Predicting NBA Longevity: The Role of Physical Attributes and Early Career Performance

Abstract

Basketball analytics has transformed talent evaluation, influencing scouting, drafting, and player development decisions in the NBA. This study applies logistic regression with Lasso and Ridge regularization, to predict whether a player will reach their fifth NBA season taking their position into account. Using data from 618 college players who debuted in the NBA between 2010 and 2019, physical attributes, college performance metrics, and early NBA career statistics were analysed. Findings indicate that physical traits alone offer little predictive power for guards and forwards, but for centers they play a more significant role. College and NBA performance metrics significantly enhance prediction accuracy. The model using college and first-year NBA data achieves over 0.7 AUC across all positions, making it the earliest reliable point for predicting long-term NBA success. The most effective model integrates college statistics and both first and second year NBA performance data, achieving an AUC of 0.86 for forwards with Lasso regularization. These insights assist NBA teams in refining their draft strategies, improving player evaluation, and identifying long-term prospects. Future research can expand this approach by incorporating additional predictive techniques and more complex models, external factors such as injuries and team fit, alternative statistical indicators and the use of counterfactual examples.