Football may be unpredictable, but its injuries aren’t. Artificial intelligence can work out when players are likely to get hurt so that coaches can let them rest.
A machine learning algorithm developed by data scientists at the University of Pisa, Italy, and FC Barcelona recently predicted 9 out of 14 injuries sustained by an elite Italian football team during a single season.
The researchers fitted the team’s 26 players with GPS sensors during their training sessions to measure how far and fast they ran, how often they accelerated and decelerated, and the impact they had with the ground and other players. They also recorded information about their age, height, weight, role on the field, injury history, and minutes played in the last game.
Advertisement
As the season progressed, the researchers’ algorithm learned to detect patterns between these variables and players getting hurt. By the end, it was able to predict about 60 per cent of injuries.
Coaches could use the algorithm to work out when they should rest a player or lighten their training load, says at the University of Pisa, who led the research. For example, if Edinson Cavani’s coach had known he was at risk of straining his calf in the recent World Cup match against Portugal, he could have replaced him sooner, he says.
Missing the power
Sports injury forecasters have been used in the past, but their precision has typically been less than 5 per cent, says Rossi. This is because they have usually relied on a single variable – like the number of balls bowled during cricket – to predict injury risk, he says. “They miss the power of combining different training workload measures,” he says.
Existing forecasting tools have also been hampered by high false alarm rates that cause them to incorrectly predict injuries, says Rossi. “Stopping players unnecessarily is a condition clubs want to avoid, especially for key players,” he says. The new forecaster halves the rate of false alarms, he says.
The Italian team that trialled the AI can’t be named because it doesn’t want to give away its competitive advantage, says Rossi. The same goes for three other top-level European football teams that have started using it, he says.
Rossi and his colleagues are now investigating whether they can make their injury forecaster more accurate by including additional variables like heart rate and degree of sweating. They are also interested in seeing whether they can predict the particular type of injury.
Microsoft launched its own in June last year, which uses GPS and heart rate data, as well as players’ self-reported sleep, mood and muscle soreness. It’s now being used by US football team Seattle Reign, Spanish football team Real Sociedad, the Seattle Seahawks NFL team, and the Australian cricket team.
Their results are mostly confidential, but Seattle Reign says it only suffered one player injury in the season after it adopted the technology.
PLOS One
