The H measure of classification performance
The H measure of classification performanceinformation on theory and software related to the H measure of classification performance
The H-measure is a coherent alternative to the Area Under the ROC Curve (AUC) for measuring classification performance. Authors: David J. Hand, Christoforos Anagnostopoulos (maintainer).
- R package: a fully documented R package may be downloaded from CRAN: http://cran.r-project.org/web/packages/hmeasure/. For manual downloads see below.
- R package (MAC OS / LINUX): hmeasure_1.0.tar.gz (see readme.txt for instructions).
- R package (WINDOWS): hmeasure_1.0.zip (instructions as above).
- Matlab v0.1 implementation: hmeasure.m and an example file hmeasure_example.m
- Hand, D.J. and Anagnostopoulos, C., A better Beta for the H measure of classification performance, arXiv preprint, http://arxiv.org/abs/1202.2564
- D.J. Hand. Measuring classifier performance: a coherent alternative to the area under the ROC curve. Machine Learning, 77:103–123, 2009.
- D.J. Hand. Evaluating diagnostic tests: the area under the ROC curve and the balance of errors. Statistics in Medicine, 29:1502–1510, 2010.
The H-measure in one picture: The H measure can be motivated in terms of a prior on the severity of the two types of misclassification costs (false alarms versus missed positive cases). Under this interpretation, the prior weights implicitly imposed on the relative misclassification cost by the popular Area Under the Curve (see leftmost plot) are classifier-dependent (rightmost plot), whereas in the case of the H-measure they can be controlled explicitly using a Beta prior (centre plot).