Generalization operators are used to relax the conditional part of a rule. This means that they will have as a probable consequence a bigger number of examples covered. On the other hand this raise on coverage could lead to an increase on the number of incorrectly covered examples. Those two important issues are captured by completeness and consistency ratios which play an important role on the evaluation function (see section 2.5) that guides the search transformations.
The drop-condition operator is one of the generalization operators that YAILS uses. Being a rule a set of conditions together with a conclusion, this operator simply drops one of the conditions of the rule, thus relaxing its conditional part.
The conditions of the rules can be numerical intervals. This operator basically enlarges the interval of one numerical condition. Bellow a series of examples of this type of generalization is presented :
x in ]20..30] => x in [20..30]
temp in ]30..50] => x in [10..50]
y in [-1..45] => y >= -1
z >= 26.7 => z > 0
Yails uses a conservative but at the same time efficient algorithm to enlarge numerical intervals. Given an interval to enlarge YAILS starts by building an ordered list of the known values of the attribute. Then it tries to enlarge the interval up to one of the values of the list with the restriction of not including a value of a negative example.