What are the Brier score and log loss?

The Brier score is the mean squared error of probability forecasts: 0 is perfect, and giving 1/3 to each outcome of the 1X2 scores 0.667. Log loss measures the same thing but punishes misplaced confidence harder.

In the 1X2, a match's Brier score is the sum, over the three outcomes, of (forecast probability − result)², where the result is 1 for what happened and 0 for the rest. The average over many matches gives the model's score.

The number only means something in comparison: against another model, against the market, or against a baseline that uses only the frequency of home wins, draws and away wins.

See also

• Model calibration
• How xG PRO calculates every match