Skip to content

Metrics: Drift Detection

H2O Model Validation offers an array of metrics in the form of graphs to understand a Drift Detection test. Below, each metric is described in turn.

Graph: Feature PSI score

The Feature PSI score graph displays all of a model's variables from top to bottom, where H2O Model Validation orders variables from highest to lowest Population Stability Index (PSI).

Note

  • Clicking on the bar of a feature will trigger the display of a heatmap for the feature. To learn more, see Graph: Distribution of variable.

  • The Feature PSI score graph displays the top 25 features.

Example

Considering the below graph (image) of a model, we can say that the variable StateHoliday stands as the variable with the highest PSI (0.577).

Graph: Feature PSI score

Graph: Distribution of variable

The Distribution of variable histogram displays the distribution values for the selected variable in the Feature PSI score graph. The histogram combines two histograms that H2O Model Validation has combined to highlight different and similar value distributions for the selected variable in the train and new dataset.

Graph: Distribution of variables