In baseball, the WAR statistic measures a player’s value to his team,
and generates much conversation during the trade deadline and MVP
voting. However, in basketball, it is difficult to develop a WAR-like
statistic due to the ‘teamwork’ element of the game. Each possession
in basketball revolves around the common goal of players working
together to score a basket, rather than an individual battle between
the hitter and pitcher. In addition, the relative skill level of each player
on the court is very different. For example, playing alongside Lebron
James is much different than playing with Darko Milicic, and playing
against Kevin Durant is much different than playing against Sasha
Palovic.

Attempts to measure player effectiveness and value, such as
Hollinger’s PER and Rosembaum’s adjusted plus/minus, have not
accounted for the positions of players on the court. Building on those
statistics, along with other various individual statistical measures, we
have developed a model of distinct player classifications. We then
regress different combinations of these classifications with NBA
championship contenders to build a model that predicts the best
combination of player classifications that wins games. We conclude by
measuring the expected salary level of players in each classification to
measure the feasibility of building the “Winning Team”.

Unfortunately, this project was not picked for the Sloan Sports Analytics Conference and stopped before we could gain any real traction. However, this project was a great dive into the world of data parsing and regex.