📈 Pre-Game Model Performance

Evaluation of the NHL pre-game win probability model, including overall accuracy, confidence bucket performance, probability calibration, and team-level trends.

Use these filters to compare model performance by season and game type. Playoff results will appear here automatically once completed games are captured.

🤖 Model Performance

59.1%

Accuracy

388

Correct Picks

657

Games Evaluated

XGBoost • Updated nightly • Through yesterday

🎯 Prediction Accuracy by Confidence Range

This table shows how often the model’s predicted winner was correct within each predicted home-team win probability range. This reflects bucketed prediction accuracy rather than true probability calibration.

Predicted Home Win Range Range	Games GP	Correct Predictions Correct	Accuracy Acc
20–30%	2 Low sample	2	100.0%
30–40%	64	44	68.8%
40–50%	203	112	55.2%
50–60%	240	133	55.4%
60–70%	132	86	65.2%
70–80%	16 Low sample	11	68.8%

Bins are grouped in 10% increments. Low sample = fewer than 30 games. Only completed games through yesterday are included.

📏 Probability Calibration

Calibration compares the model’s average predicted home-team win probability to the actual home-team win rate in each range. A well-calibrated model should have actual results that are close to its predicted probabilities.

Predicted Home Win Range Range	Games GP	Avg Predicted % Pred	Actual Home Win % Actual	Difference Diff
20–30%	2 Low sample	26.9%	0.0%	-26.9%
30–40%	64	36.6%	31.3%	-5.3%
40–50%	203	45.5%	44.8%	-0.7%
50–60%	240	54.6%	55.4%	+0.8%
60–70%	132	64.2%	65.2%	+0.9%
70–80%	16 Low sample	72.0%	68.8%	-3.2%

Difference = Actual Home Win % minus Avg Predicted %. Positive values mean the home team won more often than predicted. Negative values mean the model was overconfident in the home team.

📈 Overall Model Accuracy Over Time

This chart shows rolling model accuracy across completed games using a 25-game window.

The line shows how the model’s recent prediction accuracy has trended over time.

📊 Team-Level Model View

Select a team to see how the model has viewed that team over time, including average predicted win probability, actual win rate, and a rolling 10-game trend.

Select Team

ANA

Rolling 10-game comparison of model probability versus actual win rate.

Games Evaluated

55.7%

Avg Model Probability

59.6%

Actual Win %

21-15

Record as Favorite

7-4

Record as Underdog

8-5

Record as >60% Pick

Updating team view...

The blue line shows the rolling average predicted win probability for the selected team. The second line shows the team’s actual rolling win rate over the same 10-game span.

ℹ️ About the Pre-Game Prediction Model

This model estimates the probability that the home team will win an NHL game before it begins, using historical game data and team-level performance metrics.

The model predicts the probability that the home team will win before puck drop using historical game data and pre-game team features.
Predictions are generated before each game and are not updated live as games are played.
Model performance metrics, bucketed accuracy, and calibration are recalculated automatically as completed games are added.
Because early-season predictions are made with less current-season data, performance and reliability can shift as the season progresses.
The playoff filter will remain empty until completed playoff games are available in the nightly update pipeline.

This page is designed to show both how often the model is correct and how well its predicted probabilities align with actual outcomes.

Data sources include official NHL game results and statistics. Models are updated nightly.

💰 Daily Edge Board 📘 Best Odds 📅 Schedule 📊 Game Logs 🏆 Standings