Predicting Blood glucose levels with Phased LSTM
Fagerholm, Jimmy (2021)
Fagerholm, Jimmy
2021
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi-fe2021061838800
https://urn.fi/URN:NBN:fi-fe2021061838800
Tiivistelmä
In this thesis, a phased long short-term memory model is implemented to predict the blood glucose level in patients with type-1 diabetes with a 30-minute forecast. We will continue previous work by extending the standard long short-term memory deep neural network model with a phased LSTM cell. The model is trained on the OhioT1DM dataset from the BGLP challenge. This study will try to solve a standard LSTM model’s bottlenecks by using a phased LSTM model. Furthermore, an attention-based phased LSTM model will be implemented to achieve explainability to this research topic’s models. An attention-based phased LSTM model performs best if trained on a larger dataset. The performance is on par with previously implemented methods for predicting blood glucose levels from the BGLP dataset.