KFU. Card publication. Design of the Best Linear Classifier for Box-Constrained Data Sets.

KAZAN
FEDERAL UNIVERSITY

- RUS
- 中文
- ES

DESIGN OF THE BEST LINEAR CLASSIFIER FOR BOX-CONSTRAINED DATA SETS.

Form of presentation

Articles in international journals and collections

Year of publication

2022

Язык

английский

Authors, employees of KFU

Gabidullina Zulfiya Ravilevna, author

Bibliographic description in the original language

Gabidullina, Z.R. Design of the Best Linear Classifier for Box-Constrained Data Sets. Lecture Notes in Computational Science and Engineering Tom: 141 Stranicy: 109 - 124

Annotation

We construct a binary linear classifier for n-dimensional data sets with the special box-constrained structure. Data sets with this structure arise naturally in many real-world areas. We apply a linear separability criterion proposed in J. Optim. Theory Appl. (2012, https://doi.org/10.1007/s10957-012-0155-x). The Minkowski difference of the two data sets allows us to reduce a two-class classification problem to the problem in more easy to solve form. The greatest benefit of this reduction is that it allows to compute the parameters of a linear binary classifier by way of exact formulas. For this reason, a proposed framework has low computational costs. We rigorously prove that the developed linear classification model provides the possibility of constructing the data separator (or pseudo-separator) which really has the best estimate of its thickness. There are studied both regular and singular cases of separability arising in the theory and practice of linear classification of data sets.

Keywords

data classification, separator, pseudo-separator, thickness of the geometric margin

The name of the journal

Lecture Notes in Computational Science and Engineering

URL

https://link.springer.com/chapter/10.1007/978-3-030-87809-2_9

Please use this ID to quote from or refer to the card

https://repository.kpfu.ru/eng/?p_id=271607&p_lang=2

Full metadata record

Field DC	Value	Language
dc.contributor.author	Gabidullina Zulfiya Ravilevna	ru_RU
dc.date.accessioned	2022-01-01T00:00:00Z	ru_RU
dc.date.available	2022-01-01T00:00:00Z	ru_RU
dc.date.issued	2022	ru_RU
dc.identifier.citation	Gabidullina, Z.R. Design of the Best Linear Classifier for Box-Constrained Data Sets. Lecture Notes in Computational Science and Engineering Том: 141 Страницы: 109 - 124	ru_RU
dc.identifier.uri	https://repository.kpfu.ru/eng/?p_id=271607&p_lang=2	ru_RU
dc.description.abstract	Lecture Notes in Computational Science and Engineering	ru_RU
dc.description.abstract	We construct a binary linear classifier for n-dimensional data sets with the special box-constrained structure. Data sets with this structure arise naturally in many real-world areas. We apply a linear separability criterion proposed in J. Optim. Theory Appl. (2012, https://doi.org/10.1007/s10957-012-0155-x). The Minkowski difference of the two data sets allows us to reduce a two-class classification problem to the problem in more easy to solve form. The greatest benefit of this reduction is that it allows to compute the parameters of a linear binary classifier by way of exact formulas. For this reason, a proposed framework has low computational costs. We rigorously prove that the developed linear classification model provides the possibility of constructing the data separator (or pseudo-separator) which really has the best estimate of its thickness. There are studied both regular and singular cases of separability arising in the theory and practice of linear classification of data sets.	ru_RU
dc.language.iso	ru	ru_RU
dc.subject	data classification	ru_RU
dc.subject	separator	ru_RU
dc.subject	pseudo-separator	ru_RU
dc.subject	thickness of the geometric margin	ru_RU
dc.title	Design of the Best Linear Classifier for Box-Constrained Data Sets.	ru_RU
dc.type	Articles in international journals and collections	ru_RU