Chapter 3 Dataset

The dataset is from the Duke University Men’s Basketball SportsVu tracking data. Features were created by taking snapshots of the game every 1/25th of a second and recording the player’s location, action, team, etc. Data was collected for each season from 2013-2016; the dataset totals about 132,000 observations and 98 features. Since the data is owned by the Duke Men’s Basketball team, the data is private and cannot be shared.

The dataset was presented in 3 different XML files:

Boxscore Data: This dataset shows the overall player statistics (assists, points, rebounds) for each game.The Boxscore dataset was used as a reference for player performance when modeling the posterior means of the latent eigenvalue model.

Play by Play: This dataset provides a moment summary (dribble, foul, pass) at time t for each game. The Play by Play dataset was used to divide the raw data into possessions, which was then converted into individual passing networks.

Sequence Optical: This dataset provides the locational summary of each player for each game. The Sequence Optical dataset was used to map the passing order for each possession.