Java Data Mining Concepts - Java Data Mining: Strategy, Standard, and Practice - page 157

attributes sorted by their importance values and assigns ranks to
each attribute based on its importance value. In JDM, selecting
attributes can be based on a percentage of attributes, number of
attributes, or importance value threshold. For example, if a user
gives the importance threshold as 0.0, then native country, county, and
state attributes will be filtered from the returned set of attributes,
because they have negative importance values. Typically, attributes
with negative importance values may adversely impact model quality.
However, the interpretation of the importance values depends on the
implementation.

7.4 Association Rules Problem

7.4.1 Problem Definition: How to Identify Cross-Sell Products

for Customers

Part of ABCBank's strategy for customer retention and increased
profits is to cross-sell products to existing customers. When customers
have multiple products with the same bank, they are less likely to
attrite, if for no other reason than the effort and complexity involved in
switching to another bank. As a result, ABCBank wants to understand
which set of products are normally purchased by customers, so that
ABCBank can recommend appropriate cross-sell products to customers
missing products in that set. Refer to [DeBlasio 2001] for a real-world
cross-sell scenario in the banking industry.

7.4.2 Solution Approach: Discover Product Associations

from Customer Data

ABCBank has customer data, which includes the products or services
each customer purchased along with some usage characteristics, for
example, average checking account balance. Using this data,
ABCBank wants to identify product associations, such as if a customer
has a single account with the bank, or in what other accounts or
financial services might that customer be interested.
Product associations are discovered and measured by learning
product purchase patterns from the historical purchase data. For
example, an association model may identify a rule indicating that a
customer with an average annual checking account between $3,000
and $5,000 and savings account greater than $15,000 is likely to open
a certificate of deposit.