Evaluate the performance of your model by field.

Before you begin

Role required: admin

About this task

After you’ve created and tested your model, assess its performance to evaluate how well it's predicting what you want it to predict. For multi-output models that predict multiple fields, you may want to assess the performance of each field individually.

If you have not already created a model, see either Create a model to predict record fields or create an incident field model to make one.

Your model must be in Monitoring mode. You can set Monitoring mode on the Assess Your Model screen. For more details, see the Set your preferences step in Create a case field prediction model or Create an incident field prediction model.

Procedure

  1. Navigate to the Predictor Results (ml_predictor_results_task) by entering ml_predictor_results_task.list in the application navigator.
  2. Filter the Predicted Output Value Name field for the name of the field that you want to assess, such as "product" or "category."
  3. Group the list by Predicted Correctly by selecting the list controls icon in the top-left corner of the screen.
    List control open to group by Predicted Correctly
  4. Divide the number of records where Predicted Correctly is true by the total number of records.
    The result represents your accuracy for the field.
  5. Follow steps 2 through 4 for each field in which you would like to calculate the accuracy.
  6. (Optional) View the performance of your model on the Monitoring screen.
    1. Navigate to Task Intelligence for Customer Service > Monitoring in the application navigator.

      The Monitoring page enables you to select the model, field, and date range to display in the charts. Accuracy is measured based on whether the top autofilled value or the top three recommendations are correct.

      Monitoring page showing options to view model performance charts at the field level.

    2. Select a model.
    3. Select an output column for the field.
    4. Select a date range.

What to do next

If the accuracy of each field is acceptable, transition your model from monitoring mode to real-time predictions and deploy. If the accuracy of a field isn’t acceptable, you can remove that output field from your model, retrain, and deploy. See Edit a Task Intelligence model for more information on editing your model.