AccuracyHow often the model picked the winner. 60% means 60 correct picks out of 100 games.
AUCHow well the model ranks stronger teams above weaker ones. 0.50 is random; 1.00 is perfect. Higher is better.
Brier ScoreHow accurate the predicted percentages are. Confident wrong predictions hurt more. 0 is perfect, so lower is better.
Holdout testA season kept out of model fitting so performance can be checked on unseen games.