On the same day Dublin squeezed past Stephen Rochford’s men to retain the All-Ireland title last October, the Western Bulldogs ended a 62-year famine when they claimed their first AFL Premiership since 1954.
But unlike Mayo, the Bulldogs hadn’t been competing at the top table quite as consistently. They hadn’t even made it to a Grand Final since 1961.
It was a fairytale end to a magical story for the team based in the inner-western suburbs of Melbourne, Victoria. The Bulldogs only finished seventh in the AFL ladder but they dominated the finals series right up to their 22 point Grand Final win over Sydney.
AFL
AFL
Head coach Luke Beveridge and his players rightly took the plaudits for delivering only second Premiership cup in the club’s history, but it was also a triumph for those behind the scenes.
In 2014, the Bulldogs started doing things differently. They are one of the few clubs in Australia gaining an edge on their opponents by using technology. Less than three years ago, a deal was signed with the nearby Victoria University which gave the club access to a world-class team of sports scientists.
The Bulldogs were the first AFL side to use computer machine learning to collect data on players and provide analysis to help the club with decision-making.
Dr Sam Robertson leads a team of 19 staff and students, who use hi-tech equipment to help the club in key areas like player recruitment, training and matchday decisions.
“We’re probably different to a lot of Australian clubs where we actually go with modern machine learning where we can,” Robertson tells The42.
“I don’t think that’s the norm. I don’t think most clubs here have a real machine learning background.”
He is a specialist in machine learning and data analytics, and believes Australia may be as far as 10 years ahead of leagues in the US like the NFL and NBA in terms of their application of sports science tools like GPS.
“Sports science as we know it is still very new to the US and they’re not common place even in some sports yet. I’ve had colleagues that have gone over there and set-up programmes from nothing in the last couple of years.
“That would never happen here because every club has a programme in place already. I’m sure they’ll catch up, it’s just new. It’s not that they’re doing anything badly, it’s just it’s a lot newer.
“Australia probably has a critical mass of students, there’s so many universities that offer it and have offered it for a long, long time. We’ve been using a lot of sports science equipment, technology and practises for a long time, so it’s only natural we continue to develop those.
“GPS is a good example of that. US sports have only really had GPS for a short period of time, whereas it’s been common practice here for quite a long time now.”
Advertisement
Jonathan Bachman
Jonathan Bachman
Machine learning is the useful information computers provide from the tonnes of data it picks up from training and games. The Bulldogs players wear small devices in their jerseys so the club can track their physical performance in the gym or on the field.
“We use GPS for load monitoring and as a performance tool to see how far our players are running and at what intensities.
“We’re taking additional analysis on top of that and additionally we’re interested in the location of the player. That’s where it fits in to the coaching side of things where we can look at the formation at different game phases.”
Data is collected about the location of each player in every second of a match, like a Fitbit tracks your exercise.
Western Bulldogs
Western Bulldogs
“That’s fed back to a computer,” Robertson continues. “It’s an objective measure, we can get it in real time and it takes that burden off a trainer having to stand there and collect it all, or the player having to write it down.”
After a AFL game, the club can pull up small videos of gameplay with the positions of players and simulate the outcome of a play if they had acted differently.
Interestingly, they can also use the data to ascertain with combinations of players work best together.
“Combinations of players, we just see as a bit of a model. As we add data to that model we can increase the accuracy of it. In that sense we’ll use machine learning techniques to look at how players cluster their performances together.”
Machine learning means the club can model a player’s injury risk based on history and training load. It can even predict a player’s career longevity based on a number of different factors. But recruitment strategy is perhaps where it is at its most influential.
“When you think of that area you go to Moneyball straight away,” he continues “We weren’t doing a lot of that in Australia, but we do now. We put more resources into that around 2015.
“We have students that watch and record things from a coaching perspective around the league as a secondary scout, but we also have students that are crunching the numbers on that data to create a model.
“We look at things that model around how similar that player is to other players we already have on our list (squad). We obviously want to have a balance in our roster.”
Every AFL club must obey a salary cap, so Robertson says his team must “forecast how good some of the current players are going to be and what they’re going to cost to keep them.”
The club’s recruitment strategy is to go for versatile players who can play in multiple positions.
“We want to be predictable to ourselves, but unpredictable to the opposition. That’s the main mantra we’ve got here. If a player who’s playing a specific role for us is having trouble or being targeted by the opposition, we feel like we always have Plan B, C, D, E – a number of different plans we can use in that scenario.
“It’s our greatest strength in our list. It worked pretty well last year so it’s certainly something we want to continue.
