Deep Reinforcement Learning for NBA Player Valuation: A Temporal Difference Approach with Shapley Attribution

Author: Ben Jenkins

A Temporal Difference Approach with Shapley Attribution: This paper introduces a deep reinforcement learning framework for evaluating NBA players that learns context-dependent player value directly from game outcomes. Using temporal-difference learning with a distributional win-probability model, the approach estimates how actions and player presence influence expected outcomes across game states. We combine this with a neural Shapley value attribution method to fairly decompose team success into individual contributions while capturing interaction and synergy effects. Empirical results show improved predictive accuracy, greater stability than RAPM, and systematic identification of defensive value and player synergies that traditional metrics fail to capture.