feature-engineering

Build leak-free hockey model features by enforcing strict temporal ordering and shift-before-roll validation for rolling windows, ensuring predictions use only past game data available before the forecast.

Calculate rolling hockey form metrics by shifting data before rolling to avoid lookahead bias, ensuring recent scoring and shot attempt averages exclude the current game being predicted.

Create home-minus-away hockey game prediction features by subtracting away team stats from home team stats, incorporating rest-day splits and opponent adjustments for predictive modeling.

Yes, use Elo ratings and goalie stats as hockey model inputs by adding pre-game Elo ratings, starter save percentage, GSAA, and confirmed starter flags to improve win probability and totals modeling.

Strength of schedule adjustment works with hockey game logs by applying opponent adjustments to raw game data, creating matchup-ready features that account for opponent quality in predictive models.

Hockey models overfit with rolling windows when shift-before-roll validation is skipped, causing data leakage; enforce strict temporal ordering so rolling form metrics exclude future game outcomes.