05_DIABETES PREDICTOR

"Smart Health; Smart Choices". What if there is a platform that can predict if you are diabetic or not with accurate results?! The desktop application we build does that! Watch our video for more! :)

Comment

Inspiration

We see many people around us for whom diabetes has become a part of everyday life. Taking medicine to counteract diabetes has been normalized that nobody blinks an eye at it, but how much of their lives must people lose to this incurable disease? Hence as college students pursuing Computer Science and Artificial Intelligence and Machine Learning we have come up with an idea to predict diabetes and take measures to either prevent or control it by making appropriate lifestyle changes.

What it does

The Diabetes Predictor Desktop application gets an image of the medical report from the user in the form of jpg, jpeg or png. Once this image has been uploaded the application runs OCR operations to extract the required data from the image. The eight machine learning models in the program are trained using 90% of data and tested using 10% of data. These trained models then use the extracted data to predict if the user is diabetically positive or not.

How we built it

Once we understood our problem statement we had to think of a solution as Engineering students. Using our basic understanding we laid out a framework and guidelines as to how our application should work. The user interface was developed to be easy to use and accessible to everyone who has access to a computer.

Challenges we ran into

We faced multiple challenges along the development of the application:

  1. Development of Desktop application to be compatible with both Windows and macOS systems.
  2. OCR extraction issues.
  3. Cumulative prediction errors.
  4. Predictor report PDF generation.

Accomplishments that we're proud of

  1. Our application has the ability to run on both Windows PCs as well as macOS PCs.
  2. We built a HTML dependent Desktop Application hence any change in content to be displayed need not be edited for Windows and MacOS separately.
  3. Trained our models to have around 80% accuracy and have ROC curve AUC value of about 0.7
  4. Trained the OCR model to extract the data without ambiguity.
  5. Developed a fully functional desktop application and a front-end mobile responsive website in 2 days of time!

What we learned

Whenever we faced challenges, we as a team resolved it. We learnt about the importance of teamwork and planning throughout the whole journey! Each one of us had the opportunity to learn about frontend, Scikit-learn, pyqt5 and OCR. This hackathon gave an insightful perspective on how vast technology can be used for the improvement of our lifestyles. Our team is really going to move towards developing 'Technology for good' in major fields.

What's next for DIABETES PREDICTOR

In the near future, we plan on

  1. Increasing the sensitivity of OCR to scan any type of medical report.
  2. Provide more personal lifestyle suggestion according to the individual's age, gender and other parameters.
  3. Incorporate the front end of the website with ML and make it fully functional as the desktop application and provide public hosting.
  4. Include a chatbot in the website that can answer questions related to diabetes and the predictor tool.

Built With

Share this project:

Updates

Submission history