OneStepAhead

Repository for the One Step Ahead (working Title) Team of the CodeVsCovid19 Hackathon

Intro

We are a group of passionate createives and coders who want to help fighting the spread and the consquences of the Spread of Sars-Cov-2 better known as Corona virus or under the disease it is causing - COVID-19.

Problem description

Many problems arise with COVID-19 and the battle against it. We can only address one. We believe that one huge issue is uncertainty. How to deal with uncertainty? By planning. But for what shall we plan?

Solution description

We do not have a fortune teller sphere but we have the powerfull tools of Machine- and Deep Learning. So we gonne face this problem with prediction.

what is new about this approach?

Yes, nothing to special here. Anyway, we see another huge problem ahead of us. The Spread of the virus in developing countries and refugees camps. Where we see only few cases today. So just applying time-series forecasting is not good enough here. We want to add more features to the model. This way we can not only

• predict the number of cases but also
• explain how effective activities taken are We will accomplish this by explaining the model. Some models are explainable by default (like regression or logistic regression). Other models can be explained by special tools like lime for example.

Data

Data Sources

• Data Repository by Johns Hopkins ...https://github.com/CSSEGISandData/COVID-19
• Wikipedia
• tbd.

Features for prediction

Normally if you are doing time series forecast you need to get your data from a form lik this:

Country	01.03.2020	02.03.2020	...	27.03.2020
Italy	20	35	...	1247
France	15	28	...	968
...	...	...	...

into a form like this:

Country	sum cases last 14 days	sum cases last 7 days	sum cases next 7 days
Italy	340	538	?
France	290	421	?
...	...	...	?

we also want to add additional features (features are the explanatory or independet variables) to better predict the label (the variable to explain or dependent variable)

additional features to use

• population of the countries
• mean age of the population of the countries
• 'closing of schools since x days'
• 'quarantine since x days'
• sentiment of citizens about the quarantine
• ...

some thoughts about statistics

in time-series we have autocorrelated variables, as 'sum of cases last 7 days' is not independent of 'sum of cases last 14 days'. This is also true for other features (variables) we will add to the data model. Anyhow in Machine- and Deep Learning contraints that do normally count for hypothesis to be valid are neglected most of the time. (yes. That actaully is something experts of statistics argue a lot.)

Solution Architecture