The set of attributes, which with high possibility determines a class, is used to describe that class. A weight of evidence measure is given for each rule. This gives a measure for how certain it is that an object with the attribute values describing a class actually belongs to that class. To define the evidence of measure, we will need to define mutual information first.
measures the decrease or increase in uncertainty of assigning an object with attribute value to the class . If is positive, the uncertainty of classifying the object as belonging to class decreases, if it is negative, the uncertainty increases.
W may be interpreted as a measure of whether provides more evidence that an object belong to class rather that to the other classes. W is the difference in gain in information. If W is positive, provides positive evidence.
By using these definitions, characteristic descriptions of all classes is created. These are made up of those classification rules whose conclusion parts predict them.