SPRING 2025
Using Machine Learning to Construct Optimal Team Rosters in the Modern NBA
AUTHOR
Jaden Patel, St. Paul’s School
ABSTRACT
This study analyzes NBA roster composition over a 10-year period (2014/15 to 2023/24), aiming to identify optimal player archetypes and positional balances for maximizing team success. Detailed individual and collective performance and physical trait data from 300 distinct teams and 3557 players was used. Players were clustered into ten archetypes and three general positions (Guards, Wings, and Bigs) through k-means clustering. A supervised learning (gradient boosting) model was then employed to predict team win totals based on archetype and position profiles.
Results highlight the critical role of 3-point Specialists and Defensive Wings in modern NBA success, underscoring the value of versatile, low cost players – role players who contribute on both ends of the floor.
About the Wharton Sports Analytics Journal
ISSN 3070-4065 (Online)
The Journal is published by the Wharton Sports Analytics and Business Initiative and features original student research at the intersection of sports, business, and analytics. Explore the Journal.
