Predicting and identifying the most important factors on player performance in the National Basketball Association using Machine Learning.

Abstract

The field of data analytics in basketball has been on a meteoric rise in recent years. Improved optical tracking technology has allowed data collection on a scale that has never been seen before. As a result, a plethora of work related to how player performances get affected by various in-game factors have been published. The rush of incoming data has also allowed for teams to get a better handle on how they can mitigate certain factors to increase a player's performance. However, no one has been able to capture the relationship that extrinsic factors, such as weather, precipitation, or attendance could affect player performance, and rank which factors have the most impact. In this paper, I will describe the process of creating a novel dataset that consists of game attendance, opponent information, weather information, date information, and player form. With these data points, I also identify which group of factors have the most impact on player performance. I will also discuss the process of creating a supervised model, which has the potential to predict how well a player will perform in a game based on the input external factors for a particular game.