Fun with clusters: What K-means clusters can teach us about Knick comparisons

Playing the player comparison game is always fun, but what can a method called K-means clusters do to help make comparing Knicks to other players around the league more scientific?

The Knicks are a basketball team. I’m not so sure I’d be willing to stretch and call them a “good” basketball team, especially after this tumultuous season. But, they are certainly a basketball team. They’ve hit many ups, and many more downs. However, one of the cool things we do as fans is project. I wouldn’t call it cope, but rather hope. Hope for a better future. When we suck in year n, we  look at what the team may be in year n+1. And what matters there? The players. 

Basketball players go through slumps and streaks all alike. RJ Barrett is tearing up the league right now, yet he looked like a bench player a couple months ago. While that certainly is worth discussion, I’m going to stick to something else in this piece. Player archetypes and how they may apply to our Knicks.

We often group players together without really thinking about it too much. For example, if you saw a player like Alec Burks on your screen, you’d say just based on his play style: “Oh, he’s closer to a guy like Eric Gordon than he is LeBron James.” Fair enough. What if we had a way to quantify this, though? Oh wait, we do!

Allow me to present K-means clustering. Few of you have probably heard of this, but essentially it allows for one to create groups or clusters based on the value of several input variables. (I followed this fantastic tutorial from Alex Stern in my work, definitely worth checking out.) I wanted to see what NBA players play similar to each other, and to do that I needed to choose inputs. I focused on the 2021-22 season only, and here’s what I used (all per game stats):

• Minutes

• 3P%

• 3PA

• 2P%

• 2PA

• Steals

• Blocks

• Offensive Rebounds

• Defensive Rebounds

• FT%

• Free Throws Attempted

• Points

• Turnovers

• Assists

I thought this group of stats would give me the best look at a player’s production, tendencies, and workload without diving into too many advanced stats. We get a solid look at their scoring efficiency, defensive activity, and rebounding work. Having chosen my features, the next step was to make my clusters. I ended up with 9 total clusters, and here’s the feature importance chart for each. The easy way to interpret this is, the higher up a dot is for a feature, the more important it was for that cluster. 

I know this can be a lot to take in, so I’ll break it down by each cluster with a few phrases for what type of player each represents:

Cluster 1: High defensive activity, can shoot, solid role players

Cluster 2: High usage studs, not much else

Cluster 3: Your JAGs (Just A Guy)

Cluster 4: Playmaking, star guards that are also great volume shooters

Cluster 5: Trust them with the ball outside the arc, not in it

Cluster 6: Bigs that may not be your first option, but can definitely score

Cluster 7: Rebounders that’ll dominate the paint on both ends

Cluster 8: 3-point specialists with other plus traits

Cluster 9: Low minute forwards that’ll pass, but not your best

I won’t front at all, that was pretty hard to characterize. But hopefully this next look makes things a bit easier to understand. I’ll show the clusters and some of the big, recognizable names in each:  

Now, this is NOT to equally compare players with no context. No, Julius Randle is not LeBron James, as much as we may want him to be. Rather, it gives us an understanding of comparative roles across the league and how players may relate to each other in terms of efficiency and usage. Randle may have similar usage to players like DeRozan and LeBron, but we know his impact isn’t the same, which is also a limitation of using box score stats. That said, though, we can draw some neat conclusions, and I’ll do just that from a solely Knicks perspective. I’ll mention a bunch of key players on the Knicks and some comparisons that may be reasonable based on their respective clusters.

RJ Barrett

Who else would I start with? He’s been positive recently, and deserves the attention. RJ ended up in Cluster 5, the same one as Alec Burks. The main reason for this is honestly how cold he was before a month ago. His efficiency numbers really tanked, and that hurts his projection. However, he could have just as easily turned a page a month ago, and hopefully so. For now, though, some other players in his cluster include Tyler Herro and Tyrese Maxey. Jalen Brunson, too! The way these guys have been playing recently, there’s not much room for negativity. 

