Classifiers in the Form of Rulesets

Abstract

Some classifiers take the form of so-calledif-then rules: if the conditions from theif-part are satisfied, the example is labeled with the class specified in thethen-part. Typically, the classifier is represented not by a single rule, but by a set of rules, aruleset. The paradigm has certain advantages. For one thing, the rules capture the underlying logic, and therefore facilitate explanations of why an example has to be labeled with the given class; for another, induction of rulesets is capable of discovering recursive definitions, something that is difficult to accomplish within other machine-learning paradigms.