Brier Score

INFO

Measures the accuracy of probabilistic predictions in classification tasks. Lower values indicate better calibration.

How It Works

Brier Score is the mean squared error of predicted probabilities:

$$\text{Brier Score}=\frac{1}{N}\sum _{i=1}^N(f_i-o_i{)}^2$$

• $f_i$: Predicted probability for instance $i$
• $o_i$: Actual outcome (0 or 1)
• $N$: Total number of predictions

What to Look For

• Lower score = better calibrated predictions
• Useful for probabilistic classifiers
• Sensitive to both prediction confidence and correctness

Application Models

• Logistic Regression
• Random Forest Classification
• Gradient Boosting Machines (GBM)