Glossary
The terms used across RaceModel, each defined once and with how it applies here.
- Brier score
- A measure of how good a set of probability forecasts was; lower is better, zero is perfect.
- Calibrated probability
- A probability that matches reality in the long run: events given a 20% chance happen about 20% of the time.
- DNF risk
- The estimated probability that a driver or rider does not finish the race.
- Expected value (EV)
- What a bet is worth on average, given the model's probability and the bookmaker's odds.
- Finishing order prediction
- A predicted classification for the whole field, from first to last, rather than a tip for the winner.
- Frozen prediction
- A prediction published and locked before the race starts, so it cannot be edited once the result is known.
- Kelly criterion
- A formula for how much of a bankroll to stake on a positive-EV bet.
- Leave-one-race-out validation
- A way of testing a model in which each race is predicted by a model that never saw that race in training.
- Mean absolute error (MAE)
- The average number of finishing positions by which the prediction missed, across the whole field.
- Model version
- An identifier for the exact trained model that produced a prediction.
- Podium hit rate
- The share of predicted podium places that turned out to be correct.
- Podium probability
- The model's estimated chance that a driver or rider finishes in the top three.