Inspiration
Our inspiration was the death of a family friend who died of heart failure a little more than 2 years ago at the age of 73. As the theme was announced as health, we thought of his death and decided to make a prediction app based on his cause of death, heart failure.
What it does
Our app takes in multiple health-related factors of an individual related to the heart and uses a machine learning model to predict if this individual will suffer death from heart failure in a specified time period.
How we built it
We coded this machine learning system using Python in the Google Colaboratory platform. We used python libraries such as Pandas and Sklearn to develop our machine learning model effectively and to perform essential computations. The finished product took a lot of experimentation but we are satisfied with the finished product.
Challenges we ran into
The first main challenge we ran into was maximizing the accuracy of our system. This required us to experiment with 50 to 60 different machine-learning models which took an extensive period of time. The second main challenge we faced was creating an input function for user input to be passed through the model. After doing some research and experimenting with the code, we were able to create a working input function.
Accomplishments that we're proud of
The accomplishments we are proud of are successfully creating a heart failure prediction model with an accuracy of 93.3% and creating a function that takes in user input to be evaluated and predicted.
What we learned
We learned the machine learning applications of sklearn in building a prediction model, specifically a random forest classifier. Additionally, we learned how to optimize the model to be 93.3% accurate from its original 50% accuracy and how to take in user input, convert into a dataset, and pass it through our system which increases accessibility to users.
What's next for our Machine Learning Heart Failure Prediction System
We will continue to experiment with machine learning models and neural networks in order to try to reach an accuracy of 98%. We plan to use Streamlit and other software to create a working online application to increase user accessibility. Additionally, we hope to partner with a health-related community organization or research institution to maximize our real-world impact and continue developing the system.
Log in or sign up for Devpost to join the conversation.