​Algorithms Can Predict In-Game Actions of Sports Teams’ Players With High Accuracy

By Finn Seeley ’25

This October, scientists at Cornell University’s Laboratory for Intelligent Systems and Controls developed algorithms that can predict in-game actions of volleyball players that has been tested at an 80% success rate. The lab is now expanding the project’s applications to include the school’s ice hockey teams. These algorithms combine visual data, such as an athlete’s location on the court, with more implicit information, such as a player’s role on the team, to make predictions in regard to what that player might do next.

The leader of the project is Silvia Ferrari, the professor of mechanical and aerospace engineering. About her project, she states that the “Computer vision can interpret visual information such as jersey color and a player’s position or body posture.” In addition to using real-time information, hidden variables, such as team strategy, are also able to be integrated for the benefit of the predictions. The algorithms extract data from volleyball game videos and then apply it to new games to help them calculate these predictions. They can predict multiple actions with an accuracy of greater than 80% over a sequence of up to 44 frames.
 
In the future, these algorithms might be able to help teams prepare by evaluating game footage of their opponents, and then using their predictions to practice various game scenarios and specific plays. Ferrari is currently working with the men’s hockey team at Cornell and using videos of their games to further develop the software. The director of operations for the team, Ben Russell, believes that “this project has the potential to dramatically influence the way teams study and prepare for competition.” While it is unknown how successful these algorithms will be in predicting hockey and other sports, it is known that collegiate athletics are advancing greatly in terms of what is available for their use.