Dear all
I would be presenting on MetaCost, this friday. The following is the abstract.
Abstract:
MetaCost: A general method for making Classifiers cost-sensitive
This is a paper written by Pedro Domingos.
A variety of algorithms are available for classification. But many of these assume that the errors which are generated are all of the same cost. This is not so in real world problems. We must have costs associated with errors of varying degrees. The paper proposes a method for making the classifier, cost-sensitive by wrapping a cost-minimizing procedure around it. This is called as the MetaCost. The paper compares the results by running experiments on data from the UCI repository. It compares the MetaCost algorithm, the traditional C4.5 and the two types of stratification - undersampling and oversampling. It also breifly talks about the related and the future works of this method.
--
See you all there..
Bye
Narasimha Deepak Kolippakkam