Isaac
Last month

NBL Player Cluster Analysis

This stats work from Spatial Jam's Andrew Price is very cool.

NBL Player Cluster Analysis

It groups NBL players from the last four years into 12 clusters - Defensive Ball Handler, Offence-First Big, The Gunner, etc.

The project takes all boxscore statistics from the past four NBL seasons (2015-16, 2016-17, 2017-18 and 2018-19) and attempts to create an understanding of player types based on an individual's statistics and tendencies as opposed to simply looking at their 1 to 5 positions on the court. The modern game of basketball has significantly altered how we should assess and categorise players. It is becoming increasingly difficult to group players into the traditional 5 categories (Point Guard, Shooting Guard, Small Forward, Power Forward, Centre) - these groups have become far more fluid, with players’ skill-sets often fitting the description of multiple traditional positions.
Go take a look.

Topic #45915 | Report this topic


LV  
Last month

That's brilliant. Thanks for posting.

Reply #763502 | Report this post


PlaymakerMo  
Last month

I saw a TED talk years ago on this topic, regarding NBA player archetypes. My main criticism of that presentation is the same as this analysis in it's current form: I don't see how this information is useful beyond categorizing players - which fans, coaches and management do anyway.

I feel as though this isn't the ideal method for identifying optimal lineup/roster compositions, but happy to be proven wrong.

Reply #763506 | Report this post


Rod  
Last month

I suppose it would be helpful to have when creating a team or recruiting so you can have a good balance of players, not too many from the same cluster. Individually shows the areas you need to work on if you are in a cluster but you see yourself in a higher cluster.

also good to know for fantasy points as well

Reply #763510 | Report this post


Anonymous  
Last month

Really interesting and amazing that someone would put the time and effort into this (bonus points for r plots)

But as a researcher I must point out the potential interpretation error. Almost all recent Adelaide bigs (excluding Hodgson and Jacobson) are in one cluster. This could indicate that Joey is recruiting a very specific type of big for his system and would be fine with the current interpretation of this data. The alternative, however, is that the stats reflect the system they are in and the role they are asked from that team and NOT the potential or skill set of the player. eg. if Conklin played a season with Adelaide would his stats for that season reflect a "quality forward"?

Reply #763511 | Report this post


Perthworld  
Last month

The Lemanis Bullets2Boomers arse-licking cluster is missing.

Reply #763513 | Report this post


Captain Obvious  
Last month

Really interesting work.
I wonder what the Boomers squad would look like if you put each player in one of the clusters?

Reply #763514 | Report this post


Rod  
Last month

Is this piece saying these are the ONLY clusters or just the clusters that have appeared in the NBL over the past 5yrs?

The Boomers may have players outside of these clusters. Could there be players in this NBL season such as LaMelo with a unique skillset that would not fit into these categories?

Reply #763515 | Report this post


Anonymous  
Last month

The guys who made this would know far better than myself but typically when performing cluster analysis, the user needs to refine the number of clusters that exist. This is a bit of a balancing game as less clusters means more outliers and guys being put in groups where they don't really fit (perhaps your LaMelo yes) but more clusters means less classification power. What's the point of having 50 clusters with 4 guys in each? just can't use that.

So to answer you question in short, there are guys who might not fit in clusters and probably already are in the dataset (notice kickert perhaps a little out of place?) but they will still be put into the best available cluster, not creating their own new cluster

Reply #763516 | Report this post


Senator11  
Last month

Cool article, but I don't think you'd pick your starting 5 out of this. I think it would be more useful to see what type of player your team is missing or looking at compatibility, for example not wanting 3 imports that are "ball dominant studs" which I think teams do pretty well most of the time in the NBL.

Reply #763517 | Report this post


Anonymous  
Last month

Perthworld posting crap as usual.
Give the internet a rest.

Reply #763524 | Report this post


Isaac  
Last month

Is this piece saying these are the ONLY clusters or just the clusters that have appeared in the NBL over the past 5yrs?
I believe the methods are tweaked to get a set of reasonably even clusters, purely based on players from those years. He shows that in the graphs having distribution well before it's identifying players (though I think it'd be hard not to look ahead as a check). He's named the clusters, I believe, based on what features the set of players typically have. Maybe it holds OK for historical data or maybe you'd end up with different clusters needing different names.

Either way, I think the results are solid for the most part.

Reply #763530 | Report this post


andrewbprice  
Last month

Hey guys, thanks for the feedback

I'll try to answer a few of the questions -

Firstly this wasn't created with the intention of building rosters or picking starting fives, it's more a 'general interest' piece grouping together similar players to create discussion around the traditional five playing positions.

Down the track we will look to see if these groupings provide insight into which combinations of players in lineups perform well or poorly in certain situations, or against other combinations. A number of you have illuded to this outcome already. It was a significant undertaking getting it to this point, so we decided to publish the first stage and any resulting stages separately.

Secondly, it was an interesting point raised around whether a player fits in a cluster because of their nature, or because of the system they are in - with Adelaide being an example. While I don't have the answer to this (not sure if there is one, kind of a chicken and egg situation). My guess would be it's the former. Coaches usually know what type of player they need in their system and recruit accordingly. This usually serves to reinforce a player as a particular type as they are asked to perform the same role for different teams. We have used 4 seasons worth of data to try and smooth out any difference that may occur if players swap teams/systems, but only within the NBL. From the eye test, I think you'll agree there are very few that you could argue against the group they've been placed into.

Anon #763516 nailed it with their take on how clustering works, an individual will be placed within existing clusters and one player would never have a big enough effect on the entire league to shift the needle. A player like Lamelo will fit within one of these clusters, no problems. And good spotting on the Kickert one too, that's the one that's been bugging me too!

