Frame2Threat — Possession-Danger Prediction

Independent research · StatsBomb open data

What this is

Predicts possession danger from partial event sequences with an XGBoost + GRU ensemble, a graph neural network (SAGEConv) and a SHAP explanation layer. Executed notebooks and 49 committed figures.

Metrics

Ensemble

0.965 AUC (n=2,475)

Components

PossessionGRU 0.9524 · XGBoost 0.9505

Pass-level

0.882 AUC (n=7,344)

Honesty

360 context adds only +0.001 AUC

Stack

XGBoostPyTorch GNNGRUSHAPStatsBomb

Provenance

Repo history includes AI-assisted scaffolding (11 of 16 commits by copilot-swe-agent). The modelling and results are mine and defensible on substance; no solo authorship of the scaffold is claimed.

What I would not claim

Nothing beyond what is listed above.