Construction of Multiclass Classifier as Linear or Mixed Binary Programming Task

Kód výsledku v IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F60461373%3A22340%2F25%3A43933539" target="_blank" >RIV/60461373:22340/25:43933539 - isvavai.cz</a>

Nalezeny alternativní kódy

RIV/68407700:21340/25:00387678

Výsledek na webu

<a href="https://journals.pan.pl/Content/137602/PDF/art02.pdf" target="_blank" >https://journals.pan.pl/Content/137602/PDF/art02.pdf</a>

DOI - Digital Object Identifier

<a href="http://dx.doi.org/10.24425/acs.2025.157142" target="_blank" >10.24425/acs.2025.157142</a>

Jazyk výsledku

angličtina

Název v původním jazyce

Construction of Multiclass Classifier as Linear or Mixed Binary Programming Task

Popis výsledku v původním jazyce

Linear and mixed binary programming techniques, which arise from model linearity, are widely supported by advanced optimization solvers. In this paper, we present a comprehensive guide for transforming linear classifiers into linear or mixed binary programming tasks. Our approach employs widely used techniques, such as Kesler construction with either perfect or imperfect learning, and weight regularization using the L1 or L0 norm, enhanced by additional maximum weight constraint (i.e., L∞). Various linear optimization tasks are formulated based on performance measures such as accuracy and sensitivity. The classifiers are constructed using different weight penalizations and regularizations – specifically, the L0 norm, which yields mixed binary programming tasks with NP-hard complexity, and the L1 norm, which results in linear programming tasks with polynomial complexity, both with an additional maximum weight constraint. The proposed classifiers are compared on several UCI datasets (Iris Flower, Wine, and Seeds) and match or outperform Ridge and Lasso regression methods when applied to classification tasks.

Název v anglickém jazyce

Construction of Multiclass Classifier as Linear or Mixed Binary Programming Task

Popis výsledku anglicky

Linear and mixed binary programming techniques, which arise from model linearity, are widely supported by advanced optimization solvers. In this paper, we present a comprehensive guide for transforming linear classifiers into linear or mixed binary programming tasks. Our approach employs widely used techniques, such as Kesler construction with either perfect or imperfect learning, and weight regularization using the L1 or L0 norm, enhanced by additional maximum weight constraint (i.e., L∞). Various linear optimization tasks are formulated based on performance measures such as accuracy and sensitivity. The classifiers are constructed using different weight penalizations and regularizations – specifically, the L0 norm, which yields mixed binary programming tasks with NP-hard complexity, and the L1 norm, which results in linear programming tasks with polynomial complexity, both with an additional maximum weight constraint. The proposed classifiers are compared on several UCI datasets (Iris Flower, Wine, and Seeds) and match or outperform Ridge and Lasso regression methods when applied to classification tasks.

Druh

J_imp - Článek v periodiku v databázi Web of Science

CEP obor

—

OECD FORD obor

10102 - Applied mathematics

Projekt

<a href="/cs/project/LUC23138" target="_blank" >LUC23138: Role zánětu v progresi aterosklerózy studovaná pomocí metabolomických a proteomických nástrojů</a><br>

Návaznosti

P - Projekt vyzkumu a vyvoje financovany z verejnych zdroju (s odkazem do CEP)

Rok uplatnění

2025

Kód důvěrnosti údajů

S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů

Název periodika

Archives of Control Sciences

ISSN

2300-2611

e-ISSN

2300-2611

Svazek periodika

35

Číslo periodika v rámci svazku

4

Stát vydavatele periodika

PL - Polská republika

Počet stran výsledku

27

Strana od-do

"655–681"

Kód UT WoS článku

001653381800002

EID výsledku v databázi Scopus

2-s2.0-105029633699