Adaptive Models Reporting landing page

Use the reports on this landing page to monitor adaptive models in the ADM system. Reporting on adaptive models is based on the last data snapshot, which means time and date may be not the same as when you request the report.

Data snapshots are controlled by the ADM Data Mart settings in the Services configuration. In these facilities, you can use Update Reporting Data to take a new snapshot of the models and predictors in the ADM system.

Access this landing page by selecting the > Decisioning > Monitoring >Adaptive Models Reporting menu item.

There are two tabs:

• Models overview: Inspect active and inactive predictors.
• Predictors overview: Inspect the use of predictors for all adaptive models in the ADM system.

Available in applications that have access to the Decision Management rule sets or in applications built on PegaDM.

Models overview

Models overview displays all models in the ADM system in a bubble chart and allows you to compare models in terms of success rate and performance. You can apply filtering to focus on models for a particular direction, channel or a combination of the two.

The data used to visualize the models is displayed below the chart (model in the action context, response count, performance and acceptance rate) and you can use the reporting actions obtain the behavior profile and the performance overview.

Behavior Report

The behavior report contains the behavior profile. The behavior profile allows you to observe the treatment of predictors and the difference between positive and negative cases. The grouping is automatically determined by ADM. The behavior profile is centered around predictor grouping and predictive performance.

To see the behavior report:

1. In the data grid, locate the model for which you want to see the report.
2. Click Select Report.
3. Select Behavior report.

The report provides information about the modeled proposition:

• Predictors: Names of the properties used as predictors.
• Performance (in%): Total predictive performance.
• Bounds/Symbols: Bounds for numeric predictors or number of symbols for symbolic predictors.
• #Bins: The number of bins (influenced by the adaptive model rule's grouping settings).
• Positives: Number of positive settings.
• Negatives: Number of negative settings.
• Type: Numeric or symbolic.

List of active predictors: predictors actively used for modeling, also known as filtered behavior.

List of inactive predictors: predictors that are not used for modeling. Predictors are inactive if their performance is below the performance threshold defined in the adaptive model rule.

The Classifier section displays the model's predictive performance (unfiltered behavior).

For every predictor, use the expand control to view the breakdown of statistics per interval/category.

Performance overview

Performance overview lets you monitor predictive performance over time. To run the report:

1. In the data grid, locate the model for predictive performance analysis in which you are interested.
2. Click Select Report.
3. Select Performance overview to display the Performance over time graph.

Predictors overview

Predictors overview displays predictor information in a bar chart. This overview lets you analyze predictor use across models in the ADM system.

In the Show graph for drop-down select the criteria for reporting on predictors:

• Average Performance: overview of predictors based on the average predictive performance.
• Maximum Performance: overview of predictors based on the highest predictive performance.
• Minimum Performance: overview of predictors based on the lowest predictive performance.
• Number of Models Where Predictor is Active: overview of active predictors based on number of models.
• Number of Models Where Predictor is Inactive: overview of inactive predictors based on number of models.

The data used to visualize the predictors is displayed below the chart:

• Predictor name
• Number of models where the predictor is active
• Number of models where the predictor is inactive
• Minimum performance
• Maximum performance
• Average performance

Related information

Open topic with navigation