Specify the records used to train a classification solution, what fields trigger a prediction, and how often you want to retrain your solution.

Before you begin

Important: In the Yokohama release, models in the classification, clustering, and similarity frameworks use Workflow solutions. These are pre-trained, so a word corpus isn't needed for your new solutions. When your existing solutions with a word corpus are re-trained after upgrading, they become Workflow solutions, and the Word Corpus field is removed from the form.

About this task

A predictive model is only as good as the data that you use to train it. To select appropriate records for training, examine the table's database dictionary as well as the current quality of the record values that you want to use.

For information on using encrypted training data, see Data Encryption in Predictive Intelligence.

For information about the minimum and maximum number of records you can use for training, see Predictive Intelligence properties.

Note: Classes that have fewer than 30 records in your training dataset are excluded from solution training. When your solution is trained and complete, any excluded classes are listed in the Solution Statistics section of your ML Solution form.

Any classes that were excluded from your solution training are listed in the Solution Statistics section of your trained solution's ML Solution form.

You must create a separate solution definition for each predictive model you want to support. The following procedure explains how to create a new classification solution, but you can also copy an existing solution definition and its configuration into a new record by selecting Copy Solution Definition from the context menu. Edit the field values on the new record as needed.

From the Yokohama release, you can also create a Workflow Classification solution using a script if you want to include an analysis of the key features influencing the model's predictions. For more information see Model Explainability.

Procedure

  1. Ensure that you are in the application scope you want for your solution definition, then navigate to All > Predictive Intelligence > Classification > Solution Definitions.
  2. On the Classification Definitions list, select New.
  3. On the empty Classification Definition form, configure the fields according to the following guidance.
  4. Click the appropriate context menu option or button for your solution definition.
    OptionDescription
    Save or Save & Train Save your solution definition record so you can return to it later, or save and submit it for training.
    Submit or Submit & Train Create your solution definition record and submit it, or submit and train it.
  5. If you submitted the solution for training, click OK on the Training Activation window to confirm.
    • The system schedules the solution for training with the nearest training service. The system sends you a notification when the training completes, including any errors that may have occurred in the training. Other users can subscribe to the Predictive Intelligence Notifications category. When training completes, the system uploads the solution as an Attachment record.
    • A bubble chart populates the Solution Visualization tab of your solution form, showing the estimated precision and coverage for each of the classes covered by the solution. The size of the bubble indicates the percent of records (distribution) that belong to the class. When you point to a bubble, you can see its estimated coverage, estimated precision, and distribution.
    The estimated precision and coverage for each of the classes covered by the solution.

What to do next

In the Class Confidence section of the Solution Statistics tab in your solution, review the trained solution precision and coverage statistics.

In the Test Solutions tab in your solution, you can test the prediction output by entering values from the input fields, such as the Short Description.

How to test the prediction output for the records you used as input to the prediction by entering values from the input fields, such as the Short Description