Data Mining Process - Java Data Mining: Strategy, Standard, and Practice - page 58

deployment of a data mining solution in a call center application
may involve rebuilding models with the latest customer data on a
weekly basis at the central data center, and exporting these models
to geographically local call centers. The customer service represen-
tative (CSR) application accepts new data obtained by the CSR's
interaction with a customer. The application uses the data mining
model to generate predictions indicating anything from products
the customer is likely to buy, to whether the customer is likely to ter-
minate service or leave for a competitor, to what is the customer's
frustration index.
By virtue of being an API, a JDM program captures the sequence
of data mining operations performed, which can be useful for devel-
oping a report, or more importantly, for deploying a functioning sys-
tem. However, JDM also defines an array of data mining objects, such
as models, settings, and tasks. JDM also defines result objects, which
when combined with the other mining objects can be used to recount
the data mining process the user went through to develop the solu-
tion. JDM further supports the deployment phase through interfaces
for exporting and importing models and other mining objects, apply-
ing models to data (i.e., scoring), and, of course, building models.

3.2 A More Detailed View of Data Analysis and Preparation

We have just explored the data mining process according to CRISP-
DM, end to end. When planning a data mining project, consider that
particular phases require a disproportionate percentage of the over-
all time and effort. For example, among the data analysis, data
preparation, and modeling phases, it is often said that 80 percent of
the time is spent in data analysis and preparation, and only 20 per-
cent on modeling. Given this, no discussion of data mining is com-
plete without a discussion of data preparation. However, to do
justice to the topic of data analysis and preparation requires more
space than we will devote to it here. Indeed, entire topics are written
on the subject, as data preparation is viewed as an art as much as a
science [Pyle 1999].
Section 3.1 discusses how JDM supports both data analysis and
data preparation. However, JDM 1.1 does not support any specific
data mining transformations. The expert group decided to limit the
scope of the first release of JDM, making it more manageable and, as
a result, the general data transformation problem was deferred to a sub-
sequent Java Specification Request (JSR) or JDM release. As such,
transformations are now being addressed in JDM 2.0.