IN PROGRESS · 2026
Hybrid 3D Expected Goals (xG) Model for Soccer
Most expected-goals models treat a shot as a flat point on the pitch. I'm building one that keeps the height of the ball, the body part that struck it, and where the defenders actually stood, then checks whether that extra detail predicts goals any better.
Earlyno results to report yet
PythonPyTorchSoccer Analytics3D Geometry
THE STORY
Traditional xG models flatten the world to (x, y) shot coordinates and a few categorical features. Real shots happen in three dimensions: the ball arcs, the keeper moves, the defender contests in space. I'm exploring whether adding the third dimension materially improves goal probability estimates over the standard 2D baselines.
CODE
Repo pending · model code being consolidated