Data Mining - Professional SQL Server Analysis Services 2005 with MDX - page 732

Working with Mining Models
Analysis Services 2005 provides two types of mining models: the relational
mining model and the OLAP mining model. Relational mining models are cre-
ated directly from the relational data source and the OLAP Mining models are
created from an existing cube or part of a cube. Use of the nine types of data
mining algorithms are made within the context of relational or OLAP mining
models. In this chapter you will learn both these models by creating mining
models using a few algorithms and analyzing the results.
Relational Mining Model
The Adventure Works DW sample relational database has specific patterns to
demonstrate various algorithms available in Analysis Services 2005. In this
section you learn how to create and analyze a decision tree model and a clus-
tering model. Obviously, you need to create a new mining model to explore
and analyze the information. When you build a mining model, Analysis Ser-
vices 2005 retrieves data from the data source and stores it in a proprietary
format. When you do want to build several mining models from the data set,
there will be redundant data stored on Analysis Services 2005. In order to
share the data across several mining models, Analysis Services 2005 stores
the information about the data that can be shared across several mining mod-
els under an object called Mining Structure. Internally the information read
from relational data sources is stored as a cube in order to efficiently retrieve
the data during mining model creation. The Mining structure stores data type of
attributes metadata in the mining model, the corresponding column in the data
source, and allows you to modify the certain data mining properties that are
common across all of your mining models.
Have you received coupons in the mail? If you have a postal address, you
have. Retail companies used to send coupons to all customers and even some
who weren't customers. That was expensive and of less than optimal effi-
ciency. In order to minimize cost and maximize profit, companies now use data
mining to select targets for coupon or other special postal distributions. Based
on certain attributes, retail companies can classify customers into several
groups (Gold, Silver, or Bronze membership). By doing this they clearly identify
unique characteristics of the group. From there, targeted mailing to those
groups can be made instead of mailing to every address on file. This practice