Prediction Methodology

How the Model Works

This project uses a Bayesian statistical model to simulate and predict the outcomes of the NHL playoffs. The methodology is designed to provide robust, probabilistic estimates for both series and individual games, leveraging the full season’s game data and advanced simulation techniques.

Data Sources and Inputs

The model incorporates several key data sources:

  • Game Level Data from the official NHL API
  • All games up to current date
  • Historical playoff performance data

Prediction Components

The predictions are broken down into several key components:

Series Win Probability

Calculated using Monte Carlo simulations of the entire series, factoring in home-ice advantage shifts.

Game-by-Game Predictions

Individual game win probabilities that account for home-ice advantage, rest days, and prior

Series Length Distribution

Probability distribution of series ending in 4, 5, 6, or 7 games, derived from the game-by-game prediction model and historical patterns.

Exact Series Outcomes

Detailed forecasts of specific series scorelines, like the probability of a team winning 4-2 or 4-3.

Model Validation and Updates

The model is continuously validated against actual outcomes and updated daily throughout the playoffs to incorporate:

  • Results from completed games
  • Changing team performance metrics

Interpretation of Results

All probabilities should be interpreted as the likelihood of an outcome based on the model's current assessment. Even high-probability outcomes are not certainties, and low-probability events do occur in sports. The model provides a data-driven framework for understanding the relative likelihood of various scenarios.

Technical Notes

The model employs a combined approach using:

  • Bayesian probability frameworks
  • The model is compared to an Elo baseline for improvement.