+ 61 3 9020 2085 info@parityanalytic.com.au

Brownlow Medal Prediction Using Excel Based Machine Learning

Share this article

A spreadsheet-based regression model using Neural Networks to rank this year’s medal contenders

By Stephen Huppert and Jack Langhammer

It may not be September, and the finals are playing out far north of the MCG, but at least footy fans can take comfort that the Brownlow Medal Awards, AFL’s night of nights will go ahead, albeit virtually and without the red-carpet glamour.

The Brownlow Medal is the highest individual honour and awarded to the best and fairest player during the home-and-away season and determined by votes after each game from the people most footy fans love to hate – the umpires.

There is almost always a favourite, however this year there are some surprise entrants who have enjoyed brilliant seasons, so the race for the medal is still wide open. This got us thinking and we decided to choose the Brownlow Medal race as a test for an Excel based machine learning neural network based on the complexity of understanding the umpires’ decisions on the awarding of the votes and the abundance of high-quality current and historical data.

But first – a quick rundown of machine learning. What is it and how does it work?

Parity Project Financing

“AI is the new electricity. It will transform every industry and create huge economic value.”

– Andrew Ng

Machine Learning and Neural Networks

Machine learning is a subset of artificial intelligence using algorithms that improve automatically through experience. Some algorithms used in machine learning are computing systems modelled on the human brain and nervous system, often referred to as ‘neural networks.’

We decided to investigate using Excel to create and test machine learning neural network prototypes. If successful, the possibilities would be endless and provide a useful environment for businesses to use applications of machine learning before taking the plunge with expensive projects or high-risk decisions.

Predicting the 2020 Brownlow Medallist

With our goal in mind of using Machine Learning to predict the 2020 Brownlow Medallist we got to work.

The basic unit of computation in a neural network is the neuron or node.

In simple terms, there is a layer of input nodes where the data goes in, a layer of output nodes where the result can be found, and a number of ‘hidden layers’ which is where all the magic happens.

We started with an Excel workbook that contained:

the data set for training the algorithm (several previous seasons worth of voting data); and
the data set for the algorithm to make its 2020 season prediction (2020 AFL season data); as well as
the neural network itself.

The neural network and its training derivatives are structured in cells using formulas to calculate the node results, derivatives, and input cells for the node variables.

We used VBA to train the network on the training data set and for testing the network on the testing data set.

We sourced our data from afltables.com and normalised it against the total stats for all players.

Each row contains the data items listed below for one player for one game. That gives us approximately 60,000 rows of data!

  • Disposals
  • Goals
  • Hit Outs
  • Tackles
  • Clearances
  • Frees for less frees against
  • Contested Possessions
  • Contested Marks
  • Goal Assists
  • Team winning/losing margin

The Neural Network Nitty Gritty

The neural network we used contains two hidden layers of 20 nodes each along with the input and output layers.

The hidden layers use what is known as a leaky rectified linear unit (ReLU) activation function. The ReLU is one of the most used activation functions in practice and the leaky ReLU avoids any nodes becoming deactivated which simplifies the training.

The output is the estimated amount of votes the player would receive for each game. This is a decimal number and, in some cases, results in a small negative or a value slightly greater than 3. We compared the players votes in using the training data with the actual votes they polled in each game and tallied these results to determine the likely order of players in the Brownlow.

Who is going to bring home Charlie?

The Brownlow Medal is often referred to as Charlie, named after Geelong great, Charles “Chas” Brownlow. Like everything else in 2020, the Brownlow Medal is going virtual and will be held on Sunday 18 October.

The Parity Analytic Machine Learning algorithm has used the player data to come up with the following top five:

  1. Lachie Neale
  2. Jack Steele
  3. Clayton Oliver
  4. Christian Petracca
  5. Patrick Dangerfield

Lachie Neale, who missed out last season, looks set to take home his first Brownlow this year and is the clear favourite with both commentators and bookmakers. The others rounding out the top 5 in most predictions are Travis Boak, Christian Petracca, Jack Steele and Jack Macrae.

We list our Top 20 at the end of the article.

Test and Learn with Excel

Overall, we are happy with the outcome that mostly mirrors ‘expert’ predictions. Using Excel for our neural network also proved user friendly and relatively simple to develop and implement compared to a more industrial network, for example, Python.

Using VBA to conduct the testing and training on the Brownlow data sets was also quick- roughly 15 minutes for approximately 60,000 rows of data.

Machine learning and, more specifically, neural networks can be used to untangle and break down extremely complex relationships for more accurate predictions and decision making. 

Consider Using Machine Learning in Your Business

There are several barriers to the adoption of machine learning into business operations including cost of development and implementation, determining whether the application is feasible and knowing where to begin. It is a strategic step that requires considerable resources. It’s important to understand your objectives for your business and which of the different types of algorithms might be best for achieving them.

Developing a sample or prototype machine learning neural network in Excel before moving to a more industrial solution is a good way to explore this. Excel would be able to handle other neural network structures with respect to layers, nodes and activation functions. Excel does have some limitations including scalability issues and limits on the amount of data that can be processed and calculation speed but should suit most testing scenarios.

We will leave you with the words of Andrew Ng, founder of the Google Brain project, and co-founder of Coursera:

“AI is the new electricity. It will transform every industry and create huge economic value.  Technology like supervised learning is automation on steroids. It is very good at automating tasks and will have an impact on every sector – from healthcare to manufacturing, logistics and retail.”

If you would like a copy of the Brownlow Medal Excel model, we will gladly share!

Email us at brownlow_ml@parityanalytic.com.au.

Parity Analytic Brownlow Medal Top 20

1.   Neale, Lachie

2.   Steele, Jack

3.   Oliver, Clayton

4.   Petracca, Christian

5.   Dangerfield, Patrick

6.   Martin, Dustin

7.   Hawkins, Tom

8.   Boak, Travis

9.   Lyons, Jarryd

10. Bontempelli, Marcus

11.   Adams, Taylor

12.   Cripps, Patrick

13.   Sheed, Dom

14.   Wines, Ollie

15.   Macrae, Jack

16 . Fyfe, Nat

17.   Viney, Jack

18.   Dixon, Charlie

19.   Parker, Luke

20.   Menegola, Sam

Share this article
Brownlow Medal 2021 Prediction

Brownlow Medal 2021 Prediction

Brownlow Medal 2021 Prediction Share this articlelinkedintwitterWith the AFL confirming the Brownlow Medal ceremony is to be held on Sunday, 19th September we dug out our trusted Brownlow Medal machine learning model to see if we could predict the 2021 winner.Last...

Project Financing

Project Financing

Our client is a leading Australian independent school that regularly places in the top 10 schools for academic results in Victoria. The client needed various…


+ 61 3 9020 2085

Level 19, 15 William Street,
Melbourne VIC 3000

Home                    Careers

About Us               Contact Us

Our Work              Privacy