The Brier score is the mean squared difference between a model's predicted probabilities and the actual outcomes coded as zero or one. Lower is better, and it rewards both discrimination and calibration.
Because it is a proper scoring rule, it is minimized by reporting honest probabilities. A model cannot improve its score by exaggerating confidence.
It is most meaningful in comparison, against a baseline such as the base rate for the population, or against another model on the same data. Its absolute value depends on how often the event occurs.
The Brier score is not a percentage accuracy and should not be read like one.
Working a file that turns on this term?
Working through a diligence process?
Institutional partners evaluating a position against this platform's outcome and duration models are welcome to reach out directly.