“It’s kind of our forte to go for a older player provided he fits in with our list.”
The Bulldogs also place far more importance on technical ability and intelligence than physical characteristics.
“From a recruiting perspective it’s far more about the skill side of things than the physical. Australian football went through a phase where everyone was looking for the best athlete they could get their hands on, but it’s still a skill-based sport.
“That’s first and foremost what we’re looking for – players that can kick and handball, because that’s more fundamental to the game. It’s probably easier teach someone to run better later down the track rather than teaching them to kick.
“It helps because we’ve got a really strong idea at the club for the types of players we want to have here. It sounds obvious but a lot of clubs don’t have that. We can evaluate our system against those values we’ve got here.
“It’s just also basic intelligence, we can pick out the education we’re going to provide them and make sure they’re going to understand it quickly. If they do that they’ll be more likely to play for us earlier.”
It’s a fast-moving industry and Robertson believes we’re not far away from the introduction of tattoo-like sensors which could measure kicking, heart rate, sweat rate and plenty more.
“They would work in much the same way as wearing sensors in the gym, except you wouldn’t have to strap them onto your body.
“A little bit more complex is what type of sensors are on the body, how we get battery to them, how we wash them, these types of things. There’s already research going on in this area. They could probably do it now, it’s just about making it cost-effective.
“I can’t see a world in five or ten years time where we’re still putting big trackers on people’s backs under a jersey. I think we’ll be doing it through these sensors on the skin, no questions about it.
“It would be easier to have it under the skin but that might be a step too far.”
James Crombie / INPHO
James Crombie / INPHO / INPHO
Robertson doesn’t overstate the role machine learning had in the Bulldogs’ AFL Grand Final victory, although it did influence the team’s style of football and help them identify some of the players who contributed to the win.
The Western Bulldogs have shown the positive benefits advanced sports science can bring to a team. Australia is the world leader in this area and it may be some time before artificial intelligence makes its way to an amateur game like GAA, although it might not be as far away as you think.
Who knows, maybe post-All-Ireland final victory speeches in the not too distant future will even include a word of thanks to the county computer – for all its efforts during the year.
The42 is on Instagram! Tap the button below on your phone to follow us!
Artificial intelligence helped the AFL champions end a 62-year famine, but will we ever see it in GAA?
IT WAS A drought Mayo fans could relate to.
On the same day Dublin squeezed past Stephen Rochford’s men to retain the All-Ireland title last October, the Western Bulldogs ended a 62-year famine when they claimed their first AFL Premiership since 1954.
But unlike Mayo, the Bulldogs hadn’t been competing at the top table quite as consistently. They hadn’t even made it to a Grand Final since 1961.
It was a fairytale end to a magical story for the team based in the inner-western suburbs of Melbourne, Victoria. The Bulldogs only finished seventh in the AFL ladder but they dominated the finals series right up to their 22 point Grand Final win over Sydney.
Head coach Luke Beveridge and his players rightly took the plaudits for delivering only second Premiership cup in the club’s history, but it was also a triumph for those behind the scenes.
In 2014, the Bulldogs started doing things differently. They are one of the few clubs in Australia gaining an edge on their opponents by using technology. Less than three years ago, a deal was signed with the nearby Victoria University which gave the club access to a world-class team of sports scientists.
The Bulldogs were the first AFL side to use computer machine learning to collect data on players and provide analysis to help the club with decision-making.
Dr Sam Robertson leads a team of 19 staff and students, who use hi-tech equipment to help the club in key areas like player recruitment, training and matchday decisions.
“We’re probably different to a lot of Australian clubs where we actually go with modern machine learning where we can,” Robertson tells The42.
“I don’t think that’s the norm. I don’t think most clubs here have a real machine learning background.”
He is a specialist in machine learning and data analytics, and believes Australia may be as far as 10 years ahead of leagues in the US like the NFL and NBA in terms of their application of sports science tools like GPS.
“Sports science as we know it is still very new to the US and they’re not common place even in some sports yet. I’ve had colleagues that have gone over there and set-up programmes from nothing in the last couple of years.
“That would never happen here because every club has a programme in place already. I’m sure they’ll catch up, it’s just new. It’s not that they’re doing anything badly, it’s just it’s a lot newer.
“Australia probably has a critical mass of students, there’s so many universities that offer it and have offered it for a long, long time. We’ve been using a lot of sports science equipment, technology and practises for a long time, so it’s only natural we continue to develop those.
“GPS is a good example of that. US sports have only really had GPS for a short period of time, whereas it’s been common practice here for quite a long time now.”
