The third phase of Chan and Wong's method([CW91]) is to use the description of the classes (the rules) and their weight of evidence to classify objects not in the original training set. Such an object, obj, is characterized by . This object is matched against the classification rules, and if an attribute value satisfies the condition part of a rule predicting class , the description of obj partially matches .
Due to uncertainty or incompleteness in the training set, the description of obj might partially match more than one class. To choose the correct, or most plausible class, the weight of evidence that obj belongs to class in favor of the other classes is used. If m number of attributes of obj match the class description of , we get the following weight of evidence:
Here denote the correct class obj belongs to and denote the matching attribute. The weight of evidence thus gives a measure of the evidence that obj belongs to class compared to the evidence that it do not.
By calculating the weight of evidence above for all classes for each object, we get the evidence of the object belonging to each class. The classification given for an object, is the class which has the highest weight of evidence for that particular object.
Another similar way of choosing the best rule is by using the J-measure which is defined in [SG91] as:
This could be used in a similar way as the weight of evidence measure to find the best classification.