RJ’s still got some steps to take to reach that next level, that Jimmy Butler group of guys. He’s taken that step recently, now it’s up to him to make it stick. Even without it, if his median projection ends up like Jalen Brunson’s career, I wouldn’t complain too much. Expectations are key.

Immanuel Quickley

As seen above, Quick found himself in Cluster 8. This group of players can SHOOT. A cool comp was Patty Mills for him. While I do believe Quickley has a lot more to offer on the defensive end of the game due to his wingspan and length, the offensive comparison is definitely interesting. Mills has been an unreal shooter throughout his career and a very solid, secondary playmaker. Again, hard to be upset with that for a guy who has drafted near the end of the first round. Our good old friend Frank Ntilikina shows up in this cluster too.

Cam Reddish

The recently-acquired wing from the Atlanta Hawks, Reddish is hard to project. Not so much his fault, but he’s never really gotten a fair shake and a consistent role with a team. Atlanta didn’t have it for him, and New York’s been more of the same. Nonetheless, he was also in Cluster 5 with Quickley! There were a couple comps I liked here for him, but really liked Terrence Ross and Cameron Johnson. Johnson’s game has evolved this year with the Suns, he’s taken a step up as a shooter and even filled the scoring void recently without Devin Booker in the lineup. Ross has always been a scoring option off the bench for most of his career, known for his hot stretches. It’s hard to put a ceiling on a guy with the physical tools of Reddish, but I’d expect him to blossom into something like Johnson, with a real chance for much more. 

Mitchell Robinson

Mitch has been awesome this year… in the time he’s actually played. We know what he brings to the game, what he can do, but the injuries do hurt. If he even gets one full season healthy, that’d be massive. He got matched with Cluster 7, which included guys like Rudy Gobert, Clint Capela, and Steven Adams. While he likely won’t reach Gobert level, I could definitely see him playing an Adams type role when healthy. That “when healthy” part is key, but man it’s hard to not get excited about what he brings to the floor. The shot blocking and defensive gravity is there already, and I can’t wait to see him get to put it all together. 

Julius Randle

The last guy I’ll cover, Julius Randle might be the most polarizing player the Knicks have had in a while. He’ll have games like he did against Sacramento where he can’t be stopped, but he’ll also sometimes put up clunkers when you need him most. I think a lot of Knicks fans got too bought into the hype last year, rightfully so, when in reality that was his 100% outcome. And frankly, the names in Cluster 2 are way ahead of him in terms of impact on the court. Guys like Joel Embiid, LeBron, Ja Morant, and Giannis Antetokounmpo are much, much better than him. 

The only reason he ends up in this cluster is due to his usage with the Knicks. He’s being used as the first option, when he probably better suits a second or third option role for a contending team. Different play styles, but a Pascal Siakam role on the 2019 Raptors would be interesting to see Randle in. Don’t get me wrong, Randle has a lot of positives to his game that make him a solid basketball player, but it’s hard to see him taking a leap at this stage in his career.

There are a couple other players that didn’t qualify due to minutes played, injuries, or other factors — one being Quentin Grimes. Grimes rocks. The perfect 3-and-D player, he’ll play the Danny Green role so well on a contending team. He’s even shown flashes of his playmaking, and if that becomes consistent, look out. Burks also ended up in RJ’s cluster, but I really don’t feel like talking about Burks given the way Tom Thibodeau plays him, so I won’t. 

Overall, the Knicks are in a weird spot. They came into this season with playoff hopes, and now look destined for the lottery. Coach Thibs lost his magic touch from last year. He’s struggled to find the right rotations, ignores data in favor of our young guys, and just seems discombobulated this year. Yet, there’s bright spots. RJ’s taking a leap to the next level. Quickley’s found his shot again. Mitch has shown his ceiling multiple times. Grimes rocks. Did I mention that yet? He rocks. 

In the end, I hope this nerdy look at basketball left you with something interesting, something to maybe inspire hope in such a gloomy season. Go Knicks.