Browse Teach Curate

Confidence Calibration in Classification Models

The property that a classifier's stated probabilities match observed frequencies, so that events predicted with 90 percent confidence occur about 90 percent of the time. Covers post-hoc rescaling methods such as Platt scaling and temperature scaling and the reliability diagrams used to diagnose miscalibration.

jeremy Video

Probability Calibration using Logistic Regression in Python

Probability calibration is a statistical mechanism in machine learning that corrects for systematic miscalibration where predicted probabilities from classifiers deviate from empirical frequencies. T…