MVP-Shapley: AI model uses Shapley values to rank basketball players

A team of researchers has proposed MVP-Shapley, a novel framework for evaluating the Most Valuable Player (MVP) in basketball using Shapley values from cooperative game theory. The method ingests play-by-play data—such as assists, points, and defensive events—and trains a win-loss prediction model to quantify each player's marginal contribution to winning. By allocating Shapley values, the system produces an explainable, fair ranking of player impact. The researchers also introduced causal optimization to align the automated rankings with expert voter preferences, bridging the gap between statistical models and human judgment.

The framework was validated on two datasets: the NBA dataset and the Dunk City Dynasty dataset (a Chinese basketball competition). Results demonstrated that MVP-Shapley outperformed traditional metrics in both accuracy and explainability. Notably, the system has been deployed in a real-world industry setting, suggesting practical adoption beyond academic research. This work highlights how game theory and machine learning can solve subjective evaluation problems in sports analytics, offering a transparent alternative to heuristic or opaque MVP voting processes.