A Spatiotemporal Analysis of Professional Basketball

• Activity Number: 154
• Type: Topic Contributed
• Date/Time: Monday, August 4, 2014 : 10:30 AM to 12:20 PM
• Sponsor: International Society for Bayesian Analysis (ISBA)
• Author(s): Andrew Miller*+
• Keywords: point process ; spatiotemporal ; basketball ; nba

Abstract

We describe the analysis and modeling of NBA player tracking data, a rich (~100 GB) dataset of space-time locations of basketball players and events from hundreds of games over the last two seasons. This talk will discuss our strategy to balance data size, model expressiveness and tractability with the goal of measuring the spatial ability of individual players. Our analysis focuses on measuring each player's defensive ability. First, we develop a method to discover defensive match-ups at every moment in over one hundred thousand possessions in an unsupervised way. We then couple this information with offensive events to model defensive ability. We directly measure two different dimensions of defensive ability, shot deterrence and shot contesting, and how they vary spatially across the court. Additionally, we will discuss player representations and their role in establishing realistic priors for these models.