Project links
- Jupyter Notebook code (You need to open this with Google Colab or download it)
- Paper
Inspiration
According to the WHO, COVID-19 is a global pandemic that has affected over 42 million people worldwide and has caused over 1.1 million deaths. With the first known case in December 2019, the pandemic is still actively growing in many countries in North and South Americas, including the US. Being able to quickly detect and contain new COVID cases is crucial in putting a break to this pandemic. While widespread testing is good detection tool, doing so at a national scale is almost impossible given the current resources. Another way of detecting new cases is through the self-reporting of symptoms. With the prevalence of cell phones, it is very easy for people to report any symptoms that they have. As a result, being able to predict COVID likelihood based on these symptoms would be extremely beneficial.
What it does
I built multiple predictive models of COVID testing results based on the patient symptoms, as well as an ensemble model that combines the results from all of the single models. I also came up with a variety of metrics to measure which symptoms are the most useful in predicting COVID. Finally, I did an ablation study to understand what subset of symptoms is the best at predicting COVID.
How I built it
I built mostly of the machine learning models in scikit-learn, with the neural network being written in TensorFlow. Data visualization was done with seaborn and matplotlib.
What's next for COVID Symptom Analysis
Given the findings of this research, more work needs to be done in understanding what kind of data needs to be collected in order to reliably predict COVID results.
In addition, the time and geographical data were not used. In the future, a geographical and time analysis of how predictive the symptom scores are can be interesting. It is possible that combined with that data, we can find more useful information to predict COVID testing results.
Built With
- matplotlib
- pandas
- python
- scikit-learn
- scipy
- seaborn
- tensorflow
Log in or sign up for Devpost to join the conversation.