SFU computing science professor Oliver Schulte (right) and student Zeyu Zhao (left) were pleased to see NHL defenseman Erik Karlsson's performance being praised during the playoffs as their research predicted him to be a star player.

SFU researchers predicted NHL player’s playoff performance

May 31, 2017

While fans were heartbroken to see the Ottawa Senators get eliminated from the NHL playoffs last week, they were also content. That is because no one had expected the team to make it to the Eastern Conference finals and here they were in double overtime of Game 7, so close to advancing to the Stanley Cup finals.

One of the star players that has been the talk of the playoffs is Senators’ defenseman Erik Karlsson, whose performance is being credited for the fact that his team was able to advance this far.

“Some people talk about how his performance is a surprise,” says SFU computing science professor Oliver Schulte, who specializes in sports analytics.

“For me it wasn’t a surprise. In our paper, we pointed out Karlsson as a star defenseman whose performance is worth paying attention to.”

The paper that Schulte is referring to summarizes the research his team conducted last summer before the 2016-2017 NHL season began. The researchers used data from the 2015-2016 season to cluster hockey players based on their movement patterns. They then used location data and machine learning techniques to develop a player performance assessment system.

This system could help teams make better coaching and trading decisions as it can predict a player’s performance in the upcoming season.

Schulte presented the research at the MIT Sloan Sports Analytics Conference this March alongside co-author Zeyu Zhao, an undergraduate student who will be graduating from the computing science dual degree program in June. Zhao is also a recipient of the Dean’s Convocation Medal and will begin his graduate studies at Harvard University in the fall.

In their model, the researchers calculated the scoring impact (SI) for each player, a metric that takes into consideration the player’s actions—both offensive and defensive—and the impact they have on his team scoring the next goal.

This model placed Karlsson at the top of all players.

“He’s not only on top of the list, his SI stands out compared to the others,” points out Schulte. “This result is based on data from the 2015-2016 season and clearly predicts him to be the best defenseman in the league for the 2016-2017 season.”

As the Senators entered the playoffs and kept advancing toward the Eastern Conference finals, it was exciting for Schulte and Zhao to see Karlsson’s exceptional performance continue—especially since he was playing with hairline fractures in his left heel—as it supports their model’s prediction.

So does this mean they can predict who will win the Stanley Cup finals?

“If we take the data from the first three games between Pittsburgh and Nashville, we could have enough data from the matchup of these two teams to make a prediction,” says Schulte.

“So yes, theoretically we can. But that’s not what we’re focusing on at the moment. I’d rather watch the games to see who wins.”

One of Schulte’s other projects involves working with HockeyData Inc., a startup formed by SFU students, to analyze data from the American Hockey League (AHL) and predict player performance. This is especially helpful for NHL teams who are looking for star players to recruit. HockeyData recently reported having the Washington Capitals sign on as their first NHL client team.

Schulte’s research is conducted in collaboration with SPORTLOGiQ, a player tracking and analytics company, which collects the sports data that Schulte’s team then analyzes. This research is supported by a Natural Sciences and Engineering Research Council of Canada (NSERC) Engage Grant.


Related Video