luxury-lakehouse/psxg-predictions

PSxG Predictions — Post-Shot Expected Goals per Shot

Per-shot PSxG scores produced by the PSxG model (logistic regression on goalmouth coordinates). Used as the primary shot-stopping input to fct_goalkeeper_stats in the goalkeeper evaluation framework.

Part of the (Right! Luxury!) Lakehouse soccer analytics platform.

Quick Start

from datasets import load_dataset

ds = load_dataset("luxury-lakehouse/psxg-predictions")
df = ds["train"].to_pandas()
print(f"{len(df)} on-target shot predictions")

# PSxG faced per goalkeeper (join with shots dataset for goals_conceded)
gp = df.groupby("player_id").agg(
    psxg_faced=("psxg", "sum"),
    shots_faced=("event_id", "count"),
)

print(gp.sort_values("psxg_faced", ascending=False).head(10))

Explore interactively: Soccer Analytics App

What Is This Dataset?

This dataset contains one row per on-target shot, with the PSxG score assigned by the PSxG model. PSxG is the estimated probability that a given on-target shot becomes a goal, conditioned on goalmouth position.

Goals prevented = sum(PSxG over shots faced) − actual goals conceded.

• Positive value indicates the goalkeeper saved more goals than expected given shot difficulty.
• Negative value indicates the goalkeeper conceded more goals than expected.

Data Fields

Column	Type	Description
event_id	string	Unique StatsBomb event identifier (join key to source shot events)
match_id	Int64	Match identifier
player_id	Int64	Goalkeeper player identifier (the keeper who faced the shot)
psxg	float64	Post-Shot Expected Goals: probability the shot becomes a goal [0, 1]

Interpreting PSxG

PSxG Range	Meaning
0.80–1.00	Near-certain goal (top corner, unstoppable)
0.40–0.80	Difficult save required
0.10–0.40	Moderate difficulty
0.00–0.10	Routine save (central, low, slow)

Data Sources

Predictions are generated from the PSxG model applied to on-target shots in StatsBomb Open Data.

Source	On-Target Shots	License
StatsBomb Open Data	~15K	CC-BY 4.0

Use Cases

• Goalkeeper benchmarking: Aggregate goals_prevented = sum(psxg) - actual_goals per goalkeeper to rank shot-stopping performance
• Season analysis: Track a goalkeeper's PSxG performance over a season to distinguish form from underlying difficulty
• Squad analysis: Compare squad goalkeepers on shot-stopping contribution beyond raw save percentage
• Research: Evaluate custom PSxG models against this logistic regression baseline

Limitations

• StatsBomb only: Predictions are generated only for shots with StatsBomb goalmouth coordinates (end_location_z). No Wyscout coverage.
• Two-feature model: PSxG is conditioned only on end_location_y and end_location_z. Shot speed, trajectory, and defensive pressure are not modeled.
• No keeper position conditioning: The model does not observe the goalkeeper's starting position or reaction. Saves from unconventional positions may appear easier than they were.
• Goalkeeper attribution: player_id identifies the goalkeeper who faced the shot, derived from the StatsBomb event data keeper field.

Citation

If you use this dataset, please cite:

@article{butcher2025xgot,
  title={An Expected Goals On Target (xGOT) Model},
  author={Butcher, J. and others},
  journal={Big Data and Cognitive Computing},
  volume={9},
  number={3},
  pages={64},
  year={2025},
  publisher={MDPI},
  url={https://www.mdpi.com/2504-2289/9/3/64}
}

@software{nielsen2026psxg,
  title={PSxG Model: Post-Shot Expected Goals for Goalkeeper Evaluation},
  author={Nielsen, Karsten Skyt},
  year={2026},
  url={https://github.com/karsten-s-nielsen/luxury-lakehouse}
}

Companion Resources

Resource	Description
PSxG Model	Logistic regression PSxG model that produced these predictions
On-Target Shot Data	Input dataset: ~15K StatsBomb on-target shots with goalmouth coordinates
xG Shot Data	Full shot dataset with pre-shot xG features (StatsBomb + Wyscout)

More Information

Explore interactively: Soccer Analytics App

• Model repo: luxury-lakehouse/psxg-model
• License: CC-BY-NC 4.0
• Platform: Luxury Lakehouse Soccer Analytics