Preparing the Data

Two types of data preparation may be useful when you are using the Naive Bayes, Adaptive Bayes, and Support Vector Machine provided with Oracle Data Mining algorithms in modeling:

• Binning, or conversion of continuous numeric range fields to categories for algorithms that cannot accept continuous data.
• Normalization, or transformations applied to numeric ranges so that they have similar means and standard deviations.

Binning

IBM® SPSS® Modeler’s Binning node offers a number of techniques for performing binning operations. A binning operation is defined that can be applied to one or many fields. Executing the binning operation on a dataset creates the thresholds and allows an IBM SPSS Modeler Derive node to be created. The derive operation can be converted to SQL and applied prior to model building and scoring. This approach creates a dependency between the model and the Derive node that performs the binning but allows the binning specifications to be reused by multiple modeling tasks.

Normalization

Continuous (numeric range) fields that are used as inputs to Support Vector Machine models should be normalized prior to model building. In the case of regression models, normalization must also be reversed to reconstruct the score from the model output. The SVM model settings allow you to choose Z-Score, Min-Max, or None. The normalization coefficients are constructed by Oracle as a step in the model-building process, and the coefficients are uploaded to IBM SPSS Modeler and stored with the model. At apply time, the coefficients are converted into IBM SPSS Modeler derive expressions and used to prepare the data for scoring before passing the data to the model. In this case, normalization is closely associated with the modeling task.