How does model calibration relate to bias and variance in probabilistic predictions?

Calibration ensures predicted probabilities reflect true outcomes, reducing systematic bias in probabilistic models.

Model calibration refers to how well predicted probabilities align with actual observed frequencies. A miscalibrated model may have good accuracy but biased probability estimates. This is a form of systematic error (bias).

Variance affects calibration stability across datasets. High-variance models often produce unstable probability estimates. Techniques like Platt scaling and isotonic regression are used to recalibrate outputs.…