SQL/MX Data Mining Guide

Introduction
HP NonStop SQL/MX Data Mining Guide—523737-001
1-5
Defining the Business Opportunity
•
Usability of the results
Merely identifying patterns is not enough. The opportunity and analysis must be
structured so that any interpretation of results obtained develops into deployable
business strategy.
•
Political and organizational reaction
In assessing probabilities for organizational resistance, it is helpful to examine
similar past efforts and understand why these efforts succeeded or failed.
•
Availability of business analysts and data mining experts and technology
Are data, domain, and mining experts available to participate in the process? Is
sufficient technology, both hardware and software, available?
•
Data availability
Does preclassified data exist or can it be derived? Do sufficiently large amounts of
data exist? Both internal and external data sources should be considered.
•
Logistics
How difficult is it to collect, extract, and transport the relevant data? Is
confidentiality an issue?
Careful consideration of these factors helps to ensure that the opportunity selected is
both amenable to data mining and likely to provide significant value.
After an operation is selected, the next task is to specify it precisely. In the scenario of
building a model to predict credit card account attrition, the goal is to build a model that
will predict, as early as possible, whether a credit card customer will close their
account.
To specify this opportunity precisely, decide on an explicit definition of attrition, such as
when a customer calls and closes their account. Another option is implicit—when a
customer stops using their card. For simplicity, define attrition as a customer closing
their account or maintaining a zero balance for three months.
Another aspect of specifying the opportunity is defining what it means to predict as
early as possible when an account will be closed. For this example, choose three
months as the prediction window. This window should be long enough to allow the card
issuer to take some action to try to retain customers likely to leave, but short enough to
capture attrition-related patterns.
The goal is to build a model that will predict, as early as possible, customer attrition.
Example Business Opportunity
The precise specification of our example opportunity is to build a model that will predict
at any point in time, based on such things as current account status, account activity,
and demographics, whether a credit card customer will close their account in the
future. Note that the precise specification of the opportunity might be modified or