Inspiration
We wanted to combine the technical lense of AI into the sea of mystery that is heart disease (angina). Inspired by the apple watch and its feature to detect atrial fibrillation. More specifically we wanted to create something more accurate than the apple watch.
What it does
A machine learning model that is able to take several parameters most yielded through non-invasive methods, and produces a prediction for the likelihood of exercise-induced angina with 80% accuracy.
How I built it
Using a multitude of online resources and websites, we were able to utilise TensorFlow and pandas libraries to create a machine learning model on jupyter notebook.
A golden standard dataset from Kaggle was imported as a csv, and after the data was normalised, we were able to order this data into columns that TensorFlow could use. After that, we isolated for the dependent variable and independent variables. From there, the model was created and was trained to predict the dependent variable. Following the creation of the model, we listed the predictions and compared them to the real-life data to yield the accuracy of our model.
Challenges I ran into
-Using tensorflow libraries and pandas libraries was more complicated than we are used to, and we ran into several errors in the implementation of these libraries
-Properly inputting the independent variables into the model using the sklearn.model_selection library.
-Setting up jupyter notebook in an isolated virtual environment with a specific version of TensorFlow
-Finding unbiased, appropriate data to train the model on
-We had to deal with errors that we have never seen before, and it was quite difficult to overcome them.
-Understanding how the AI model was being trained and deployed was quite a hurdle, especially when it came to using the input function
Accomplishments that I'm proud of
For both me and my partner this was our first hackathon and our first time using machine learning. We're both extremely proud that we were able to make a functional model with TensorFlow to predict if a patient will experience exercise-induced angina with 80% accuracy. Learning how to manipulate data in the context of machine learning also provided a strong sense of accomplishment. We are particularly proud of being able to use TensorFlow effectively as it was extremely daunting at the start.
What I learned
Learned how to utilise TensorFlow in order to create a machine learning model Learned how to utilise pandas library in order to organise data and to structure it for a machine learning model Learned more data manipulation methods such as Train_test_split that are used to train the model- that was quite satisfying to understand!
What's next for Project Lionheart
Create an interface such that a user is able to input their own data and the model will give them a personalized prediction as opposed to predictions only being tied to a preset data set.
Log in or sign up for Devpost to join the conversation.