As the relative of someone who has diabetes, I've seen firsthand that without precise management of glucose levels a diabetic suffers. Over 29 million people in the U.S. rely on regular blood sugar readings to avoid severe complications associated with abnormal glucose levels. However, while most diabetics are aware of the lifestyle factors that influence blood sugar, accurately predicting the behavior of an individuals glucose level is extremely difficult. Introducing DiabeTech, an app that uses machine learning models to predict the future blood sugar levels of a diabetic!

What it does

The app we developed has the user enter their most recent blood glucose reading as well as a variety of activity indicators that can influence a diabetics glucose level. When the user enters a new reading, DiabeTech connects to our machine learning model to predict the blood sugar level of the user over the next three hours, providing the predicted values in an easily understandable chart. Our app also gives the user the option to request a push notification alerting them at a time predicted by the model. The user will be notified that their blood sugar is predicted to be out of range and that they should check their current glucose level and take appropriate action. Additionally, our machine learning model trains to our user data periodically to optimize accuracy. DiabeTech uses machine learning as a preventive health tool to take the guessing out of blood glucose management.

How we built it

We created a machine learning algorithm based on a deep neural network trained for regression. The Neural network has been developed entirely in python and using the library Tensorflow and Numpy. The app was developed for iOS in Swift.

Challenges we ran into

We ran into multiple challenges such as creating a proper cloud database that both the client and the machine where we are running the ML model can access. Moreover, the data set we found to train our model was not clean and presented high dimensionality, which caused difficulties.

Built With

Share this project: