Elaboration on Two Points Raised in "Classifier Technology and the Illusion of Progress"
- Robert Holte, Department of Computing Science, University of Alberta
This short note elaborates two points raised in David Hand’s target article. First, I provide additional evidence that simple classification rules should be given serious consideration in any application and that there are often diminishing returns in considering increasingly complex classifiers. Second, I refine Hand’s basic argument that small improvements in performance are irrelevant because of the uncertainty about many aspects of the situation in which the classifier will be deployed. In particular, I briefly describe a recently developed method for analyzing and comparing classifier performance when the class ratios and misclassification costs are unknown. This does not refute his general argument, but it does provide an important exception to it.
Citation
R. Holte. "Elaboration on Two Points Raised in "Classifier Technology and the Illusion of Progress"". Statistical Science, 21(1), pp 24-26, January 2006.Keywords: | machine learning |
Category: | In Journal |
BibTeX
@article{Holte:StatisticalScience06, author = {Robert Holte}, title = {Elaboration on Two Points Raised in "Classifier Technology and the Illusion of Progress"}, Volume = "21", Number = "1", Pages = {24-26}, journal = {Statistical Science}, year = 2006, }Last Updated: June 04, 2007
Submitted by Staurt H. Johnson