Identifying High Frequency Shooting Zones for NBA Teams using Clustering

摘要

We examine National Basketball Association (NBA) data to understand areas on the basketball court that teams tend to shoot successfully versus not. We call these areas high frequency shooting zones (HFSZs). These are found with play-by-play and box-score data. For any team, we define an HFSZ as a region on the basketball court that has high probability of a team making a successful shot. To achieve this purpose, we use unsupervised algorithms like k-prototype clustering and expectation-maximization (EM) clustering. We examine factors such as the player location, shot type, field goal percentage, game scenarios and the defense of the opposition team. For the purpose of reducing bias in the algorithm, we repeat the algorithm for different number of initial clusters. Our aim is to use the results to obtain trends associated with certain teams. This research has potential to assist teams in floor movement and in-game decision making.