Building Models with Oracle Data Mining

Oracle model-building nodes work just like other modeling nodes in IBM® SPSS® Modeler, with a few exceptions. You can access these nodes from the Database Modeling palette across the bottom of the IBM SPSS Modeler window.

Data Considerations

Oracle requires that categorical data be stored in a string format (either CHAR or VARCHAR2). As a result, IBM SPSS Modeler will not allow numeric storage fields with a measurement level of Flag or Nominal (categorical) to be specified as input to ODM models. If necessary, numbers can be converted to strings in IBM SPSS Modeler by using the Reclassify node.

Target field. Only one field may be selected as the output (target) field in ODM classification models.

Model name. From Oracle 11gR1 onwards, the name unique is a keyword and cannot be used as a custom model name.

Unique field. Specifies the field used to uniquely identify each case. For example, this might be an ID field, such as CustomerID. IBM SPSS Modeler imposes a restriction that this key field must be numeric.

Note: This field is optional for all Oracle nodes except Oracle Adaptive Bayes, Oracle O-Cluster and Oracle Apriori.

General Comments

  • PMML Export/Import is not provided from IBM SPSS Modeler for models created by Oracle Data Mining.
  • Model scoring always happens within ODM. The dataset may need to be uploaded to a temporary table if the data originate, or need to be prepared, within IBM SPSS Modeler.
  • In IBM SPSS Modeler, generally only a single prediction and associated probability or confidence is delivered.
  • IBM SPSS Modeler restricts the number of fields that can be used in model building and scoring to 1,000.
  • IBM SPSS Modeler can score ODM models from within streams published for execution by using IBM SPSS Modeler Solution Publisher.