Inspiration: I have family members with possible diabetes, so I thought of this app to predict diabetes chances for aging people in a simple way, without any invasive tests.
What it does: Diabetes app predictor predicts whether a person has type 2 diabetes based on the answers provided to survey questions.
How we built it: I used Type 2 Diabetes dataset published by CDC as dataset to build, train and test a machine learning model. This model uses logistic regression. Model take features as survey questions from web user interface and helps predict chances of diabetes in person.
Python, sklearn package, Pycharm IDE, HTML, Flask are used to build this model.
Challenges we ran into: I ran into Data quality issues related to cleaning data like datatypes and format from source data. It was also challenging to perform multiple experiments to identify good features that would be give about 75% model accuracy and precision.
Accomplishments that we're proud of: Easy to use interface for everyone to predict the chances of diabetes.
What we learned: Basic Machine learning concepts, use of python packages for machine learning, flask basics to build user friendly user interface, how to perform experiments in data science.
What's next for Non-Invasive ML Type 2 Diabetes Predictor: Automate model to take new data and self-train periodically from other sources. Explore other machine learning models and see if it can do better than logistic regression. Provide mobile app interface for users to predict chances of diabetes.
Log in or sign up for Devpost to join the conversation.