Submitted partial fulfillment of the requirements for the master’s degree in computer science
Loading...
Date
2025
Journal Title
Journal ISSN
Volume Title
Publisher
Faculty of Sciences
Abstract
The detection and classification of mental health conditions present considerable challenges due to the variability of symptoms and the complexity of integrating diverse data sources. This study aims to enhance the precision and reliability of mental health detection by leveraging artificial intelligence techniques. Using a dataset comprising textual data from social media posts and behavioral data from wearable sensors, we developed and evaluated multiple models, including traditional machine learning approaches such as Random Forest, alongside advanced deep learning architectures like Long Short-Term Memory (LSTM). The models' performances were assessed using metrics such as accuracy, precision, recall, F1-score, and ROC-AUC score. The experimental results demonstrated that deep learning models significantly outperformed traditional methods, with the LSTM achieving an accuracy of 93% and the Random Forest reaching 73%, supported by an AUC of 0.7228 for the latter. These findings indicate a substantial improvement in diagnostic performance, suggesting that AI-based systems can play a pivotal role in reducing diagnostic errors and enhancing the efficiency of mental health detection. Implemented with Streamlit 5.6, the system offers a user-friendly interface, with proposed enhancements including additional data modalities and clinical scalability. Implications for healthcare include improved early intervention and personalized care, paving the way for integrating AI into mental health workflows.