DATA 621 Blog 4: NBA Player Similarity using PCA

David Quarshie

Intro

For the last three blogs I looked at how to predict NBA players efficiency. The results showed that how a player performed offensively had the most bearing on predicting efficiency. But now that we know how a player can become more efficient, are their certain NBA players someone can model their game around to become more useful? Basketball fans can watch a healthy amount of basketball to know which players the best is according to certain skills. Using cluster analysis, they can also see which of these players are most alike according to their stats.

The main goal behind Principle Component Analysis (PCA) is to drill down the data in the dataset into similar groups and see how much these groups differ from each other. For this blog PCA plot, the color code was kept for the player’s original position and their PER, when large shows how efficient a player is, is used for sizing.

PCA Plot

Looking at this PCA plot from the NBA dataset we can see most players grouped together at left side, but some players are branching away. For example, Russell Westbrook is furthest away from the main group showing that he has metrics that are very different from the other players. This makes sense as Westbrook was the MVP of the leauge and registered record numbers in the 2017 season. Another player, James Harden is also far from the group, his playing style led him to become the 2018 NBA MVP. We can also see a group of centers begining to cluster in the upper left. These players play the game extremely different from the others but also play the same way as each other, making them cluster in the same area. Using a PCA plot allows us to see players that play alike.