I am not immediately recalling seeing any substantial work to cluster and characterize the most common NBA lineup types, besides David Sparks' work based of player types found from traditional boxscore stats viewtopic.php?f=2&t=116&p=193&hilit=lin ... ology#p193 and a few brief previous mentions of the possible topic by me? What have I missed or forgotten?
What would machine learning find using team level and player level data?
Just brain free-lancing, I can name a few (straight-ahead or somewhat silly but still serious):
2 PG,
3 Guard
4out, 1 in
Classic
Featuring stretch 4
Stretch 5
Smallball thru PF
Supersmallallthewayball
Big or at least post capable at the 3 ball
2 PFs
Inadequate from 3 pt land ball
No post ball
High post
Low post
Perimeter player in the post
Wing initiated offense instead of PG
Rebound the hell out of it ball
Move the rock, passing collective
Dribble drive obsessed
Get some stops dammit, dam the flow emergency crews
Make a shot or die lineup
One man shows
Positionless jam bands
I didn't really have a coherent plan this string but I can pretend to and I can rack up 600-900 lineups used this way
Garbage time from here
Tank job
Lets run
We love the 3
We love the 3 and anything at the rim or ft line
We're trying 3 and D
We're trying 3 and OR
Bad boys or just grit and grind
The assemble of the one way players
Two way strong as we can get
D is our best O
O is our best D
The fresh leg mass substitute 5
Vet dominant
Speed PG
4 guys and a thug
4 guys and a has been we feel we have to play
Super dominant PG
Lots of us can't or won't pass
No go to guy right now
Extremely high combined usage lineup
Inadequate usage lineup
We are mostly long y'all
The big 2 or 3
The young guns (for now or the future)
Pick n roll
Pick n pop
More sure hands
Tweaked for most recent mistake or need
Does any team ask or require a coach to give his post game rationale for each lineup move? Why not? Do coaches systemically self analyze in private? Is that sufficient? If you reported in real time and gave feedback in real time or at timeouts or quarter breaks via a bench, behind the bench or wired remote analyst could the team make better decisions? Is it a team game and an external analytics informed game for everybody or everybody but many head coaches? Can you evaluate coaching fully without measuring results against known intentions? If the coach made a change for better shooting, rebounding, passing, ball control, getting more fould or committing less, does that jive with the actual lineup performance to date or against this opponent or this specific type of opposing lineup? Yeah you could overdo this, but you can certainly under do it too.