Artificial Intelligence learns the value of teamwork to form efficient football teams
Machine learning experts from the University of Southampton are optimising football team selection by using AI to value teamwork between pairs of players.
The new approach uses historic performance data to identify which player combinations are most important to a team, generating insights that can help select teams' most efficient line-ups and identify suitable transfer targets.
The study, led by PhD student Ryan Beal in the Agents, Interaction and Complexity (AIC) Group, has developed a number of teamwork metrics that can accurately predict team performance statistics, including passes, shots on target and goals.
Researchers presented their findings and hosted an AI in Team Sports workshop at this month’s Association for the Advancement of Artificial Intelligence (AAAI) Conference in New York.
"We have tested our methods from games in the 2018 FIFA World Cup and the last two seasons of the English Premier League," Ryan says. "We found that we could select teams using the AI in a similar fashion to human managers and then also suggest changes that would improve the team.
"When looking at the results for the Premier League, the teamwork analysis identified Aymeric Laporte as one of the key players for Manchester City. He has been injured for much of this season which may explain their downturn in form compared to last season."
The Southampton team have used a number of machine learning techniques to assess teamwork values from the historic data and found that teams with higher teamwork levels are more likely to win. They then trained an optimisation method to assess the teamwork between pairs of players and compute a number of new metrics that they compare in their latest paper.
"While this work could be used as a tool to assist football managers, we think that the approach could also be extended into other domains where teamwork between humans is important, such as emergency response or in security," Ryan says.
Ryan has also presented his work to sporting industry experts at the StatsBomb Innovation in Football Conference at Stamford Bridge in October.
Ryan's work is supported by UK Research and Innovation (UKRI) and AXA Research Fund. The work was done in collaboration with Narayan Changder (NIT Durgapur), Professor Tim Norman and Professor Gopal Ramchurn.
Team sport performance is one of a several innovative AI research topics being explored in the AIC Group. In 2018, Gopal and Dr Tim Matthews revealed how machine learning algorithms can accurately predict team and player performance to finish in the top 1% of the Fantasy Premier League game, outperforming close to six million human players.