This project demonstrates the development of two deep learning models — a Transformer and an LSTM — two models to predict psychosis-related outcomes in youth using brain imaging, demographic, and clinical data as features.
- Built custom LSTM and Transformer models to analyze complex sequential and cross-sectional data.
- Focused on extracting meaningful patterns from temporal and snapshot data in this dataset.
- Dataset is private and not included.
- LSTM: Captures temporal dependencies in the data.
- Transformer: Uses attention mechanisms to model relationships across sequences.
Achieved 95.7% accuracy and 82.5% AUC on a highly imbalanced dataset.