Building a predictive model

When building a predictive model, the only required specifications are a data source and target field. You can also choose options for preparing and partitioning data, and for selecting the fields and records to use. The model is generated by applying a range of methods as appropriate to the target type, and automatically identifying the specific technique or combination that performs best on the selected data.

When scored, the model returns one or more fields containing predictions, as well as confidence values associated with those predictions. For example, if the target field is a flag field named Response, the model returns a field named XF-Response containing predicted values for that field. The predictions or scores returned by the model can then be used as input to rules, prioritization, and other points in the application as appropriate.

1. Specify a data source. This can be any analytical data source that records the outcome you want to predict.
2. Specify a target field that records the result you want to predict. For example, the target field might indicate which customers have churned, responded to past offers, raised fraudulent claims, and so on.

   All other fields will automatically be included as inputs that may be useful in predicting this value.

3. Specify optional settings as desired. See the topic Optional model settings for more information.
4. If desired, click the Data Overview icon to see an overview of the data that will be used to build the current model. See the topic Data overview for more information.
5. Click Build Model.

   You can close the browser or work on other tasks while the model builds. The model can be accessed from the Gallery once building is complete. See the topic Gallery for more information. If the model is taking longer to build than expected, click the Stop button to revert to the previously-saved version.

   When model building is complete, results are displayed. You can choose to view the Combined model results (performance charts for the combined model that was currently built) or the Individual model results (high level details of the individual models that make up the combined model).

   When viewing individual model results, the name of each model is a link that opens a new dialog containing information about the model such as the target field and input fields. If the selected model is a tree model, then a tree viewer will also be displayed.

6. Optionally, use the Evaluate and Test features to see how the model performs on your sample data. See the topic Evaluating models for more information.
7. Save the model before closing the model builder or returning to the application.
8. Click Use Model, and select the model field you want to use. For example if you want to use the value predicted by the model as input to a rule, select the field that contains the predictions.