People of all ages in all parts of the world suffer from diabetes. Current patients often have trouble tracking their glucose and weight levels over time while others at times wonder if they're becoming diabetic due to recent changes in their glucose, blood pressure or weight levels.

DiaBuddy, a simple web application, aims to solve this problem and make the lives of current and future diabetes patients easier. Firstly, it's machine learning engine allows new patients to enter their glucose, blood pressure and weight levels in order to predict if they're diabetic. The prediction is based on a model created using Microsoft Azure's machine learning platform, which was trained using the Pima Indians Diabetes Data Set from the "National Institute of Diabetes and Digestive and Kidney Diseases". On the other hand, the application's health management system enables patients to track their diabetes levels over time using in-built graphs/charts. The application also messages important alerts and updates to users using the Twilio API.

The application was built using Flask and Cockroach DB on the back-end and AngularJS on the front-end. Linode instances were used to set up multiple Cockroach DB clusters. The machine learning model was created using a Two-Class Deep Support Vector Machine while graphs were created using the HighCharts API.

The application's machine learning model, which was tested using a 230 row record-set, responded with an accuracy rate of 78.3% and a precision rate of 79.4%.

Future work for this project involves using a larger training data-set, making a mobile app to track levels on the go and integrating with diabetes tracking hardware.

Share this project: