Player evaluation with player-dependent inputs from tracking data

Develop a principled player evaluation methodology for American football that validly uses tracking-derived features that are functions of the players themselves—such as distances, speeds, orientations, and model-based space ownership—when assessing performance.

Background

The literature commonly constructs evaluation features directly from tracking data (e.g., distance, velocity, acceleration, orientation) and via model-based constructs (e.g., space ownership via Voronoi tessellation).

Despite these advances, the authors emphasize that evaluating players using such player-dependent inputs remains unresolved, indicating the need for methods that can leverage these features in a valid and interpretable way for player assessment.