Stroke Predictability and Leading Factors

Inspiration

Strokes are the second leading cause of death internationally and fifth in the United States. We wanted to create an impactful program that can accurately predict strokes and provide insight into individual stroke risks. Hopefully we can help people reduce their likelihood of suffering a stroke.

What it does

Trains a machine learning model to predict if someone will have a stroke or not. Our project requires an input of age, bmi, and glucose level, and it returns the individual's actual stroke prediction based off our model, their risk level, their deviation from non-stroke patients in each of the inputted variables, and if the stroke prediction is yes, then it will also display the minimum probability of suffering a stroke in the next 10 years.

How we built it

We cleaned our stroke dataset CSV found on Kaggle, then we used sklearn to create a machine learning model to predict stroke outcomes. We also created new columns for classifying age, bmi, and glucose levels as well as a final column classifying their summed risk levels for the inputted information.

Challenges we ran into

We had problems with implementing Bayes theorem, finding a suitable workspace to work and collaborate, and implementing interactive visualizations from new libraries.

Accomplishments that we're proud of

We are proud of the time and effort we put in to make this project happen. We are proud of our implementation of Bayes theorem, the individual function, the risk classifier system, and even how accurate our models truly were.

What we learned

We learned a lot. We learned everything from new libraries, effectively debugging, and working together in a group to in depth machine learning and behind the scenes as to why things work the way they work.

What's next for Stroke Predictability and Leading Factors

Eventually we would like to create a more detailed individual function to get more accurate and personalized results, a more precise Bayes theorem making an actual percentage instead of a minimum percentage, and a new weighted classifier function for a more personalized risk classification.

Built With

• kaggle
• python