Beginning with Pega Platform 8.4, Prediction Studio informs you when an adaptive model has an unrealistically high predictive performance score, and as a result, might be overoptimistic. Increase the accuracy of the model's predictions by identifying and addressing the root cause of the issue. The possible root causes include the model continuing to collect a sufficient number of responses, or feeding the historical data to the model instance in a certain order, for example, negative responses first, and then positive responses.
Prediction Studio triggers the notification for an adaptive model that has a performance score above 95 AUC (Area Under the Curve).
Adaptive Model modelName Model performance is very high.
Recommended next steps
Identify and address the root cause of the issue by performing the following procedures:
Verify the number of registered responses
- In Prediction Studio, open the adaptive model by clicking the corresponding notification.
- On the Monitor tab, for the affected model instance, click Model report.
- Expand the Model details section, and then, in the Number of responses section, sum the number of responses that the model instance has registered for each response type:
- If each response type has less than 100 responses, wait until the model receives more responses.
- If each response type has more than 100 responses, see Reset the model.
- In the header of Dev Studio, click Configure > Decisioning > Model Management.
- On the Model Management landing page, click Adaptive.
- Select the adaptive model for which you received the notification by selecting the corresponding check box.
- Click Clear.