Diabetes predictive AI Model

Inspiration

I was first inspired by the emerging of AI systems and applications and wondered if I could make my own. I was first confused on how to start. I didn't know anything about making AI models. That is until I started learning and finally managed to get a really good understanding on AI models. After a short while I got to develop an AI model that is capable of predicting the likelihood of diabetesin an individual.

What it does

The model predicts if a person is likely to have diabetes based on a number of factors including (but not limited to): Age, BMI, Blood glucose, Hb1AC levels and more. The model is meant as a support tool for doctors and learners not as a replacement for medical professionals.

How we built it

The AI model was primarily built on Python and a number of libraries, namely: Scikit-learn as the main backbone, Pandas for visualization, NumPy for data handling, Streamlit for UI and Render to deploy it. It was trained on 100,000 records based on a dataset from Kaggle. The raw dataset was not feasible and needed some changes: I used one-hot encoding in order to change text into numerical values and compared both Logistic Regression and RandomForest algorithms. Then I compared their recall values, false positives, negatives etc. and found that RandomForest was the better choice so I went with that. The AI model works on most devices as of this writing and anyone can test it out. The link is attached below with the GitHub repository.

Challenges we ran into

I ran into a number of challenges: version mismatches between the libraries, evaluating the model based on accuracy rather than recall, overlooking basic bugs and some hardware difficulties. Despite all the obstacles I've come across I managed to (eventually) overcome them all and finally deploy the AI model I was proud to call mine.

What we learned

I learned that obstacles are not meant to be avoided but rather learned from. I also learned that asking questions and developing a curious mind is the key to success (and efficient debugging). And ironically, I learned that mistakes are the best things you can do - it is a great way to learn and it challenges your brain to fix it and through time you will be able to learn anything.

Built With

• jupyter
• numpy
• pandas
• python
• render
• scikit-learn
• streamlit