EasyDeepLearn

What is the Brier score and how does it compare to log loss?

medium

Answer

  • Brier = 1/n * sum (pi    yi)2(p_{i}\; - \;y_{i})2 — mean squared error between predicted probability and true label.
  • Range [0, 1] for binary.
  • Like log loss, it's a proper scoring rule that penalizes miscalibration.
  • Differences: Brier is bounded and quadratic (less punishing on confident mistakes than log loss), decomposable into calibration + refinement + uncertainty terms (Murphy decomposition), and interpretable in probability units.
  • Prefer Brier when confident misprediction shouldn't dominate the metric.
Check yourself — multiple choice
  • Brier is the same as accuracy
  • MSE between predicted probability and true label; bounded, decomposable, less harsh than log loss on outliers
  • Brier is only for regression
  • Brier score can be greater than 1

Brier = MSE(p, y). Proper scoring rule, decomposable, more forgiving than log loss.

#metrics#classification#calibration

Practise Supervised Learning

215 interview questions in this topic.

Related questions