Applications of machine learning : basketball strategy

Abstract

While basketball has begun to rapidly evolve in recent years with the popularization of the three-point shot, the way we understand the game has lagged behind. Players are still forced into the characterization of the traditional five positions: point guard, shooting guard, small forward, power forward, and center, and metrics such as True Shooting Percentage and Expected Shot Quality are just beginning to become well-known. In this paper, we show how to apply Principal Component Analysis to better understand traits of current player positions and create relevant player features based on in-game spatial event data. We also apply unsupervised machine learning techniques in clustering to discover new player categorizations and apply neural networks to create improved models of effective field goal percentage and effective shot quality.