This is a project submission for the Hackathon Track: Novice
What it does and Why
Our creation uses a neural network to predict whether a patient has or is at risk for diabetes-2 based on certain factors, like BMI, Glucose levels, Skin Thickness, etc. We hope that this tool can be applied to patient databases to warn people who are at risk of diabetes earlier, so that they don't have to find out about it for the first time when more dangerous complications and problems ensue.
How we built it
We first built a neural network that trained on the Pima Indians Diabetes Database, originally from the National Institute of Diabetes and Digestive and Kidney Diseases. Then, we split up to simultaneously begin building a website and a way for individual patients to access and use the neural network from a website. Finally, we added a feature on the website to edit entire csv files with patient records to add a prediction and probability column.
Challenges we ran into
On our first day, our biggest problem was increasing the accuracy of our neural network. We had to first tweak the number of layers and nodes to reach an accuracy of approximately 60%. Then we standardized our dataset before training, allowing us to reach 77% accuracy. It also took us some time to figure out how to link the form for patient inputs to the neural network. However, when we finally solved these challenges, we found that they were accomplishments we could be proud of.
What's next for DIA-Gnosis
We hope to further improve our neural network, hopefully training it on a more diverse and larger database, and perhaps increasing the number of layers. We also hope to make it easier for hospitals and health care providers to use our tool by improving its ability to take in and add predictions to entire databases.
Log in or sign up for Devpost to join the conversation.