NHL’s new game head-to-head probabilities combine data, technology to enhance the viewer’s experience

The NHL is applying in-game head-to-head probability to broadcasts, a significant leap forward for the league in terms of data analytics and technology.

“Head-to-head Probability” leverages data collected by NHL Edge, the league’s player and player tracking technology, to create a graphic showing the chances of a player winning a head-to-head match or a team winning the right own the ball.

This is one of the first machine learning stats the league has developed in partnership with Amazon Web Services, whose artificial intelligence can generate in-game probabilities at sub-zero speeds. second.

“This is the first time that NHL and AWS have come together to build something that will happen before an event and give the probability of whether that event will happen,” said Dave Lehanski, NHL executive vice president of development. business development and innovation, told ESPN on Monday. “Typically we’re taking data from an event and quickly analyzing it to come up with some kind of insight. Even if we’re doing it in real time, we’re not actually doing it yet. appear before an event.”

Priya Ponnapalli, senior manager at Amazon’s Machine Learning Solutions Lab, says Face-off Probability uses more than 70 data points ranging from historical and in-game stats, as well as data context. Ponnapalli says the artificial intelligence takes 10 years of head-to-head results – more than 200,000 draws for all players in the league today – and uses data including a player’s success rate based on Head-to-head positions, home vs. away matches and specific opponent’s head-to-head history. It also affects personal data such as forehand, height and weight.

The NHL then adds in-game head-to-head stats to round out the data. In both historical and in-game statistics, there is additional context such as game situations, scores and head-to-head times.

Artificial intelligence uses the NHL player tracking system to determine who can face both teams and then instantly runs that data to generate probabilities, which are shared with broadcasters and fan.

Ponnapalli said there are challenges in creating this technology for hockey compared to other sports in which AWS already operates.

“The head-to-head prediction model has to be flexible to generate predictions as the game situation changes,” she said. “For example, if a player is disqualified from the head-to-head round due to a violation, the predictions must be updated to a new match based on real-time streaming sensor data. The predictions also happen at a high level. sub-second delay and can be triggered at any time. All this complexity has to be built in and the resulting solution needs to be flexible.”

The NHL believes its tracking technology provides a way to further educate fans about the game and gives broadcasters more opportunities to tell stories. With 50 to 70 face-offs per game, Lehanski says, and up to 20 seconds between downtime and head-to-head, there’s going to be plenty of opportunity to tell that story.

“If there’s a major key confrontation, we want to be able to come up with the probability who could win and how that probability might change if someone else does the confrontation. That would be extreme. It’s fascinating and really valuable to the viewer,” he said.

With machine learning metrics already in place, Lehanski says the underlying technology could be applied to other aspects of hockey to generate probabilities and predictions for broadcasts.

“Take historical data, combine it with real, live, in-game data, process it to develop an analysis or probability, and then display that data on the screen as a graphic,” he said. in less than a second. “It certainly opens up unlimited opportunities for us to broaden the way we look at all the events that happen in the context of a hockey game.”

These probabilities, and all the data collected by NHL Edge technology, have another intriguing application: sports betting.

The NHL predicts that sportsbooks will eventually generate additional backing bets around data gathered from player tracking. The federation has formal data licensing agreements with MGM Resorts International, FanDuel, William Hill and PointsBet. It also has a 10-year agreement with Sportradar as the official NHL and betting data rights partner.

Lehanski says that “the technology is there to bet on head-to-head matches,” but he cautioned that this type of bet is only one possibility at this point.

“If there is a betting organization that wants to assign a new bet type to head-to-head matches and can develop an odd number for the match, based on the probability of the outcome, there is, in theory, a reasonable amount of time. long enough through a mobile app, you can press a button and decide if you want to bet on the outcome of a head-to-head,” he said.

Lehanski said the challenges with real-time betting like head-to-head is the difference in time between a bet in the arena or by fans at home watching the broadcast with a few seconds delay. , as well as whether the NHL will ever allow bets on outcomes such as head-to-head matches.