Written by :
Published on :
(AP) - The NHL is launching a new faceoff probability stat that uses a decade's worth of data and in-game analytics to determine the likelihood of each player in the circle winning the draw.
The league and Amazon Web Services unveiled the new feature Tuesday. It's the latest innovation in the puck and player tracking sphere and could be just the start of machine-learning technology in hockey with implications for gambling and countless other statistics.
Using information from the past 10 years of games in combination with the location of a faceoff, game situation and player performance, AWS software spits out percentage chances of who's coming away with the puck in a matter of seconds before the linesman drops it.
"Hockey is such a series of so many events that lead up to an outcome," NHL executive vice president of development and innovation Dave Lehanski said. “We’ve identified the faceoff as being this key component — the beginning of some of these incredible events, and that’s what we want to start analyzing.”
Faceoff probability graphics are set to debut on Sportsnet in Canada this week and on ESPN and Turner later in March. Down the road, the stat could become a gambling opportunity for real-time, in-game betting.
“The technology will enable that,” Lehanski said, deferring to league officials about what might become betting options in the near future. “The technology is at a place and will soon get to a place where almost any in-game event could be processed to potentially be a bet type — something that someone could take a bet on. ”
Priya Ponnapalli, senior manager of Amazon Machine Learning Solutions Lab, said the NHL and AWS built the model using data from thousands of faceoffs from the NHL's Hockey Information and Tracking System (HITS).
“This data source was significantly varied and complex, incorporating such information as a player’s home and away faceoff statistics, head-to-head matchup history, player characteristics such as height, weight, and handedness and game context such as the faceoff location, game score, and faceoff time,” she said. “This combination of historical success rates, player matchup characteristics, and game context brings together HITS and puck and player tracking data to provide a complete perspective for a faceoff’s dynamics.”
