Inspiration
The United States really could have been in the end phases of COVID lockdown by now, but ineffective leadership and lack of adherence to health advisories have exacerbated the epidemic and caused an unprecedented amount of unnecessary deaths. One way people could look out for their own health without falling victim to the decisions of others would be to have a better understanding of the COVID situation on the county level, and predict what will happen in these areas in the future.
What it does
Our app uses current data to build an adaptive SIR model of counties in New Jersey. These models are used to predict future conditions in each county and identify hotspots to notify users to avoid.
How we built it
First, we designed a set of ordinary differential equations that would incorporate natural deaths, interaction between sick and infected people, recovery rate, and deaths due to COVID. With an initial guess for these parameters and a dataset from JHU's COVID database we were able to train a model in python to have a relatively close fit using the scipy package. After identifying the parameters we used an iterative solution to predict the infected population days into the future. The python analysis was presented as an interactive web application in which users could visualize past, present, and future COVID-19 cases on a map provided by SimpleMaps. Django, a python-based web framework was used in order to integrate the back-end analysis with the front-end user interface. A mixture of Javascript, HTML, and CSS were used to develop the front-end, and Django's built-in template language was helpful in allowing us to pass python-generated data to the user.
Challenges we ran into
We wanted the app to be useful to people locally to give them a better sense of the situation around, so we looked for county datasets. However, only the infected and death deltas per day were available. This meant we had to calculate relevant numbers like the susceptible population, total infected, and recovered while making assumptions that introduced a degree of error. In fact, the first model we built failed for some of the data and we had to remake it. The time constraint on the challenge meant that we had limited time to try different things out. Trying to find a proper API for generating maps was difficult, but we believe that we ultimately ended up finding something that worked for us.
Accomplishments that we're proud of
Having a pretty good fit for our model to actual numbers while using some math magic to calculate numbers required to build our model. Leveraging both our understanding of biological math modeling and our computer science understanding to develop a very basic, yet potentially useful, application.
What we learned
One of us learned python today by trying to convert from things they knew in MATLAB, We developed our collaborative GitHub skills and were able to streamline the development process because of that.
What's next for Predicting COVID-19 Cases in NJ Using an Adaptive SIR Model
Our next step is to apply this to more states to extend our user base. We also want to make the model more specific so we can do towns within counties. Something we would like to do in the future (or would have liked to do over the past day) is to use the Google AI to predict movement among counties during certain times of the year. For example, tracking such as outlined in this post could accurately predict which location will have a high population concentration and how that concentration changes one a day to day. Then, we could modify our SIR model to reflect how a nice summer day led to a longer COVID threat.
Log in or sign up for Devpost to join the conversation.