PK is typically not interestin for predictions. Primary keys have distinct occurrences in a data set (unique values for each row), and a single occurrence is not enough to learn patterns about a column. Also, in general primary keys do not hold any significance in their values -- a primary key may be a unique integer or a Guid.
However, if your primary key actually holds significant information (numerical values, I would guess?) then you may duplicate it in your mining model. For example, you could create a model which uses the PK column as a key, then right click the Mining Structure object and select Add Column, and add the PK column once more.
bogdan crivat [sql server data mining] / http://www.bogdancrivat.net/dm- 已标记为答案Jin ChenMSFT, 版主2009年8月26日 9:26
-