A study on the effect of class distribution using cost-sensitive learning

Kai Ming Ting

doi:10.1007/3-540-36182-0_11

A study on the effect of class distribution using cost-sensitive learning

Kai Ming Ting

Research output: Chapter in Book/Report/Conference proceeding › Conference Paper › Research › peer-review

7 Citations (Scopus)

Abstract

This paper investigates the effect of class distribution on the predictive performance of classification models using cost-sensitive learning, rather than the sampling approach employed previously by a similar study. The predictive performance is measured using the cost space representation, which is a dual to the ROC representation. This study shows that distributions which range between the natural distribution and the balanced distribution can also produce the best models, contrary to the finding of the previous study. In addition, we find that the best models are larger in size than those trained using the natural distribution. We also show two different ways to achieve the same effect of the corrected probability estimates proposed by the previous study.

Original language	English
Title of host publication	Discovery Science
Subtitle of host publication	5th International Conference, DS 2002 Lubeck, Germany, November 24-26, 2002 Proceedings
Place of Publication	Berlin Germany
Publisher	Springer
Pages	98-112
Number of pages	15
ISBN (Print)	3540001883
DOIs	https://doi.org/10.1007/3-540-36182-0_11
Publication status	Published - 2002
Event	International Conference on Discovery Science 2002 - Lubeck, Germany Duration: 24 Nov 2002 → 26 Nov 2002 Conference number: 5th

Publication series

Name	Lecture Notes in Computer Science
Publisher	Springer
Volume	2534
ISSN (Print)	0302-9743

Conference

Conference	International Conference on Discovery Science 2002
Abbreviated title	DS 2002
Country/Territory	Germany
City	Lubeck
Period	24/11/02 → 26/11/02

Access to Document

10.1007/3-540-36182-0_11

Cite this

@inproceedings{d785d3f6ff3f4e67925d4dad9cc99548,

title = "A study on the effect of class distribution using cost-sensitive learning",

abstract = "This paper investigates the effect of class distribution on the predictive performance of classification models using cost-sensitive learning, rather than the sampling approach employed previously by a similar study. The predictive performance is measured using the cost space representation, which is a dual to the ROC representation. This study shows that distributions which range between the natural distribution and the balanced distribution can also produce the best models, contrary to the finding of the previous study. In addition, we find that the best models are larger in size than those trained using the natural distribution. We also show two different ways to achieve the same effect of the corrected probability estimates proposed by the previous study.",

author = "Ting, \{Kai Ming\}",

year = "2002",

doi = "10.1007/3-540-36182-0\_11",

language = "English",

isbn = "3540001883",

series = "Lecture Notes in Computer Science",

publisher = "Springer",

pages = "98--112",

booktitle = "Discovery Science",

address = "Switzerland",

note = "International Conference on Discovery Science 2002, DS 2002 ; Conference date: 24-11-2002 Through 26-11-2002",

}

Ting, KM 2002, A study on the effect of class distribution using cost-sensitive learning. in Discovery Science: 5th International Conference, DS 2002 Lubeck, Germany, November 24-26, 2002 Proceedings. Lecture Notes in Computer Science, vol. 2534, Springer, Berlin Germany, pp. 98-112, International Conference on Discovery Science 2002, Lubeck, Germany, 24/11/02. https://doi.org/10.1007/3-540-36182-0_11

A study on the effect of class distribution using cost-sensitive learning. / Ting, Kai Ming.
Discovery Science: 5th International Conference, DS 2002 Lubeck, Germany, November 24-26, 2002 Proceedings. Berlin Germany: Springer, 2002. p. 98-112 (Lecture Notes in Computer Science; Vol. 2534).

Research output: Chapter in Book/Report/Conference proceeding › Conference Paper › Research › peer-review

TY - GEN

T1 - A study on the effect of class distribution using cost-sensitive learning

AU - Ting, Kai Ming

N1 - Conference code: 5th

PY - 2002

Y1 - 2002

N2 - This paper investigates the effect of class distribution on the predictive performance of classification models using cost-sensitive learning, rather than the sampling approach employed previously by a similar study. The predictive performance is measured using the cost space representation, which is a dual to the ROC representation. This study shows that distributions which range between the natural distribution and the balanced distribution can also produce the best models, contrary to the finding of the previous study. In addition, we find that the best models are larger in size than those trained using the natural distribution. We also show two different ways to achieve the same effect of the corrected probability estimates proposed by the previous study.

AB - This paper investigates the effect of class distribution on the predictive performance of classification models using cost-sensitive learning, rather than the sampling approach employed previously by a similar study. The predictive performance is measured using the cost space representation, which is a dual to the ROC representation. This study shows that distributions which range between the natural distribution and the balanced distribution can also produce the best models, contrary to the finding of the previous study. In addition, we find that the best models are larger in size than those trained using the natural distribution. We also show two different ways to achieve the same effect of the corrected probability estimates proposed by the previous study.

UR - https://www.scopus.com/pages/publications/30144439931

U2 - 10.1007/3-540-36182-0_11

DO - 10.1007/3-540-36182-0_11

M3 - Conference Paper

SN - 3540001883

T3 - Lecture Notes in Computer Science

SP - 98

EP - 112

BT - Discovery Science

PB - Springer

CY - Berlin Germany

T2 - International Conference on Discovery Science 2002

Y2 - 24 November 2002 through 26 November 2002

ER -

A study on the effect of class distribution using cost-sensitive learning

Abstract

Publication series

Conference

Access to Document

Other files and links

Cite this