SQL/MX Data Mining Guide

Introduction
HP NonStop SQL/MX Data Mining Guide—523737-001
1-11
Knowledge Deployment and Monitoring
Descriptive tasks involve finding patterns describing the data. The most common are:
•
Database segmentation (clustering)—Map a case into one of several clusters.
•
Summarization—Provide a compact description of the data, often in visual form.
•
Link analysis—Determine relationships between attributes in a case.
•
Sequence analysis—Determine trends over time.
You use a variety of algorithms, and the models they produce, to perform these
predictive and descriptive tasks.
For example, classification can be done by building a decision tree model, where each
branch of the tree is represented by a predicate involving attributes in the mining data
set and where each branch is homogeneous with respect to whether the predicate is
true or false. The main task in classification is to determine which predicates form the
decision tree that predicts the goal. The most common algorithms for classification
come from the field of machine learning in computer science.
Typically, the model building step involves the use of client-mining tools that require the
interactive participation of the user to guide the investigation. A description of these
special-purpose tools is beyond the scope of this manual.
For further information, see Section 4, Mining the Data.
Knowledge Deployment and Monitoring
The last two steps of the knowledge discovery process involve deploying and
monitoring discovered knowledge. Deployment can take many different forms. For
example, deployment might be as simple as documenting and reporting the results, or
deployment might be embedding the model in an operational system to achieve
predictive results.
Most data mining tools support model deployment either by applying a model to data
within the tool or by exporting a model as executable code, which can then be
embedded and used in applications. In the credit card attrition example, one form of
model deployment is to periodically use the model to identify profitable customers that
are likely to leave, and then to take some action, such as lowering interest rates or
waiving fees, to try to retain these customers.