Publication detail

PSO-based Constrained Imbalanced Data Classification

HLOSTA, M. STRÍŽ, R. ZENDULKA, J. HRUŠKA, T.

Original Title

PSO-based Constrained Imbalanced Data Classification

English Title

PSO-based Constrained Imbalanced Data Classification

Type

conference paper

Language

en

Original Abstract

The paper deals with classification of highly imbalanced data with accuracy constraints for the minority class. We solve this problem by our proposed meta-learning method that uses cost-sensitive logistic regression to generate initial candidate models. These models can be used as an initial solutions for various optimization algorithms. This paper is aimed for using Particle Swarm Optimization (PSO) to handle the constrained imbalanced classification problem. Experiments, comparing with Genetic Algorithm (GA), show that the swarm intelligence approach is suitable for this problem and outperforms GA.

English abstract

The paper deals with classification of highly imbalanced data with accuracy constraints for the minority class. We solve this problem by our proposed meta-learning method that uses cost-sensitive logistic regression to generate initial candidate models. These models can be used as an initial solutions for various optimization algorithms. This paper is aimed for using Particle Swarm Optimization (PSO) to handle the constrained imbalanced classification problem. Experiments, comparing with Genetic Algorithm (GA), show that the swarm intelligence approach is suitable for this problem and outperforms GA.

Keywords

Data mining, imbalance classification, constraints, PSO, Genetic Algorithm

RIV year

2013

Released

05.11.2013

Publisher

The University of Technology Košice

Location

Spišská Nová Ves

ISBN

978-80-8143-127-2

Book

Proceedings of the Twelth International Conference on Informatics INFORMATICS'2013

Edition

NEUVEDEN

Edition number

NEUVEDEN

Pages from

234

Pages to

239

Pages count

6

URL

BibTex


@inproceedings{BUT103558,
  author="Martin {Hlosta} and Rostislav {Stríž} and Jaroslav {Zendulka} and Tomáš {Hruška}",
  title="PSO-based Constrained Imbalanced Data Classification",
  annote="The paper deals with classification of highly imbalanced data with accuracy
constraints for the minority class. We solve this problem by our proposed
meta-learning method that uses cost-sensitive logistic regression to generate
initial candidate models. These models can be used as an initial solutions for
various optimization algorithms. This paper is aimed for using Particle Swarm
Optimization (PSO) to handle the constrained imbalanced classification problem.
Experiments, comparing with Genetic Algorithm (GA), show that the swarm
intelligence approach is suitable for this problem and outperforms GA.",
  address="The University of Technology Košice",
  booktitle="Proceedings of the Twelth International Conference on Informatics INFORMATICS'2013",
  chapter="103558",
  edition="NEUVEDEN",
  howpublished="print",
  institution="The University of Technology Košice",
  year="2013",
  month="november",
  pages="234--239",
  publisher="The University of Technology Košice",
  type="conference paper"
}