SQL Server 2005 Key Columns

Specifying the Column Usage

After you specify the case table and the nested tables, you determine the usage type for each column in the tables that you will include in the mining structure. If you do not specify a usage type for a column, the column will not be included in the mining structure. Data mining columns can be one of four types: key, input, predictable, or a combination of input and predictable. Key columns contain a unique identifier for each row in a table. Some mining models, such as sequence clustering and time series models, can contain multiple key columns. Input columns provide the information from which predictions are made. Predictable columns contain the information that you try to predict in the mining model.

For example, a series of tables may contain customer IDs, demographic information, and the amount of money each customer spends at a specific store. The customer ID uniquely identifies the customer and also relates the case table to the nested tables, therefore you would use it as the key column. You could use a selection of columns from the demographic information as input columns, and the column that describes the amount of money each customer spends as a predictable column. You could then build a mining model that relates demographics to how much money a customer spends in a store. You could use this model as the basis for targeted marketing.

The Data Mining Wizard provides the Suggest feature, which is enabled when you select a predictable column. Datasets frequently contain more columns than you want to use to build a mining model. The Suggest feature calculates a numeric score, from 0 to 1, that describes the relationship between each column in the dataset and the predictable column. Based on this score, the feature suggests columns to use as input for the mining model. If you use the Suggest feature, you can use the suggested columns, modify the selections to fit your needs, or ignore the suggestions.
Specifying the Content and Data Types

After you select one or more predictable columns and input columns, you can specify the content and data types for each column.