Where on Ice? Algorithmically Deconstructing NHL Shot Locations as a Method for Player Classification
Department
Statistical and Actuarial Sciences
Program
Ph.D. Statistics
Year
5
Supervisor Name
Douglas Woolford, Charmaine Dean
Supervisor Email
dwoolfor@uwo.ca
Abstract Text
Where do hockey players shoot from? How does this vary from player to player? We present the results of a study that uses data-driven statistical methods to investigate these questions. The locations of shots by National Hockey League (NHL) players from 2011 to 2017 are analyzed using a combination of an image recognition algorithm and spatial statistical methodology. An unsupervised classifier is applied to output from a spatial point process model in order to determine which shot locations best characterize a given player. We define the number of regions a priori, but the image recognition algorithm chooses the shape and location of shot regions. By analyzing how many shots a player takes from within each algorithm-defined region, we can compare the shot preferences across all players in the NHL. An interactive app has been developed to combine estimates and visualize the locations of shots from any offensive line (i.e. a left wing, a centre, and a right wing player). These results can inform coaches and goalies about probable shot locations from potential line combinations of an opposing team either when preparing for an upcoming game or even during a game. These results can also provide a metric for classification of players into different play styles and to differentiate the play styles of top players from other players in the same position.
Study completed
Supervisor Consent
yes
Where on Ice? Algorithmically Deconstructing NHL Shot Locations as a Method for Player Classification
Where do hockey players shoot from? How does this vary from player to player? We present the results of a study that uses data-driven statistical methods to investigate these questions. The locations of shots by National Hockey League (NHL) players from 2011 to 2017 are analyzed using a combination of an image recognition algorithm and spatial statistical methodology. An unsupervised classifier is applied to output from a spatial point process model in order to determine which shot locations best characterize a given player. We define the number of regions a priori, but the image recognition algorithm chooses the shape and location of shot regions. By analyzing how many shots a player takes from within each algorithm-defined region, we can compare the shot preferences across all players in the NHL. An interactive app has been developed to combine estimates and visualize the locations of shots from any offensive line (i.e. a left wing, a centre, and a right wing player). These results can inform coaches and goalies about probable shot locations from potential line combinations of an opposing team either when preparing for an upcoming game or even during a game. These results can also provide a metric for classification of players into different play styles and to differentiate the play styles of top players from other players in the same position.