Thanks for the feedback team! Glad it's creating conversation

Reply #763532 | Report this post


The Ibek Way  
Last month

The only way I can think of to determine that is if you basically do as I suggested above, "if Conklin played a season with Adelaide would his stats for that season reflect a "quality forward"?".
So that would involve tracking players who change clubs and seeing if they're stats for that year fall into a different cluster based on the coach

Adelaide have that really obvious clustering but there is the potential there are other clubs displaying that effect albeit, less dramatically. For instance, maybe just one position rather than all bigs.

For the most part I agree with you though, they do look like they fit their clusters really well and damn it's satisfying when they do (for those that don't know cluster analysis usually takes a long time to give you hot garbage). And if there is a system effect going on I do think it looks like it's impacting the "big" clusters more than the outside players

Reply #763533 | Report this post


PyroCross  
Last month

Andrew - great work here. Love how guys like yourself can translate data/analytics in a format fans can understand.

Things I'm intrigued about:
1. The fact the NBL has plenty of local, backup guards - enough so that we could make a separate category for them.
2. Seeing the import spreads is also interesting. Would be intriguing to see how the different clusters sat on teams' rosters to see what the trends are.
3. Interested to see which players were borderline between two clusters, as there seem to be some odd pairings.

Will need more time to unpack the insights here.

Reply #763537 | Report this post


Rod  
Last month

Love your work Andrew and Robert - thought provoking for sure!

Reply #763564 | Report this post


Anonymous  
Last month

Surely for all this to make it viable, time on the court has to be taken.

Reply #763574 | Report this post


Senator11  
Last month

"Hey guys, thanks for the feedback"

Awesome work Andrew, it would be interesting to see if players who change club change their "cluster" group. Conger had very different roles with Hawks and 36ers, with Hawks really exploiting his strengths and 36ers using him more in a "role". Shawn Long could alter clusters slightly by moving to United, but presume he would stay similar with just less volume. Gliddon from Taipans to Bullets, Creek from 36ers to Phoenix, etc etc. Interesting stuff.

Reply #763584 | Report this post


UseTaHoop  
Last month

Could a smart coach use this analysis to determine changing needs in the league? Ie look at which clusters are more strongly represented in the more successful teams.

Reply #763626 | Report this post


Senator11  
Last month

"Could a smart coach use this analysis to determine changing needs in the league? Ie look at which clusters are more strongly represented in the more successful teams."

Here's your ideal team to aim for. :P

Wildcats 2018/19

Ball dominant studs: nil
The Gunner: Cotton, Terrico
Defensive Ball Handler: Martin
Backup Gaurd: Dech
3pt Specialist: Steindl
Verstile Floor Spacer: Wagstaff, Norton
3 & D: Hire?, Vague
Q Forward: Brandt, Kay
Defence First Big: nil
Offence First Big: nil
Traditional Big: nil
High Usage Big: Jervis

Reply #763636 | Report this post


D2.0  
Last month

There's several notable players that simply don't match the classification given.
There's also some bizarre inclusions in most clusters.

I suspect that the NBL is probably too small and diverse for this kind of analysis.
I think you would also need to factor in the following:
Career point/age; Redhage would have been in the last year or two of his, and basically occupying the 10~11th spot out of convenience.
Position; If looking at somebody like Loughton, surely need to factor in that he played predominately as a 4/5.
Role; The same player, used as 25min starter, or as 10min bench impact scorer, will produce different results. Similarly players behaviour will be moulded according to the team requirement. Kenny was hired to backup Martin, and was expected to play the same role. Had he been hired by another team, he might have looked different.
Volume; Trying to include guys who played a couple of minutes as DPs, is always going to be difficult.

Reply #763641 | Report this post


Anonymous  
Last month

This data analysis is excellent, it is outstanding work. Fans of the game should be thankful for this kind of insight. Few people are working with basketball data in Australia and this work should be praised at every opportunity. But it will always need additional explanation and interpretation. Everyone that works with data knows this.

Always need to add context and nuance to the data. Data helps to visualise patterns, but the patterns still need to be explained. Small samples, like 20-30 game leagues or bench players, are much more volatile and more susceptible to chance and hot streaks.

Also, the data collected is only as accurate as the people that input the data at the games. There are always weaknesses to data, and it is just one of the tools available.

Reply #763715 | Report this post


Perthworld  
Last month

Few people are working with basketball data in Australia and this work should be praised at every opportunity.

Few are in this country and that doesn't include this website as Andrew of Spatial Jam is from New Zealand.

I remember when the website was first launched years ago and thinking "it took a Kiwi to come up with a NBL analysis website!"

Thank you Andrew.

Reply #763765 | Report this post




 

Reply to this topic

Random name suggestion for anonymous posters: Rhea 20

Rules:You must read the Terms of Use. No spam, no offensive material, no sniping at other clubs, no 'who cares?'-type comments, no naming or bashing under 18 players. Learn how to embed YouTube videos or tweets

Please proof-read your post before submitting as you will not be able to edit it afterwards.



Close ads
Dunk.com.au - Custom basketball uniforms
PickStar - The best place to book sports stars
Westside Basketball Association

Advertise on Hoops to a very focused, local and sports-keen audience. Email for rates and options.

Recent Posts



Invoicing clients? Stay productive with Punch, the insightful time tracker that earns you more.

Special offer: $30/month Pay $100 for lifetime access. Sign up now!

.


An Australian basketball forum covering NBL, WNBL, ABL, Juniors plus NBA, WNBA, NZ, Europe, etc | Forum time is: 3:54 am, Wed 23 Oct 2019 | Posts: 797,031 | Last 7 days: 1,560