Injury prediction

Using a range of performance data to predict athlete injuries

Premiership Football, Sport

Given the high intensity and accident-prone lifestyle of football players, health monitoring is a top priority for them.

Download PDF

Goal

Football players are among the best-conditioned athletes in all of sport. Still, the demands of practice and competition over the course of a 38-game regular season (not including the post-season) can take a toll on an athlete’s body. The fitness coaching team of one premiership football club was seeking new methods to evaluate the health of its players and identify the physical metrics that might signal impending injuries. In particular, the team wanted to evaluate whether certain strength and flexibility tests, plus saliva samples, could assess the likelihood of injury.

Insight and Action

QuantumBlack built a surrogate model to determine the onset of injuries based on a player’s history. Using objective medical markers and information from prior injuries, we identified the features that correlate to injury onset in the hamstring, upper leg, and lower leg. We used blind historical testing across two years of data and four football squads to hone the model, which correctly forecast 170 out of 184 non-impact muscle injuries.

Results

  • 90Percentimprovement in accuracy of forecasting non-impact injuries
  • 64Percentcorrect categorisation of (actual) injured players (type II error)
  • 73Percentcorrect categorisation of (actual) healthy players (type I error)

The impact of our analytics model on the team was significant: around two-thirds (64 percent) of the time, it was able to provide the correct categorisation of injured players. Similarly, it determined the correct categorisation of nearly three-quarters (73 percent) of the healthy players.