Inspiration
The Corona crisis had an enormous impact on the social life all over the worlds: our consumer behaviour changed, our working times and our form of working changed, and also the preferred way of transportation changed a lot. Machine learning models and many other methods of forecasting rely on patterns learned from past behaviour.
What it does
One important method to adjust to the new situation now is going back to classical data science and looking at descriptive statistics to get an idea of the quality of the changes - to find out which patterns from the past are still the same and which aren't. And to find out to what extent at least in Germany we are already on a way back to pre-Corona patterns. In the first exploratory step we therefore visualized the changes with different datasets depicting the changes in the behavior due to Corona on a website. Overall we looked at data on the sales of different product groups of a bakery branch provided my Meteolytix, the rentals from a bike sharing service in Kiel (Sprottenflotte from the KielRegion), pedestrian movement in main shopping streets of Hamburg and Berlin, and the usage of the IT Help Desk of the city of Kiel. Finally, we prepared data for the bakery branch sales, bike sharing usage, and pedestrian movement. The graphs used to depict the changes show the percentage of the corresponding value in comparison to the same value one year ago. In order to get more stable values that are less volatile (e.g., due to holidays), we used moving averages, in which the daily values represent the average of two or six weeks (we played around with different values for different datasets) before the day depicted in the graph. Therefore, the observed changes are somewhat less extreme but clearer and more stable to recognize.
How we built it
We setup a website using Firebase to visualize the data with chart.js. The data sets themselves were imported from the different data sources and re-calculated and aggregated to the needed values using R, and an additional data preparation step was done in Scala to finish the prototype in time. The link to our GitHub-Repo is the following: https://github.com/nikitaDanilenko/changes2020
Challenges we ran into
Since there was no time (and no money :-) ) to host a web application on premise, we had to find a web-based solution that fit our prerequisites for displaying the graphs for the different datasets and allowed us to import the data in the needed format, which was one main task. The further task was to find a meaningful aggregation of the data and way to depict the data that actually shows the underlying changes in the patterns. We were playing around with different forms like bar charts and line charts of the actual values, which were not very useful though to detect patterns beyond the "crash" due to the lock-down in mid of March. Also we had to notice that many of the available datasets were not perfectly suited since the observed data points were not as need; for example, not covering the year before the Corona crisis (to avoid seasonal effects in the comparisons), or not including the latest data (e.g., only yielding time point until March).
Accomplishments that we're proud of
The members of our team had quite different backgrounds: from a programmers with a the frontend or backend experience, persons familiar with quantitative statistics, and others with a little bit of everything. However, we fit perfectly as a team, with everyone always having his specific task. And the final result is a true team effort, where everyone has its share.
What we learned
We learned a lot about the many possible forms to look the impact of the corona, and in particular how many forms of data representations there are where you actually cannot see anything - in fact, simply because there is too much information. Finding out what works and was not works was therefore a valuable experience.
What's next for The Impact of Corona on our Life
The data from the bike sharing service Sprottenflotte was only available until end of march since dataset updates are only available every 3 month, we would like to upload the updatet dataset including the interesting time span of March, April, May to compare the changes in that dataset to the other ones.
Log in or sign up for Devpost to join the conversation.