Java Data Mining Concepts - Java Data Mining: Strategy, Standard, and Practice - page 159

If customer has savings Then customer will have checking

Consequent

Antecedent

Figure 7-10 Association rule.

to limit the number of rules, and hence reduce model size, allowing
users to focus on a more relevant subset of possible rules.
Rule quality can be specified using one or more metrics, such as
support, confidence , and lift . Figure 7-10 shows the rule “If customers
have a savings account, then they have a checking account” derived from
the data in the Table 7-10. A rule consists of a condition part, called
the antecedent , and a result part, called the consequent . Using this
example we illustrate the rule quality metrics. Figure 7-11 shows two
sets, the first represents the cases that conform to the antecedent and
the second represents the cases that conform to the consequent. The
intersection of these two sets represents the cases that conform to
both antecedent and consequent, that is, the rule.
The support of a rule is the ratio of cases that match the rule when
compared to the total number of records in the dataset. In this exam-
ple, there are two customers, 1 and 3, that conform to the rule out of
five customers, so support for this rule is 2/5 0.4.

The confidence of a rule is the ratio of the number of records that
include all items in the rule to the number of records that include all
items in the antecedent part of the rule. In this example, there are four
customers, 1, 2, 3, and 5, that match the antecedent portion of the rule,

Customers who have
both checking and
savings accounts

Customers
who have
savings
account

Customers
who have
checking
account

2, 5 1, 3 4

Figure 7-11 Associations.