Creating a predictive model

Use predictive models to compute a prediction that you can use to make a decision. You create predictive models in decision services, to then use them in decision models.

For example, let's say you have a decision model that decides whether a loan that is requested by a customer should be approved. You have a machine learning model that can predict how likely it is that the customer reimburses the loan, based on a database of past loans. To use this prediction in your decision model, you must encapsulate it in a predictive model first. Then, you can inject this predictive model in your decision model.

Before you create a predictive model, you must have knowledge about the machine learning model:

List the data that the model requires to make a prediction. For example, the prediction might be based on the age and monthly salary of the customer, and on the amount and duration of the loan.
Find out the expected form of the prediction. It might be a number in the range 1 - 100, where 100 means that is it certain that the customer will reimburse the loan, and 1 means he will not.
Verify the range of the values that the model was trained with. For example, if the age of the customer is used to make the prediction, verify the range of ages that the model was trained with. By ensuring that the input data that you provide to the predictive model is within this range, you increase the reliability of its predictions.

To create a predictive model, you generate a template based on a machine learning model. It contains several nodes that are automatically generated for you:

An input data node that represents one of the input data types that the machine learning model requires to make a prediction.
A decision node for the mapping of the input data. It contains rules that you write to map the input data types of the machine learning model to data types of your data model.
A decision node that contains the rule to invoke the machine learning model.
A decision node for the mapping of the output data. It contains rules that you write to map the output data type of the machine learning model to a data type of your data model.

A machine learning sample is available with Automation Decision Services to help you get started with predictive models. For more information, see Samples and tutorials in GitHub.