Disney develops new analytics tool for basketball, soccer

Chuck Bednar for redOrbit.com – Your Universe Online
Coaches and athletes are increasingly turning to sports analytics in order to gain a competitive advantage, and new computer models developed by experts at Disney Research Pittsburgh can accurately predict what players will do in basketball games and soccer matches.
In one study, the researchers demonstrated that they could use player tracking data from more than 600 basketball games from the 2012-13 National Basketball Association (NBA) season to build models capable of making accurate in-game predictions as to whether a player is more likely to pass the ball to a teammate or to take a shot based on specific situations.
In another, the same researchers used a full season’s worth of ball and player tracking data from a professional soccer league to analyze team behavior instead of that of individual players. They reviewed over 400 million data points to create a system which could detect and visualize team formations accurately enough to identify teams based on play style 70 percent of the time.
Both studies will be presented at the IEEE International Conference on Data Mining. which will be held in Shenzhen, China from December 14 through December 17. Patrick Lucey, an associate research scientist at Disney Research Pittsburgh and Yisong Yue, an assistant professor of computing at California Institute of Technology, were among those involved in the research.
In a statement, Lucey explained that that this type of automated, data-driven analysis could be used to back up a coach’s own intuition, but it could also be used to educate players during practice, to scout opposing teams or to plan for specific in-game situations.
During the basketball study, the authors were presented with the challenge of modeling the behaviors not just of the player in possession of the ball, but of his teammates and members of the opposite team playing defense and their shifting on-court positions as well.
Yue and his fellow researchers used “a machine learning approach in which the models were trained based on the tendencies of each player to take shots or pass or receive passes in certain locations,” they explained. “It also incorporated such factors as how those tendencies varied in the presence of opposition and on the duration of their possession of the ball.”
This approach used what is latent factor model, which not only makes accurate predictions but also determined what factors could be interpreted and linked with known intuitions of basketball game. The authors of the soccer study faced a different challenge, since so few points are scored. It required an understanding of what takes place during the game in non-scoring situations.
Lucey noted that simply focusing on individual players can be somewhat misleading, as the athletes can sometimes change positions during the course of the game, and the decisions they make will be based on the responsibilities and expectations that come with each position. Their decisions will also differ based on the relative positions of their teammates and opponents.
“Simply evaluating the player without considering that context will yield meaningless statistics,” the Disney researchers noted, adding that they “developed a ‘role-based’ representation of teams that doesn’t track individuals but instead automatically identifies the players in each position and how they play that position. This provides a view of the team’s behavior as a whole and also provides more meaningful, contextual information about individuals.”
—–
Follow redOrbit on Twitter, Facebook, Instagram and Pinterest.