Prediction Studio notification types
Prediction Studio provides various notifications that describe the status of adaptive and predictive models. Use the details that are gathered in these actionable insights to monitor your models and update their configuration as necessary, to improve the predictions that support your customer-oriented strategies.
For more information about predictive performance metrics, see Metrics for measuring predictive performance.
| Notification | Pega Platform version | Description |
|---|---|---|
| No monitoring information for model in the last 2 days. | 8.3 and later | The model instance has not had any monitoring data for the last two days. |
| No updates to model in the last 2 days | 8.3 and later | The model instance has not been updated for the last two days. The number of responses gathered between two days ago and the time when the monitoring data was last saved has not changed. This notification occurs only when historical monitoring data is saved. |
| Model performance is at its minimum of 50 AUC. | 8.3 and later | The model instance area under the curve (AUC) performance equals 50. |
| Zero responses received for adaptive model. | 8.3 and later | The model instance has not received any responses yet. |
| 20% less responses received within the current week compared to last week. | 8.3 and later | This week’s monitoring data for the model instance contains at least 20% fewer responses compared to the previous week. |
| Performance has dropped by 20% compared to the previous day. | 8.3 and later | The model performance has decreased by more than 20% in the current monitoring data, compared to the previous day. |
| Success rate has dropped by more than 50% compared to the previous day. | 8.3 and later | The model success rate has decreased by more than 50% in the current monitoring data, compared to the previous day. |
| Model performance is low for adaptive model. | 8.3 and later | The model instance performance is between 50 and 52 AUC. When the model performance is 50, the status is covered by the Model performance is at its minimum of 50 AUC notification. |
| Notification | Pega Platform version | Description |
|---|---|---|
| Model performance is low for the current week. | 8.3 and later | The weekly performance of the binary outcome model is lower than 52 AUC, based on sufficient evidence of the least frequent outcome with more than 100 responses. |
| ROC curve is under the diagonal. | 8.3 and later | The ROC curve for the binary outcome model is under the diagonal, with the AUC value below 50. This status means that the labels for the outcomes or the output score of the model should be inversed. |
| Performance of the model over the last month is less than 20% of the expected performance. | 8.3 and later | The model monthly performance for each version is below 20% of the expected performance. The measurement is based on the sufficient evidence of more than 100 positive and 100 negative responses for the binary outcome models, and more than 500 responses in the case of categorical and continuous outcome models. |
| 20% less responses received within the current week compared to last week. | 8.3 and later | This week’s monitoring data for the model instance contains at least 20% fewer responses compared to the previous week. |
| Performance has dropped by 20% compared to the previous day. | 8.3 and later | The model performance has decreased by more than 20% in the current monitoring data, compared to the previous day. For continuous outcome models, this notification means that the root-mean-square error (RMSE) value has increased by more than 20%, which reflects the lower performance. |
| Success rate has dropped by more than 50% compared to the previous day. | 8.3 and later | The success rate for the binary outcome model has decreased by more than 50% in the current monitoring data, compared to the previous day. |
For information about how to access these notifications, see Viewing model notifications.
Previous topic Introduction to adaptive models monitoring Next topic Using historical data to predict customer behavior