Deakin University
Browse

File(s) under permanent embargo

Clustering aided support vector machines

conference contribution
posted on 2017-01-01, 00:00 authored by G Ristanoski, R Soni, Sutharshan RajasegararSutharshan Rajasegarar, J Bailey, C Leckie
Support Vector Machines (SVMs) have proven to be an effective approach to learning a classifier from complex datasets. However, highly nonhomogeneous data distributions can pose a challenge for SVMs when the underlying dataset comprises clusters of instances with varying mixtures of class labels. To address this challenge we propose a novel approach, called a cluster-supported Support Vector Machine, in which information derived from clustering can be incorporated directly into the SVM learning process. We provide a theoretical derivation to show that when the total empirical loss is expressed in terms of the combined quadratic empirical loss from each cluster, we can still find a formulation of the optimisation problem that is a convex quadratic programming problem. We discuss the scenarios where this type of model would be beneficial, and present empirical evidence that demonstrates the improved accuracy of our combined model.

History

Volume

10358

Pagination

322-334

Location

New York, USA

Start date

2017-07-15

End date

2017-07-20

ISSN

0302-9743

eISSN

1611-3349

ISBN-13

9783319624150

Publication classification

E Conference publication, E1 Full written paper - refereed

Copyright notice

2017, Springer International Publishing AG

Editor/Contributor(s)

Perner P

Title of proceedings

MLDM 2017 : Proceedings of the 13th International Machine Learning and Data Mining in Pattern Recognition Conference

Event

Machine Learning and Data Mining in Pattern Recognition. Conference (13th : 2017 : New York, USA)

Publisher

Springer

Place of publication

Cham, Switzerland

Series

Lecture Notes in Artificial Intelligence

Usage metrics

    Research Publications

    Exports

    RefWorks
    BibTeX
    Ref. manager
    Endnote
    DataCite
    NLM
    DC