Diabetes Risk Model for Native Women

Inspiration

Native Americans in modern America have a life expectancy about 4.4 years less than the average American of all races. Furthermore, Native Americans die at a higher rate due to preventable illnesses such as chronic liver disease, respiratory disease, and diabetes. In addition to larger barriers to healthcare, Native women are often times hesitant to seek guidance from physicians due to high rates of sexual assault at medical workplaces unaddressed by the Indian Health Service. Access to healthcare for Native Americans is largely ignored in the US, and this project hopes to draw attention to the issue while providing a useful tool for native women.

What it does

This web app uses machine learning models to evaluate the risk of diabetes for Native women using the Pima Indians dataset. Complete a brief survey and are presented with a percentage risk for the onset of diabetes.

How we built it

The backend was built through training random forest and convolutional neural network classifiers on the Pima Indians dataset. The models parameters were optimized through a brute force approach using GridSearchCV. The models were cached using pickle and was called to make predictions in Streamlit-powered front-end.

Challenges we ran into

Since this was our first hackathon and first shot at making a web app, it was challenging for us to figure out where to start. We asked around for help and were lucky enough to get some helpful advice from others. Additionally, we were able to get going through consulting Youtube videos. Another challenge we faced was integrating the model into the front end. For several hours, we were stuck as to why we couldn't call the model in our front end code. After a while, we realized that we were calling our optimization model instead of our classifier and updated our code to adjust it.

Accomplishments that we're proud of

We're proud of successfully making a product that had both a front and back end. This was a very new experience to us, and all of us had come to the hackathon originally to learn from the seminars. After we got together as a group, though, we were excited to push out the project.

What we learned

We learned the basic ideas of full stack development and how to connect an analytic model to an interactive website.

What's next for Diabetes Risk Model for Native Women

Next, we will figure out how to select which model to use based off of the user's inputs. Currently, the user is given choice for the model, but eventually we will want to figure out automatically which model is more accurate given the input.