Med-GRUD

This is the project I had been working on during the research training. It's a GRUD model for medical analysis data imputation, to be used instead of forward / mean fill.

• Conducted a research project on clinical time-series data imputation, developing a GRU-D–based model to replace traditional mean and forward-fill methods
• Preprocessed 500k+ laboratory analyses from 40k+ patients in the MIMIC-III dataset, grouping irregular time intervals into structured sequences for model training
• Achieved 96 MSE on MIMIC-III and 141 MSE on a proprietary dataset, demonstrating strong model generalization across domains
• Reviewed and integrated insights from state-of-the-art medical ML papers to improve model architecture and preprocessing strategies

Used: Python, PyTorch, Pandas