I’ve used hierarchical clustering for three separate basketball projects: once, to group Hall of Famers by Win Shares, then, to group historic rosters by VORP distributions, and, most recently, to group offensive roles using play-type data.
1. Grouping Hall of Famers
Here’s how the clustering worked out for the collection of Hall of Fame players.
The clusters reflect combinations of career longevity and average Win Shares per season.
From there, I found HOF comps for the 2016 inductees.
2. Grouping Historic Rosters
Here I grouped NBA rosters by the way talent was distributed between players.
Below is one example of the VORP distribution from one team in each cluster of rosters.
Many roster constructions have yielded championship success, but
For a given level of overall team quality, an unbalanced roster has been more likely to produce a championship than a balanced one.
3. Grouping offensive roles
I used play-type frequencies and hierarchical clustering to define three positions,
and 18 roles.