Computing Science Researcher Uses Machine Learning to Analyze Sports
By: Andrew Ringer
In the world of sports, teams and players are constantly adapting to try to get an advantage over their competition. While this innovation is often thought of in terms of sports nutrition and training, sports analytics has also been a key factor in terms of helping teams, players and fans perform and understand sports at the highest level.
For athletes, sports analytics research can help them in improving their play, as well as negotiate salaries. For coaches and management, this research can help identify players and strategies that will improve the team. For fans, sports analytics can be used to greater understand the sports that they follow.
SFU computing science professor Oliver Schulte is a leading researcher in sports analytics. Using machine learning, he and his students build models to evaluate players and teams. Their models estimate how likely a team is to score at any point in the match. Players are ranked by how much they increase or decrease their team’s chance of scoring. These models are mostly applied to professional ice hockey but have also been applied to soccer and volleyball.
What makes their machine learning models unique is that they take into account match context, meaning that goal-scoring opportunities may be weighted differently depending on the time left in the game and the current score. Similarly, actions are weighted differently depending on the context. For example, a pass close to the opponent’s goal is worth more than one close to the team’s own goal. They are also working to have their models consider player context as well, meaning that goal-scoring opportunities will be considered differently depending on which player takes the shot.
The complexity of compiling and using the data from sports games is what originally made Schulte interested in researching sports analytics.
“Sports analytics is a more complex domain than what we typically look at in machine learning,” says Schulte.
“You have different types of agents involved: different players in different positions and coaches. It also happens in time and space which makes for a dynamic and rich data set.”
Recently, Schulte and his team received funding from the Canadian Statistical Sciences Institute (CANSSI) to continue their research in sports analytics in collaboration with other sports analytics researchers across Canada.
“This funding will help our students in moving forward in their research,” says Schulte.
“It also makes us part of a sports analytics community that we can contribute to and learn from.”
Going forward, Schulte and his students will work as part of this research community to continue to address some of the problems and challenges in sports analytics.