Practical Applications of Data Mining

3.3 separate-and-Conquer approaCh 83

and adding tests to the rule until all the subset of instances covered by the rule

are “pure” (e.g., all members belong to only one class). The covered subset

of instances is then excluded from further processing. The rule-generation

process continues until no more unclassified instances are left in the dataset.

The advantage of this approach is time efficiency as a result of the following

two characteristics: first, it creates knowledge rules directly without inducing

an intermediate decision tree; secondly, it immediately excludes instances

covered by a newly created rule from further induction.

The criteria used for test selection and the standard of “purity” vary

among different algorithms ...