How Alphr's NFL Predictions Work

1. The Model Ensemble

Three models work together for each game:

• Win model, a classifier outputting each team's moneyline probability.
• Margin model, a regression model estimating the winning margin, used for against-the-spread tips.
• Totals model, forecasting the combined points for over/under tips.

2. Team ELO Ratings

Every NFL franchise carries a continuously-updated ELO rating that rises and falls with results and margin of victory. ELO captures team strength more responsively than win/loss records and is a core input to all three models.

3. Engineered Features

Each game is described by engineered features including ELO and ELO difference, rolling form over recent games, home-field advantage, rest days and short-week effects, divisional and conference context, head-to-head history and the live market line. All features use only data available before kickoff.

4. The Edge Filter

The model's probability is compared to the bookmaker's implied probability (1 ÷ decimal odds). Only games with a positive edge become published tips, so the feed favours value over volume. See NFL best bets for the highest-edge picks.

5. Transparent Tracking

Every NFL tip is recorded before kickoff and graded afterwards, so the published strike rate and ROI reflect genuine out-of-sample performance. Read more about validation on the model validation page.