The Blood Glucose Level Prediction (BGLP) Challenge was again organized this year at the 5th International Workshop on Knowledge Discovery in Healthcare Data (KDH). The workshop was held at the 24th European Conference on Artificial Intelligence, which due to COVID-19 was completely digital. The BGLP Challenge is interesting because it provides data for multiple participants from continuous blood glucose monitoring devices and activities such as sleep, food intake, etc in the OhioT1DM dataset.
Ananth and Priyanshu led the work on the challenge and did some interesting work to think of the problem as a time-series analysis problem. We used sequence-to-sequence models, which were individualized to each patient, and predicted blood glucose values fairly accurately. We were among the few online models, which was one of our priorities because we wanted to make our models generalizable and usable on real-world apps. Our paper 50 featured in the preliminary rankings of the papers for the challenge. You can read our full paper here: Blood Glucose Level Prediction as Time-Series Modeling using Sequence-to-Sequence Neural Networks
The presentation can be seen here: