Training Predicted Attributes

Training a predicted attribute creates a prediction model that can be used to predict the value of this predicted attribute for different periods. The training is based on the target and parameters specified in the predicted attribute definition.

To ensure accuracy of predictions, retrain the predicted attribute to reflect data changes over time. Typically, this could be once a month, after all the reconciliations for the month are complete. Training can be scheduled or performed manually (when the predicted attribute is created or on-demand). Scheduled training is performed during system maintenance.

When training a predicted attribute, you must specify a period. The system uses up to the last one year's historical data during the training. After training, an Expected Accuracy is generated indicating the overall accuracy of predictions made by this predicted attribute, and with the specified parameters.

To train a predicted attribute on-demand:

  1. On the Predicted tab, select the required predicted attribute.
  2. Click the Train Now icon. Or, from the Actions menu of the predicted attribute, select Train Now.

    The Train Predicted Attribute dialog appears.

  3. Under Period, select a period for the training job. It is recommended that the period should have a full set of reconciliation data.

    Locked periods are not displayed in the Period list.

  4. Click Apply.

    The Train Attribute dialog appears and a Train Predicted Attribute job is started. The Stage shows the status of the training job. After the training completes, the Training Time is displayed.

  5. Click Close to close the dialog. The training job runs in the background and can be accessed from the Jobs card.

To schedule training for a predicted attribute:

  1. On the Predicted tab, click the required predicted attribute to open the Edit Predicted Attribute dialog.
  2. In Repeat Training On Period, specify the frequency at which the predicted attribute must be trained. Select one of the following and enter a number:
    1. Start Date
    2. End Date
    3. Closed Date

Note:

The system prevents training on targets that have a high percentage of unique values.