Glucose Level Prediction

0 ×

Glucose Level Prediction

This workflow reads in the diabetes subset from the Dubosson data set, which contains glucose measurements, as well as several sensor measurements like heartrate, breathing rate, activity and acceleration.

The exploration Component generates an overview of all the different measurements and lets the user filter the data by date and patient. To update the view without closing the window, the "Refresh Button" can be used.

Afterwards an LSTM network is trained using 15 minutes of blood glucose measurements. The goal is to predict the next 15 minutes of blood glucose levels for each patient.

The data is split into training and test set by taking 80% of the data for each patient from the top as training set and the remaining 20% as test set. "Diabetes_Patient_008" is not included in the training/testing process. This patient will be used for validation purposes.

At the end, the model performance is displayed in an interactive view. In this view different Patient IDs can be selected and the corresponding Line Plot, as well as the RMSE score is shown. Also the results for "Diabetes_Patient_008", which is used for validation, is shown via a Line Plot.

To learn more about this workflow, please read: https://www.knime.com/blog/predict-blood-glucose-lstm-cgm-data

The data is available for download from:
https://zenodo.org/records/5651217/files/diabetes_subset_pictures-glucose-food-insulin.zip?download=1
https://zenodo.org/records/5651217/files/healthy_subset_sensor_data.zip?download=1

URL: Dubosson Dataset https://github.com/irinagain/Awesome-CGM/wiki/Dubosson-(2018)
URL: The open D1NAMO dataset: A multi-modal dataset for research on non-invasive type 1 diabetes management https://www.sciencedirect.com/science/article/pii/S2352914818301059

Nodes

• Row Filter16 ×
• Column Filter15 ×
• Component Input14 ×
• Component Output14 ×
• Loop End9 ×
• Show all 49 nodes

Extensions

• FeatureKNIME Base nodes
• FeatureKNIME Conda Integration
• FeatureKNIME Data Generation
• FeatureKNIME Deep Learning - Keras Integration
• FeatureKNIME Deep Learning - TensorFlow Integration
• Show all 10 modules

Links

• No links available