Machine learning is the useful information computers provide from the tonnes of data it picks up from training and games. The Bulldogs players wear small devices in their jerseys so the club can track their physical performance in the gym or on the field.
“We use GPS for load monitoring and as a performance tool to see how far our players are running and at what intensities.
“We’re taking additional analysis on top of that and additionally we’re interested in the location of the player. That’s where it fits in to the coaching side of things where we can look at the formation at different game phases.”
Data is collected about the location of each player in every second of a match, like a Fitbit tracks your exercise.
“That’s fed back to a computer,” Robertson continues. “It’s an objective measure, we can get it in real time and it takes that burden off a trainer having to stand there and collect it all, or the player having to write it down.”
After a AFL game, the club can pull up small videos of gameplay with the positions of players and simulate the outcome of a play if they had acted differently.
Interestingly, they can also use the data to ascertain with combinations of players work best together.
“Combinations of players, we just see as a bit of a model. As we add data to that model we can increase the accuracy of it. In that sense we’ll use machine learning techniques to look at how players cluster their performances together.”
Machine learning means the club can model a player’s injury risk based on history and training load. It can even predict a player’s career longevity based on a number of different factors. But recruitment strategy is perhaps where it is at its most influential.
“When you think of that area you go to Moneyball straight away,” he continues “We weren’t doing a lot of that in Australia, but we do now. We put more resources into that around 2015.
“We have students that watch and record things from a coaching perspective around the league as a secondary scout, but we also have students that are crunching the numbers on that data to create a model.
“We look at things that model around how similar that player is to other players we already have on our list (squad). We obviously want to have a balance in our roster.”
Every AFL club must obey a salary cap, so Robertson says his team must “forecast how good some of the current players are going to be and what they’re going to cost to keep them.”
The club’s recruitment strategy is to go for versatile players who can play in multiple positions.
“We want to be predictable to ourselves, but unpredictable to the opposition. That’s the main mantra we’ve got here. If a player who’s playing a specific role for us is having trouble or being targeted by the opposition, we feel like we always have Plan B, C, D, E – a number of different plans we can use in that scenario.
“It’s our greatest strength in our list. It worked pretty well last year so it’s certainly something we want to continue.
“It’s kind of our forte to go for a older player provided he fits in with our list.”
The Bulldogs also place far more importance on technical ability and intelligence than physical characteristics.
“From a recruiting perspective it’s far more about the skill side of things than the physical. Australian football went through a phase where everyone was looking for the best athlete they could get their hands on, but it’s still a skill-based sport.
“That’s first and foremost what we’re looking for – players that can kick and handball, because that’s more fundamental to the game. It’s probably easier teach someone to run better later down the track rather than teaching them to kick.
“It helps because we’ve got a really strong idea at the club for the types of players we want to have here. It sounds obvious but a lot of clubs don’t have that. We can evaluate our system against those values we’ve got here.
“It’s just also basic intelligence, we can pick out the education we’re going to provide them and make sure they’re going to understand it quickly. If they do that they’ll be more likely to play for us earlier.”
It’s a fast-moving industry and Robertson believes we’re not far away from the introduction of tattoo-like sensors which could measure kicking, heart rate, sweat rate and plenty more.
“They would work in much the same way as wearing sensors in the gym, except you wouldn’t have to strap them onto your body.
“A little bit more complex is what type of sensors are on the body, how we get battery to them, how we wash them, these types of things. There’s already research going on in this area. They could probably do it now, it’s just about making it cost-effective.
“I can’t see a world in five or ten years time where we’re still putting big trackers on people’s backs under a jersey. I think we’ll be doing it through these sensors on the skin, no questions about it.
“It would be easier to have it under the skin but that might be a step too far.”
Robertson doesn’t overstate the role machine learning had in the Bulldogs’ AFL Grand Final victory, although it did influence the team’s style of football and help them identify some of the players who contributed to the win.
The Western Bulldogs have shown the positive benefits advanced sports science can bring to a team. Australia is the world leader in this area and it may be some time before artificial intelligence makes its way to an amateur game like GAA, although it might not be as far away as you think.
Who knows, maybe post-All-Ireland final victory speeches in the not too distant future will even include a word of thanks to the county computer – for all its efforts during the year.
The42 is on Instagram! Tap the button below on your phone to follow us!
GAA Analytics Roundtable: How numbers have changed the game
What started as a backpacking trip turned into wearing Jim Stynes’ number at Melbourne FC
To embed this post, copy the code below on your site
AFL analytics Aussie Rules GAA machine learning