Melbourne Datathon 2020 - the covid edition

Challenge A

Can electricity consumption patterns tell us anything about the pandemic?

The global pandemic has created a surge in demand for forecasting modellers as business and government try to predict what is going to happen. A major problem is that many of the existing models are not going to be accurate as the current conditions have not been experienced before. Can we be more creative and supplement traditional models with additional information not previously used?

Electricity consumption is one such potential data source. When we use electric kettles, lights, air-conditioners and heaters, this activity is recorded in the form of meter readings and when aggregated is a record of what is happening in a region.

Electricity demand data can be of a very high resolution  (e.g. every 5 minutes) and can be publicly available. Most economic indicators are backwards looking and updated quite infrequently (ie monthly), which makes it difficult to judge the impact of policy decisions. Other metrics, such as retail spend only capture a small proportion of the economy. If electricity demand were to be a good proxy for economic activity, then its higher resolution may make it an important metric to use for faster and better decision making. But we don’t know if this is the case.

During the pandemic there has been a shift in behaviour; industry has been disrupted and people are working from home. The purpose of this challenge is to examine historical electricity consumption data to determine if such shifts can be detected. The hope is then that this may potentially be used  as a barometer of activity.

Background

There are a few key drivers that determine the amount of electricity consumed at any point in time:

1) the weather – we use electric heating and cooling
2) seasonality – we use electric lighting
3) business cycles – there is considerably less consumption at weekends, on public holidays and during school holidays
4) base load – the underlying population size and level of business activity (ie growth/contraction)

Under ‘pre-covid’ conditions, the weather was the only real variable as the other drivers were consistent or could be pre-determined (growth would be variable but slow moving). During the pandemic, many regions around the world have seen a disruption in business activity, and hence the balance of these drivers would have shifted to some extent.

The Challenge

We want you to analyse some electricity consumption data and determine if a change can be detected that may be attributable to covid-19.

Deliverables

Insights category

An online report detailing your investigation and any conclusions. This main report should be no more than a 4 minute read and contain no more than 6 charts. Any extra work you have performed that you want to include (such as code, more specific detail, a video describing your journey etc. ) can be included as hyper-links to other documents. The main report will be the first filter in the judging process, but any other linked work will be considered when determining the winners. Any analysis that is done should be fully reproducible – any included code should be as a linked document rather than in the main report. 

The platform you use for the report is up to you, but it needs to be online and accessible to anyone. 

Data2App category

The Data2App category has a rich history within the Melbourne Datathon over the past 3 years, with over 20 winners of this category placed as interns at ANZ.

The deliverables for the Data2App category is an interactive application or dashboard that tells a story about any relationship between electricity consumption and COVID. As in previous years, the most successful entries in this category define a target audience for the app, and guide that audience to answer those questions on their own.

Your entry in the Data2App category must include:

  • A link to a public code repository containing any code required to run the app (including data sourcing and preparation) along with a readme file including directions on how to get it to run (including the installation of dependencies if required).
  • A link to an online video (eg on YouTube, 4 minutes max run time) introducing and demonstrating the app

You may use any technology you wish to process your data, and to run the visualization (e.g. python, flask, react, R, RShiny , Tableau, PowerBI). The app does not have to be published online (the video can be of a locally run version). 

We are assuming your app will help to explain any findings you have made about electricity consumption and the pandemic through an interactive visualisation. You may want to overlay other data such as when any restrictive measures were introduced, employment rate, retail spend etc. and compare geographic regions that have implemented different strategies to deal with the pandemic.

Data

You will have to discover your own data source for this challenge. It does not have to be from Australia. It could be a national source or just your own smart meter data.

But I don’t know anything about electricity or forecasting

We don’t necessarily expect many people will have direct experience, so don’t let that prevent you from participating. Demonstrating that you have the aptitude to have a go at any problem is what employers are looking for.

Can you give us a clue where to start?

Build a model to predict electricity consumption on data that was pre-covid. The predictor variables will be be things like the weather, day of week etc. Use this model to make predictions for the covid period – does it work or have things changed?

This is just a suggestion, not a requirement as you may have your own ideas.

Can we enter as a team?

Yes you can, but the maximum team size is 3 people. In order to be eligible for any internship prizes you must enter as an individual and it must be all your own work. You can only be part of 1 entry for each part of the challenge.

Deadline

Insights category: 11 pm, Sunday 18th October

Data2App category: 11pm, Sunday 8th November

How to Submit

We’ll let you know how to submit your entry closer to the deadline. Come back here then!

What if I want to be considered for an internship?

If you are wanting to be considered for an internship, then your resume should be attached to the submission. At the very end of your resume (ie nothing should appear after it) and in the same document there should be a section on the 2020 Melbourne Datathon with a clear hyperlink to your  report.

Prizes

The prizes below are a minimum

  • At least $1,000 total prize pool
  • The opportunity to interview for one of 2 six month internships with Etika potentially leading to a full time role
  • The opportunity to interview for one of up to 5 internship positions at ANZ

The submissions will be assessed by the judges who will determine the top 3 places in each category with these winners each  receiving a cash prize. The CV’s of the top 5 entrants in each category requesting an internship will be forwarded to the relevant companies who will then be responsible for taking the process from there.

Announcements

The winners will be announced once both categories have been judged any internship offers have been accepted.

There are 2 categories, how will it work?

You will first have to come up with a story to tell before developing an app to communicate  it, hence we envisage most teams will want to enter both categories. You do not have to enter the insights category to enter the data2app though (if you miss the first deadline then you still have a 2nd chance of winning). If you enter both categories, you can win both categories.

If you enter as part of a team for the insights, but the team members want to enter individually for the data2app (so you can still be eligible for internships) then that is allowed. You must have permission to use any code written by other team members (such as data preparation and modelling) that is consumed by your app and it must be noted in the code repository any code written by others.

Caveat…

We reserve the right to adjust any of the rules at any time if we see fit (we don’t plan to, but we might have missed or overlooked some things!)

Judges - Insights

Rob Hyndman

Phil Brierley

Judges - data2app

Terence Siganakis

CEO Growing Data

Anurag Soin

Director Data Science, ANZ