Skip to main content
Version: Next

Metrics: Calibration score

Overview​

H2O Model Validation offers an array of metrics to understand a calibration score test. Below, each metric is described in turn.

Settings​

Calibration curve​

The calibration curve graph displays the curve of realized event frequencies against the probabilities predicted by the model. Each point on the graph shows the values for a group of records binned by prediction ranges. The predicted probabilities order the points. Each curve represents the calibration of the target class or all target classes in the case of a multi-class model.

  • X-axis: Actual probability values
  • Y-axis: Predicted probability values
  • Dashed diagonal line: Perfect calibration line

Graph: Calibration score

Calibration scores​

The calibration scores table displays the Brier score of the target class or all target classes in the case of a multi-class model.

Table: Calibration scores

Feedback