Skip to main content


         This documentation site is for previous versions. Visit our new documentation site for current releases.      
 

Viewing a model report

Updated on May 4, 2022

To analyze an adaptive model, you can view a detailed model report that lists active predictors, inactive predictors, the score distribution, and a trend graph of model performance. You can also zoom into a predictor distribution.

  1. In the navigation pane of Prediction Studio, click Models.
  2. Click an adaptive model that you want to edit.
  3. On the adaptive model form, click the Monitor tab.
  4. Optional: To refresh the model details with the latest reporting data from the Adaptive Decision Manager (ADM) server, click Refresh reporting data.
    The data in the bubble chart comes from data snapshots that are taken on the Adaptive Decision Manager (ADM) server.
  5. In the grid that contains model data, find the model that you want to report on.
  6. In the View column, click Model report.
  7. Click one of the following tabs:
    • To analyze predictors for the selected model, click Predictors.

      Correlated predictors are automatically grouped under the best-performing predictor whose status becomes Active. The remaining predictors in each group are Inactive. You can expand each group to view all predictors that belong to that group.

      Predictors that have a univariate performance under the performance threshold setting also become Inactive.

    • To display generated score intervals and their propensity, click Score distribution.
    • To display the performance of the selected adaptive model over time, click Trend.

      Click this tab to identify sudden changes in the performance of your model when new propositions or predictors are added.

  8. Optional: To view detailed distribution metrics, click a predictor.
    You can view detailed metrics for positive and negative responses, propensity, z-ratio, and lift. For more information, see Predictor report details.
  9. Optional: To export the model report as a CSV or PDF file, click Export and select the applicable format.
    Note: You must choose the model report of the top model, otherwise you get a message that the model data is empty.

Model report details

The model report provides information about the predictor data for the selected model.

Name
Provides the names of the properties used as predictors. Click the name of the predictor to display additional details.
Status
Shows whether a predictor is used or not used by the adaptive model. Predictors can also be inactive if their performance score falls below the threshold or they are highly correlated to another predictor that has a higher performance score.
Type
Indicates the predictor type (numeric or symbolic).
Performance (AUC)
Indicates the total predictive performance that is expressed in the Area Under the Curve (AUC) unit of measurement.
Positives
Shows the number of positive responses.
Negatives
Shows the number of negative responses.
Range/# Symbols
Shows ranges for numeric predictors or the number of symbols for symbolic predictors.
# Bins
Shows the number of bins. The number of bins is affected by the group settings in the adaptive model.

Predictor report details

You can access a detailed report for a predictor from a model report. By viewing detailed statistical data for specific a predictor, you can assess that predictor's performance.

Histogram chart

By zooming into a predictor from a model report, you can inspect the correlation between the percentage of responses for each predictor bin and the associated propensity value.

Predictor performance data

Range/Symbols
The ranges for numeric predictors or the number of symbols for symbolic predictors.
Responses (%)
The percentage of all responses for a predictor bin.
# Positives
The number of positive responses for a predictor bin.
Positives (%)
The percentage of positive responses for a predictor bin.
# Negatives
The number of negative responses for a predictor bin.
Negatives (%)
The percentage of positive responses for a predictor bin.
Propensity (%)
The predicted likelihood of positive behavior (for example, the likelihood of a customer accepting an offer).
Z-Ratio
The difference between the propensity in a predictor bin and the overall propensity that is expressed in the number of standard deviations.
Lift
The propensity in a predictor bin divided by the overall propensity.

Downloading the gradient booster model

For models using the gradient boosting technique, you can export the model as a JSON file for further inspection.

  1. In the navigation pane of Prediction Studio, click Models.
  2. Click an adaptive model that you want to export.
  3. On the adaptive model form, click the Monitor tab.
  4. Open the adaptive model rule.
  5. Select the row for which the proposition is empty (the top model row).
  6. In the View column, click Model report.
    Viewing adaptive gradient boosting model report
    Selecting model report from the top model row.
  7. From the Actions menu, click Download model.
Result:
Predictor importance in the model report
Importance column in the predictor importance tab of the model report.
You download a JSON file that contains a full description of the gradient boosting adaptive model for close inspection.

Related articles

Adaptive models monitoringViewing a model reportViewing a model report

Tags

Pega Platform 8.7 Decision Management

Have a question? Get answers now.

Visit the Support Center to ask questions, engage in discussions, share ideas, and help others.

Visit the Support Center

Did you find this content helpful?

Yes
No

Want to help us improve this content?
Suggest an edit

We'd prefer it if you saw us at our best.

Pega.com is not optimized for Internet Explorer. For the optimal experience, please use:

Close Deprecation Notice
Contact us