Football Analytics with Python & R

Chapter 8. Principal Component Analysis and Clustering: Player Attributes

In this era of big data, some people have a strong urge to simply “throw the kitchen sink” at data in an attempt to find patterns and draw value from them. In football analytics, this urge is strong as well. This approach should generally be taken with caution, because of the dynamic and small-sampled nature of the game. But if handled with care, the process of unsupervised learning (in contrast to supervised learning, both of which are defined in a couple of paragraphs) can yield insights that are useful to us as football analysts.

In this chapter, you will use NFL Scouting Combine data from 2000 to 2023 that you will obtain through Pro Football Reference. As mentioned in Chapter 7, the NFL Scouting Combine is a yearly event, usually held in Indianapolis, Indiana, where NFL players go through a battery of physical (and other) tests in preparation for the NFL Draft. The entire football industry gets together for what is essentially its yearly conference, and with many of the top players no longer testing while they are there (opting to test at friendlier Pro Days held at their college campuses), the importance of the on-field events in the eyes of many have waned. Furthermore, the addition of tracking data into our lives as analysts has given rise to more accurate (and timely) estimates of player athleticism, which is a fast-rising set of problems in and of its own.

Several recent articles provide discussion ...