Validate absolute performance on an independent set of matches

Confirm the absolute performance of the Delaunay-adjacency, SEAConv-based GNN-LSTM match-phase classifier on an independent set of football matches, ideally drawn from additional competitions.

Background

The study uses a 70/30 split by complete matches into training and validation data, but does not reserve a separate test set. The validation set is used for reporting, and the authors caution that the resulting absolute performance estimates may be somewhat optimistic for unseen matches.

The authors therefore leave unresolved whether the reported performance levels will be confirmed on a genuinely independent set of matches, particularly matches from competitions beyond those represented in the data set.

References

Confirming the absolute performance on an independent set of matches, ideally from further competitions, is left to future work.

— Automated Detection of Match Phases in Football from Spatio-Temporal Tracking Data Using Graph Neural Networks  (2610.11571 - Renner et al., 8 Oct 2026) in Discussion, Section 6, paragraph beginning “Our proposed approach comes with several limitations”