Java Reference

In-Depth Information

so the confidence is 2/4

0.5. Confidence is directional; when we

invert the rule, for example, “
If a customer has a checking account, then

they have a savings account
,” we will get a different confidence value.

For the inverted rule, there are only three customers, 1, 3, and 4, that

satisfy antecedent, so the confidence value of this rule is 2/3

0.67.

Note that the inverted rule has greater confidence than the original.

The
lift
is the ratio between the rule confidence and its
expected

confidence.
Expected
confidence
is the frequency of the consequent in

the data. Lift measures how much more likely the consequent is

when an antecedent happens. In this example, there are three

customers, 1, 3 and 4, that have a checking account, so the expected

confidence is 3/5

0.66.

In addition to the rule quality metrics, JDM allows users to specify

taxonomy per attribute (Section 4.5), and settings that include the

maximum number of rules in the model and inclusion or exclusion of

model items. Section 9.7 will discuss more about these setting when

we discuss the API usage.

0.6. Hence
lift
for this rule is 0.4/0.6

7.4.5

Use Model Content: Explore Rules From the Model

An association model primarily contains the association rules and

their support, confidence, and lift details. Even with the model rule

quality thresholds, this model may contain a large number of rules

based on the number of items and the relationships among these

items. To explore the rules, users often need a filter and to order the

rules to get an interesting or manageable subset. To this end, JDM

provides rule filtering capabilities.

Filtering criteria may include rule support, confidence, and lift

thresholds; inclusion or exclusion of the specified items from the rule

or specific rule components, that is, antecedent and consequent; and

rule or rule component length. Section 9.7 will discuss more about

the various types of filtering criteria using JDM.

7.5

Clustering Problem

7.5.1

Problem Definition: How to Understand Customer

Behavior and Needs

ABCBank has thousands of customers whose profiles and needs

widely differ from each other. ABCBank wants to understand customer

segments to design new products and personalize campaigns to

Search WWH ::

Custom Search