Abstract
We explore the use of the supra-classifier framework in the
construction of a classifier knowledge base. Previously, we
introduced this framework within which labels produced by old
classifiers are used to improve the generalization performance of a
new classifier for a different but related classification task.
We showed empirically that a simple Hamming
nearest neighbor is superior to other techniques (e.g. MLP, decision
trees, Naive Bayes, Combiners) as a supra-classifier. Here, we
describe theoretically how the probability that the Hamming nearest
neighbor supra-classifier will predict the true target class
approaches certainty at an exponential rate as more classifiers are
reused. The scalability of the Hamming nearest neighbor with large
numbers of previously created classifiers makes it a good choice as a
supra-classifier in the application of building a repository of domain
knowledge organized as a classifier knowledge base.