Weighted one-against-all

Abstract

The one-against-all reduction from multiclass classification to binary classification is a standard technique used to solve multiclass problems with binary classifiers. We show that modifying this technique in order to optimize its error transformation properties results in a superior technique, both experimentally and theoretically. This algorithm can also be used to solve a more general classification problem "multi-label classification," which is the same as multiclass classification except that it allows multiple correct labels for a given example. Copyright © 2005, American Association for Artificial Intelligence (www.aaai.org). All rights reserved.