From Hardwood to Heatmap: A Bayesian Dive into the Spatial Dynamics of Basketball Shot Selection

Statistical Seminar

Organizer：

Yunan Wu 吴宇楠 (YMSC)

Speaker：

胡冠宇 副教授密歇根州立大学概率统计系

Time：

Fri., 14:00-15:00, Dec. 12, 2025

Venue：

C548, Shuangqing Complex Building A

Title:

From Hardwood to Heatmap: A Bayesian Dive into the Spatial Dynamics of Basketball Shot Selection

Abstract:

Basketball shot charts offer rich spatial and contextual information that reveal nuanced patterns in player behavior and game strategy. In this talk, we present a novel Bayesian framework for modeling and interpreting the spatial dynamics of shot selection. We begin with a log Gaussian Cox process (LGCP) model to jointly analyze the locations and outcomes (made/missed) of shots across multiple games, capturing spatially varying covariate effects through hierarchical Gaussian processes. To facilitate efficient inference, we design a custom Markov chain Monte Carlo (MCMC) algorithm using a kernel convolution approach.

Building on this foundation, we introduce a complementary modeling strategy using Functional Bayesian Additive Regression Trees (FBART), which provides flexible, nonparametric regression capabilities and uncertainty quantification. To improve scalability and accuracy, we further propose the Adaptive Functional BART (AFBART) model, which employs adaptive basis functions to better capture the nonlinear and nonstationary nature of shot selection behavior.

Through extensive simulation studies and a real-world case study examining the shot charts of NBA legends Stephen Curry, LeBron James, and Michael Jordan, we showcase the power of our approach in extracting actionable insights. Our framework reveals how spatial context, playing conditions, and opponent characteristics influence shooting efficiency—offering practical tools for analysts, coaches, and players seeking a competitive edge.