Predictive models

You can develop a predictive scoring agent (a specialized type of Java™ agent) if you want to retrieve a scoring configuration from a predictive model. The predictive service must be deployed on an SPSS® Collaboration and Deployment Services Server.

SPSS originally stood for Statistical Package for the Social Sciences, but is now named IBM® SPSS Statistics. A predictive scoring agent is used to retrieve a score from an SPSS service to show that something is likely to happen. A score can help you to plan and carry out strategies to improve your decision making. Such a score might, for example, refer to the likelihood that a passenger is going to miss a flight that is based on the following factors:

• Capacity of the security screening infrastructure at the airport.
• Ability to handle fluctuating daily and seasonal loads.
• Passenger age.
• Passenger special service requirements.

A predictive scoring agent might, for example, pass details of a flight checkin time to the SPSS scoring configuration. The scoring configuration might then return the likelihood of the passenger to miss their flight, which is then stored in the passenger entity. A rule agent, of a lower priority than the predictive scoring agent might also be triggered by the same event. When the predictive scoring agent processing is complete, the rule agent can use the score. If the probability is high enough, the agent can notify customer services that the passenger is late and is likely to miss the flight.

To develop a predictive scoring agent, you must complete the following steps:

1. Add the entities and events that are used in the agent to the BOM project.
2. Create an SPSS predictive agent plug-in project.
3. Use the project wizard to create a connection to an SPSS Collaboration and Deployment Services Repository, and select the required scoring configuration.
4. Use the agent descriptor to define the event types, and bind the agent to an entity type.
5. Complete the scoring agent Java class to call the SPSS when an event is received.
6. Add the SPSS connection properties to a solution properties file.
7. Add endpoints for the agent to the solution connectivity definition (.cdef) file.

The export and deployment of a predictive scoring agent is similar to the export and deployment of other Java agents. The main difference between a predictive scoring agent and a Java agent is in the Java class templates that are generated by the wizards.

For more information about SPSS, see IBM SPSS Collaboration and Deployment Services.