To this end, data mining technology can be applied to identify the
profiles of respondents who answered one or more questions a certain
way. It can also be used to understand how respondents are grouped
or segmented. Such results enable focused sales and marketing efforts
to well-defined groups, as well as corrective action [SPSS 2003].
Data in surveys is often structured but can also be unstructured .
Structured data contains discrete responses such as multiple choice
questions involving demographic data, for example, age and income
level, or statements to be rated from “strongly agree” to “strongly
disagree.” Unstructured data can consist of short free-form
responses, for example, the “other” category with a line to specify a
value, or longer free-form questions such as “What was your best
and worst experience with our company?” where the user responds
To address free-form responses, text mining can be introduced by
extracting important terms from the text and combining it with
structured data before mining. Some data mining vendors will per-
form this term extraction automatically, others may require explicit
Every time you apply for a loan, mortgage, or credit card, your credit
history is checked and your current financial situation is assessed to
determine a “credit score.” This score indicates the type of risk you
represent to the financial institution issuing the credit. Credit scoring
takes into account various information such as customer demo-
graphics, loan history, deposits and assets, total credit line, outstand-
ing debt, etc.
The more accurate the score, the more likely the financial institu-
tion is to make a correct decision on a customer. Although there are
always unexpected factors for individuals defaulting on loans, such
as job loss or illness, the credit score provides important input to the
loan decisioning process.
Historically, statistical methods were employed for credit scoring.
Today, data mining plays an increasingly important role in determin-
ing credit worthiness due to the large number of predictor attributes
that exist on customers.
A typical approach for building credit scoring models uses super-
vised learning, where first a credit score for each of a set of customers