Instance selection improves geometric mean accuracy: A study on imbalanced data classification

Research output: Contribution to journalArticlepeer-review

Standard Standard

Instance selection improves geometric mean accuracy: A study on imbalanced data classification. / Kuncheva, Ludmila; Arnaiz-Gonzalez, Alvar; Diez-Pastor, J.F. et al.
In: Progress in Artificial Intelligence, Vol. 2019, No. 2, 06.02.2019.

Research output: Contribution to journalArticlepeer-review

HarvardHarvard

APA

CBE

Kuncheva L, Arnaiz-Gonzalez A, Diez-Pastor JF, Gunn I. 2019. Instance selection improves geometric mean accuracy: A study on imbalanced data classification. Progress in Artificial Intelligence. 2019(2).

MLA

VancouverVancouver

Kuncheva L, Arnaiz-Gonzalez A, Diez-Pastor JF, Gunn I. Instance selection improves geometric mean accuracy: A study on imbalanced data classification. Progress in Artificial Intelligence. 2019 Feb 6;2019(2).

Author

Kuncheva, Ludmila ; Arnaiz-Gonzalez, Alvar ; Diez-Pastor, J.F. et al. / Instance selection improves geometric mean accuracy: A study on imbalanced data classification. In: Progress in Artificial Intelligence. 2019 ; Vol. 2019, No. 2.

RIS

TY - JOUR

T1 - Instance selection improves geometric mean accuracy: A study on imbalanced data classification

AU - Kuncheva, Ludmila

AU - Arnaiz-Gonzalez, Alvar

AU - Diez-Pastor, J.F.

AU - Gunn, Iain

PY - 2019/2/6

Y1 - 2019/2/6

N2 - A natural way of handling imbalanced data is to attempt to equalise the class frequencies and train the classifier of choice on balanced data. For two-class imbalanced problems, the classification success is typically measured by the geometric mean (GM) of the true positive and true negative rates. Here we prove that GM can be improved upon by instance selection, and give the theoretical conditions for such an improvement. We demonstrate that GM is non-monotonic with respect to the number of retained instances, which discourages systematic instance selection. We also show that balancing the distribution frequencies is inferior to a direct maximisation of GM. To verify our theoretical findings, we carried out an experimental study of 12 instance selection methods for imbalanced data, using 66 standard benchmark data sets. The results reveal possible room for new instance selection methods for imbalanced data.

AB - A natural way of handling imbalanced data is to attempt to equalise the class frequencies and train the classifier of choice on balanced data. For two-class imbalanced problems, the classification success is typically measured by the geometric mean (GM) of the true positive and true negative rates. Here we prove that GM can be improved upon by instance selection, and give the theoretical conditions for such an improvement. We demonstrate that GM is non-monotonic with respect to the number of retained instances, which discourages systematic instance selection. We also show that balancing the distribution frequencies is inferior to a direct maximisation of GM. To verify our theoretical findings, we carried out an experimental study of 12 instance selection methods for imbalanced data, using 66 standard benchmark data sets. The results reveal possible room for new instance selection methods for imbalanced data.

KW - Imbalanced data; geometric mean (GM); instance/prototype selection; nearest neighbour; ensemble methods

M3 - Article

VL - 2019

JO - Progress in Artificial Intelligence

JF - Progress in Artificial Intelligence

SN - 2192-6352

IS - 2

ER -