- Dominik Frey
- Fabian Nick
- Manuel Lang
- Arne Fornell
- Todor Kostov
- Rainer Heintzmann
- Oliver Laitenberger
- John Flynn
- Martin Colte
- Luisa De Guglielmo
- Anish Gupta
Crisis teams in each country communicate actions to stop the COVID-19 pandemic, however, new cases are registered every day. Hospitals become like a trench war for doctors and nurses. Hundreds of thousands have been infected and increasing numbers are dying. Some are working with no gloves due to shortages. The same holds for intensive care units. To avoid the collapse of the healthcare system, it is crucial to forecast where the virus will infect more people to distribute the necessary resources for doctors and patients. To this point corona crisis teams lack valid information on infection spread in their region and its impact on vital resources. To remove some of their uncertainty the crisis team members have the possibility to see the evolution of the infection in their regions and plan and distribute their medical resources accordingly.
What it does
We have developed a visualization and simulation of future infection rates. Machine learning is used to predict the number of diseases for various countries and various regional levels. Together with additional data such as the number of ICUs, officials are immediately informed on resource shortages. In this way, crisis teams can immediately start short-term activities to alleviate the situation by distributing scarce resources.
How we built it
- We collected the German data from Robert Koch Institute (cases, deaths) and DIVI (ICUs, beds) and found daily refreshed data from Switzerland and Italy on GitHub. Also, we have French data, but did not have the time to integrate it yet.
- The data is parsed and enriched with environmental data (geo-coordinates and population) and then stored into the MongoDB hosted on an AWS EC2 instance.
- Our custom SEIR model takes the input data from the database and simulates the infections and the facilitations for the upcoming 30 days.
- Next to the SEIR model, we have also implemented a Latent Variable Model and a Progressive State Model that, however, did not make it to the final cut yet.
- The simulated data is stored in the database together with the historic data.
- The frontend, that is implemented with React and kepler.gl and hosted on AWS S3, fetches the data from our Flask backend then visualized the infections, the facilitations as well as the ICU usage over time from February until one month in the future from now.
- We have implemented the functionality for a crontab to fetch the latest data, run the simulation and insert all data into the database, however, it is not live yet and will need some further testing.
- As an alternative to the kepler.gl frontend, we also have a visualization based on Leaflet.js, because we discovered a breaking bug in the kepler.gl source code (that we then solved by ourselves).
Challenges we ran into
- The kepler.gl setting exporter does not update the weight parameter of the heatmap correctly. We implemented a change in the source code to load it correctly. However, it took us a while and led to some frustration finding the issue.
- We found data for Italy and France only in their respective language, that made it for us (mainly German natives) difficult to understand. Our great team diversity was the key to overcome this hurdle.
- We had to shift our morning meeting on Sunday (hellooouuu daylight saving time -.-)
Accomplishments that we're proud of
We gathered an international team of people, working together applying their high expertise in various technical and non-technical areas.
We implemented our idea and vision in a "running prototype". We incorporated multiple international data sets from Switzerland, Germany and Italy as the basis for our visualisation and simulation.
We enabled regional corona crisis teams to get good forecasts; then evaluate the ICU situation of the hospitals in their area; and then determine some action (such as distribution of resources).
We used several state-of-the-art modeling techniques to find the best fitting model for the simulation.
We used the latest cloud technologies for visualization and publication of the final result.
What we learned
As an interdisciplinary team of highly motivated people, we can work closely together to provide corona crisis teams in different countries a valuable instrument for forecasting their local situation. This helps them manage the crisis which is far from over yet. What could be achieved if this could be scaled even further?
We can all make an impact even when staying at home to fight the coronavirus pandemic.
What's next for Track the Virus
We want to enable hospital officials to enter their own data (e.g. amount of beds, ICUs). This will help them to understand their own situation in relation to the forecast. This will allow hospitals to share their resources to meet their needs.
We want to add more data about other shortages in the database to provide an even more complete view of the situation for crisis teams.
Try It out