A similarity evaluation technique for data mining with an ensemble of classifiers

Evaluation of similarity is very important in data mining with an ensemble of classifiers. Similarity between instances is used to recognize the nearest neighbors of an instance, similarity between classes is necessary to derive the misclassification errors in the learning phase, and similarity between classifiers is used to evaluate the classifiers when they are integrated. In the similarity evaluation we use a training set consisting predicates that define relationships within the three sets: the set of instances, the set of classes, and the set of classifiers. We consider two ways to derive similarities.