Specify the Training Data (Data Mining Wizard)
Use the Specify the Training Data page to identify which columns to include in the mining structure. You can select columns to include in the structure even if you do not use them in all models. For example, if you want to drill through to the columns from the mining model, you can include them in the structure but not in the model.
At least one key column is required for each table that is included in the structure. The column that you pick as the key depends on whether the table is a case table or a nested table. If the table is a nested table, the key is often also the predictable column, not the relational foreign key. To learn about nested keys, see Nested Tables (Analysis Services - Data Mining).
Different mining algorithms use keys differently. To learn about the different kinds of keys, see Content Types (Data Mining).