Predictive scoring

To provide a health score for predictive models, code is required in the event processing flow.

A scoring service requires a defined set of inputs and returns a result. The score returns either a numerical value, a recommendation, or both. The sources of data for the input to the scoring service are the Event, KPI (key performance indicator), and Profile tables. Code transforms the data that is needed to supply the exact set of input parameters that are required by the scoring service. The scoring service is called by a web service call from IBM® Integration Bus.

When the results are returned from the scoring service, they are written back as new events. Measurement types and profile variables can be defined for these events.

For example, a health score and recommendation can be recorded as an event_observation.measurement and event_observation.observation_text. In addition to being stored in the event tables, this score and recommendation can be aggregated for IBM Cognos® Business Intelligence Reporting by defining two profile_variables and the corresponding profile updates in the profile adapter configuration of an orchestration.

To aggregate the health score, define a profile_variable and profile_adapter configuration for the Measurement of type calculation.

To aggregate the occurrences of a specific recommendation, one needs to define a profile_variable and Profile_adapter configuration for a Text contain calculation and set the comparision_string attribute of profile_variable and profile_adapter to the name of the recommendation.

The processing of an event that contains the result of a predictive scoring service can invoke a second scoring service.

How scoring is triggered

Scoring for predictive models is triggered based on the service adapter configuration defined in the orchestration xml file. To construct any custom scoring, the orchestration xml file must be defined accordingly.