Data Mining Process - Java Data Mining: Strategy, Standard, and Practice - page 79

Compare
Test
Results

Retrieve Monthly
Data from Data Warehouse
@ 00:00 first of each month

Prepare
Data

Build
Model 1

Test
Model 1

Request
Approval

Build
Model 2

Test
Model 2

Deploy
Model
@03:00 second
of each month

Yes

Approved?

Build
Model 3

Test
Model 3

No

Submit to
Data Analyst
for Review

Figure 3-15 A data mining workflow.

To ensure against deploying an inadequate model, which, for
example, could impact customers contacting a call center, we intro-
duce a manual approval step. If the model is approved, it is
deployed to the operational system. If not approved, the problem is
submitted to a data analyst for review. Since approval or disap-
proval is expected within one day, we have a time dependency to
deploy the model at 3:00 A . M . on the second day of the month. If a
decision is not made in time, the task does not execute.

3.6 Advances in Automated Data Mining

“Why can't I just point a data mining tool at my data warehouse and
say, 'Find something interesting' or 'Solve my business problem'? We
can map the human genome, but I still have to spend countless
hours, days, weeks, even months, extracting knowledge from my
data warehouse or solving data mining problems, many of which
must have been solved before. With all the intelligence put into
applications today, why don't the data mining vendors produce soft-
ware that automates this data mining process?”
These are the rantings of many a novice in data mining. In fact,
more and more applications are building in data mining-based intel-
ligence, providing industry-specific and problem-specific interfaces,
without users even being aware of data mining's presence. However,
there are many situation-specific problems that still warrant custom-
ized solutions. For these, data mining software is increasingly getting