The scene is familiar to anyone who has ever watch a baseball game: A batter is inching his way from first base, slowly, slowly. The pitcher sees him and fakes a throw, sending the batter scuttling back to his safe position. Moments later, the batter begins to inch his way out again, further, further. Then he breaks into a run, steals and slides safely into second base.
Now, though, fans tuning into a stream of a game can hear MLB broadcast partners analyze the probability that a runner will be able to successfully steal. MLB and its partner Amazon Web Services have developed a deep neural network that can predict stolen base success using metrics such as a runner’s speed and burst, a catcher’s pop time, a pitcher’s velocity and handedness, and lead-off distance.
The predictive model uses data from MLB Statcast, a league-wide player-tracking product launched in 2015 that leverages cameras and tracking technology in all 30 stadiums to capture the location of players, umpires, and the ball up to 10,000 times per second.
“Twenty-one inferences were computed in 128 milliseconds using Amazon Sagemaker [a machine learning platform]. And we were able to do that quickly enough so it could be into live game content,” said Justin Burks, AWS’s director of sports marketing and strategic programs, at the San Francisco 49ers and SportTechie Horizon Summit this week. “The fan reaction was super positive.”
The same type of AWS technology is also used to power the NFL’s Next Gen Stats. In 2017, the NFL announced that AWS’s machine learning technologies and cloud-based data analytics would be used to visualize gameday data collected from the Zebra RFID chips located every player’s shoulder pads.
The league has since incorporated deeper insights into its storytelling using real-time location, speed, and acceleration data. One of its AWS-powered prediction statistics is pass completion probability. Using measurements that include time to throw and the separation between a receiver and defensive back, the NFL can now calculate the likelihood that the receiver will catch the ball.
SportTechie Takeaway
Amazon Web Services provides a machine learning tool that is helping leagues to not only make sense of the mass quantities of player-tracking data they’re now collecting, but also to set predictions. As fans continue to crave real-time data about players on the field and as gaming, whether fantasy or real-money gambling, expands in popularity, the type of storytelling AWS can power will likely become increasingly valuable.
MLB and the NFL already have robust player-tracking systems in place, allowing them to easily use AWS for predictive statistics. Other leagues could soon be in the position to leverage AWS, too. The NHL is expected to begin chipping its players next season, and last weekend, MLS Commissioner Don Garber said he’ll be seeking a more liberal use of player data in the league’s next CBA negotiations when the current contract expires in January.