Regression approaches for multi-class cost-sensitive classification
Date Issued
2009
Date
2009
Author(s)
Tu, Han-Hsing
Abstract
Cost-sensitive classification is an important research problem in recentears. It allows machine learning algorithms to use the additional cost informationo make more strategic decisions.tudies on binary cost-sensitive classification have led to promising resultsn theories, algorithms, and applications. The multi-class counterpart islso needed in many real-world applications, but is more difficult to analyze.his thesis focuses on multi-class cost-sensitive classification.xisting methods for multi-class cost-sensitive classification usually transformhe cost information into example importance (weight). This thesis offers different viewpoint of the problem, and proposes a novel method. Weirectly estimate the cost value corresponding to each prediction using regression,nd outputs the label that comes with the smallest estimated cost.e improve the method by analyzing the errors made during the decision.hen, we propose a different regression loss function that tightly connectsith the errors. The new loss function leads to a solid theoretical guaranteef error transformation. We design a concrete algorithm for the loss functionith the support vector machines. The algorithm can be viewed as a theoreticallyustified extension the popular one-versus-all support vector machine.xperiments using real-world data sets with arbitrary cost values demonstratehe usefulness of our proposed methods, and validate that the cost informationhould be appropriately used instead of dropped.
Subjects
multi-class cost-sensitive classification
cost information
regression
support vector machines
Type
thesis
File(s)![Thumbnail Image]()
Loading...
Name
ntu-98-R96922139-1.pdf
Size
23.32 KB
Format
Adobe PDF
Checksum
(MD5):e55fc2802e1ae25de7f82d11656376c4