xg-model-building

An expected goals model converts NHL play-by-play shot events into goal probability estimates using event coordinates, shot type, rebound and rush features, and strength state to quantify shot quality accurately.

Hockey shot quality modeling requires event coordinates, shot type, rebound flags, rush indicators, and angle or distance metrics derived from play-by-play data to accurately estimate goal probability.

Separate strength-state models are needed because shot quality and game dynamics shift significantly between even strength and power play situations, requiring distinct calibration checks for accurate team xGF and xGA aggregation.

Yes, xgboost is commonly used to train expected goals models on hockey shot data, requiring subsequent calibration checks to ensure the predicted goal probabilities align accurately with actual outcomes.

You aggregate shot-level xG into team metrics by summing the individual goal probability estimates for all shots taken and allowed, creating team-level expected goals for and expected goals against totals.

Relying on shot volume ignores shot quality context like location, shot type, and rebound or rush situations, leading to skewed evaluations, whereas expected goals weights these factors to provide a true performance measure.