Performance measures are used in various stages of the process aimed at solving a classifi- cation problem. Unfortunately, most of these measures are in fact biased, meaning that they strictly depend on the class ratio – i.e. on the imbalance between negative and positive samples. After pointing to the source of bias for the best known measures, novel unbiased measures are defined which are able to capture the concepts of discriminant and characteristic capability. The combined use of these measures can give important information to researchers involved in machine learning or pattern recognition tasks, in particular for classifier performance assessment and feature selection.

A direct measure of discriminant and characteristic capability for classifier building and assessment

ARMANO, GIULIANO
2015-01-01

Abstract

Performance measures are used in various stages of the process aimed at solving a classifi- cation problem. Unfortunately, most of these measures are in fact biased, meaning that they strictly depend on the class ratio – i.e. on the imbalance between negative and positive samples. After pointing to the source of bias for the best known measures, novel unbiased measures are defined which are able to capture the concepts of discriminant and characteristic capability. The combined use of these measures can give important information to researchers involved in machine learning or pattern recognition tasks, in particular for classifier performance assessment and feature selection.
2015
Classifier performance measures; Confusion matrices; Feature ranking/selection; Artificial Intelligence; Software; Control and Systems Engineering; Theoretical Computer Science; Computer Science Applications1707 Computer Vision and Pattern Recognition; Information Systems and Management
File in questo prodotto:
File Dimensione Formato  
2015-INF-SCIENCES-armano.pdf

Solo gestori archivio

Tipologia: versione editoriale
Dimensione 2.43 MB
Formato Adobe PDF
2.43 MB Adobe PDF   Visualizza/Apri   Richiedi una copia

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11584/134257
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 13
  • ???jsp.display-item.citation.isi??? 10
social impact