Automatically Recognizing On-Ball Screens

Read full paper here

See slides here

Former coach Stan Van Gundy once said of his Magic team, “[The pick and roll is] what we’re going to be in when the game’s on the line. […] I don’t care how good you are, you can’t take away everything”. A staple in the modern NBA, the pick and roll is used dozens of times every game. A deep quantitative analysis of the pick and roll could have a dramatic impact on how it is used in offenses as well as how it is defended. However, with over 50,000 pick and rolls occurring in a single season, such analysis is labor intensive. Using machine learning techniques, we leverage the massive amount of data collected by the new player tracking systems – installed across all 30 NBA arenas – to develop a system for automatically identifying when a pick and roll occurs. The system we present is both a tool that will allow us to more easily identify statistical trends with respect to the pick and roll, and a tool that can help us determine which teams run the pick and roll similarly. More generally, it is the basis of a framework for recognizing other patterns of play in the NBA.

Back to Videos