This paper presents a new formulation for cost-sensitive learning that we call the One-Benefit formulation. Instead of having the correct label for each training example as in the standard classifier learning formulation, in this formulation we have one possible label for each example (which may not be the correct one) and the benefit (or cost) associated with that label. The goal of learning in this formulation is to find the classifier that maximizes the expected benefit of the labelling using only these examples. We present a reduction from One-Benefit learning to standard classifier learning that allows us to use any existing error-minimizing classifier learner to maximize the expected benefit in this formulation by correctly weighting the examples. We also show how to evaluate a classifier using test examples for which we only the benefit for one of the labels. We present preliminary experimental results using a synthetic data generator that allows us to test both our learning method and our evaluation method.
By: Bianca Zadrozny
Published in: RC23760 in 2005
LIMITED DISTRIBUTION NOTICE:
This Research Report is available. This report has been submitted for publication outside of IBM and will probably be copyrighted if accepted for publication. It has been issued as a Research Report for early dissemination of its contents. In view of the transfer of copyright to the outside publisher, its distribution outside of IBM prior to publication should be limited to peer communications and specific requests. After outside publication, requests should be filled only by reprints or legally obtained copies of the article (e.g., payment of royalties). I have read and understand this notice and am a member of the scientific community outside or inside of IBM seeking a single copy only.
Questions about this service can be mailed to reports@us.ibm.com .