Classifying teams in the NBA with player behavioral data

Abstract

I use SecondSpectrum play-by-play data from the 2016-2017 NBA season to assemble behavioral event data for each player. Behavioral data includes propensity to dribble/pass/shoot, and also the resulting quality of shot when players decide to shoot or make another pass. I apply a k-means clustering algorithm to cluster teams based on their starting lineup behavior data; the clusters show different team makeups within the behavioral data collected. In particular, the clustering identified pass-heavy vs dribble-heavy offenses, and good shot-decision making teams.