based approach utilizing a fine-tuned BERT (Bidirectional Encoder Representations from Transformers) model for multi-class mental health classification through textual analysis. The system classifies text into four categories—depression, anxiety, stress, and normal using advanced natural language processing (NLP) techniques. It features a real-time interface developed using Streamlit, offering an accessible and intuitive tool for clinicians and users. Evaluation on the DAIC-WOZ and Reddit-based datasets yields an accuracy of 93%, demonstrating the model’s strong performance. A user-friendly Graphical User Interface (GUI) was developed to facilitate real-time classification, allowing users to input text and receive immediate feedback. Evaluation through classification metrics confirmed the model’s effectiveness. The proposed system offers a scalable and automated approach to mental health assessment, supporting early intervention and reducing the burden on healthcare systems.
Keywords: Mental Health, Deep Learning, BERT, Text Classification, Streamlit, NLP