This is a review of the neural networks research conducted by Kuan-Chieh Wang and Richard Zemel.
Offensive strategies in basketball are complex and dynamic, ever changing in response to the defense. Not only do the plays evolve but also players can change roles as the play on the court continues. A further complication is that plays are variable in how long they take to set up and how long they take to execute as they respond not only to the defense but also to where teammates are placed on the court. A turn of events like fouls and player errors further complicate the process. Recognizing the multitude of offensive plays of a team requires a deep understanding of the game. This information cannot be recorded as easily as other data such as number of shots and shot efficiency and therefore requires a more advanced process.
Player tracking data from SportVU was entered into a neural network in picture form. Neural networks are relatively easy to use and are capable of solving difficult problems. Neural networks do not need the various individual features of each play to be coded; instead, they are capable of learning the features themselves by analyzing the data. The pictorial data was separated into individual steps and information regarding the position of every player on the court at that moment was noted. The position of the ball in relation to each player was also recorded. The individual steps were then combined in player position sequences, which are presented as lines on a map of a court. From these lines, the model is able to determine which offensive plays are being utilized. The data used came from the Toronto Raptors 2013-2014 season. Eleven classes of offensive plays were included and 1435 sequences from the data fit into these categories.
The neural network model that was developed based on the 2013-2014 season’s data was tested in the 2014-2015 season. The model proved that with limited data, it was able to distinguish the different offensive plays.
The model requires more information to analyze offensive plays than a scout does. However, they are capable of watching multiple games in a short period of time and are able to provide the usable data within a fraction of the time required by its human counterparts.
Being able to determine and classify the offensive plays allows analysts to scout for talent that would be a good fit for their team. This information will allow coaches to analyze their team’s effectiveness in using the various offensive plays, which will allow them to refine their playbook and provide information to their players in order to help them improve their play. The information will also create scouting reports that contain more detailed and useful information.
How mature is your team’s analytics program? Take the Sports Analytics Maturity Assessment.
Learn about the Groundbreaking Sports Analytics Model coaches and sports analysts are talking about!
Learn all about Sports Analytics here.