Co-training with relevant random subspaces

Yusuf Yaslan; Zehra Cataltepe

doi:10.1016/j.neucom.2010.01.018

Co-training with relevant random subspaces

Yusuf Yaslan^*, Zehra Cataltepe

^*Corresponding author for this work

Department of Computer Engineering

Istanbul Technical University

Research output: Contribution to journal › Article › peer-review

78 Citations (Scopus)

Abstract

We introduce the relevant random subspace Co-training (Rel-RASCO) algorithm which produces relevant random subspaces and then does semi-supervised ensemble learning using those subspaces and unlabeled data. Ensemble learning algorithms may benefit from diversity of classifiers used. However, for high dimensional data choosing subspaces randomly, as in RASCO (Random Subspace Method for Co-training, Wang et al. 2008 [5]) algorithm, may produce diverse but inaccurate classifiers. We produce relevant random subspaces by means of drawing features with probabilities proportional to their relevances measured by the mutual information between features and class labels. We show that Rel-RASCO achieves better accuracy by this relevant and random subspace selection scheme. Experiments on five real and one synthetic datasets show that Rel-RASCO algorithm outperforms both RASCO and Co-training in terms of the accuracy achieved at the end of Co-training.

Original language	English
Pages (from-to)	1652-1661
Number of pages	10
Journal	Neurocomputing
Volume	73
Issue number	10-12
DOIs	https://doi.org/10.1016/j.neucom.2010.01.018
Publication status	Published - Jun 2010

Funding

This work was partially supported by Tubitak (The Scientific and Technological Research Foundation of Turkey) research project 109E162 and Istanbul Technical University BAP (Scientific Research Projects) ‘Co-training on High Dimensional Datasets’ project. Authors also would like to thank the anonymous reviewers whose comments greatly improved the quality of the paper.

Funders	Funder number
Istanbul Technical University BAP
Scientific and Technological Research Foundation of Turkey	109E162

Keywords

Co-training
Multiple classifier systems
Random subspace methods
RASCO
Relevant subspace method
Semi-supervised learning

Access to Document

10.1016/j.neucom.2010.01.018

Cite this

@article{c4ea3ab3ed47455f92ca28ec7ca13cfe,

title = "Co-training with relevant random subspaces",

abstract = "We introduce the relevant random subspace Co-training (Rel-RASCO) algorithm which produces relevant random subspaces and then does semi-supervised ensemble learning using those subspaces and unlabeled data. Ensemble learning algorithms may benefit from diversity of classifiers used. However, for high dimensional data choosing subspaces randomly, as in RASCO (Random Subspace Method for Co-training, Wang et al. 2008 [5]) algorithm, may produce diverse but inaccurate classifiers. We produce relevant random subspaces by means of drawing features with probabilities proportional to their relevances measured by the mutual information between features and class labels. We show that Rel-RASCO achieves better accuracy by this relevant and random subspace selection scheme. Experiments on five real and one synthetic datasets show that Rel-RASCO algorithm outperforms both RASCO and Co-training in terms of the accuracy achieved at the end of Co-training.",

keywords = "Co-training, Multiple classifier systems, Random subspace methods, RASCO, Relevant subspace method, Semi-supervised learning",

author = "Yusuf Yaslan and Zehra Cataltepe",

year = "2010",

month = jun,

doi = "10.1016/j.neucom.2010.01.018",

language = "English",

volume = "73",

pages = "1652--1661",

journal = "Neurocomputing",

issn = "0925-2312",

publisher = "Elsevier B.V.",

number = "10-12",

}

TY - JOUR

T1 - Co-training with relevant random subspaces

AU - Yaslan, Yusuf

AU - Cataltepe, Zehra

PY - 2010/6

Y1 - 2010/6

N2 - We introduce the relevant random subspace Co-training (Rel-RASCO) algorithm which produces relevant random subspaces and then does semi-supervised ensemble learning using those subspaces and unlabeled data. Ensemble learning algorithms may benefit from diversity of classifiers used. However, for high dimensional data choosing subspaces randomly, as in RASCO (Random Subspace Method for Co-training, Wang et al. 2008 [5]) algorithm, may produce diverse but inaccurate classifiers. We produce relevant random subspaces by means of drawing features with probabilities proportional to their relevances measured by the mutual information between features and class labels. We show that Rel-RASCO achieves better accuracy by this relevant and random subspace selection scheme. Experiments on five real and one synthetic datasets show that Rel-RASCO algorithm outperforms both RASCO and Co-training in terms of the accuracy achieved at the end of Co-training.

AB - We introduce the relevant random subspace Co-training (Rel-RASCO) algorithm which produces relevant random subspaces and then does semi-supervised ensemble learning using those subspaces and unlabeled data. Ensemble learning algorithms may benefit from diversity of classifiers used. However, for high dimensional data choosing subspaces randomly, as in RASCO (Random Subspace Method for Co-training, Wang et al. 2008 [5]) algorithm, may produce diverse but inaccurate classifiers. We produce relevant random subspaces by means of drawing features with probabilities proportional to their relevances measured by the mutual information between features and class labels. We show that Rel-RASCO achieves better accuracy by this relevant and random subspace selection scheme. Experiments on five real and one synthetic datasets show that Rel-RASCO algorithm outperforms both RASCO and Co-training in terms of the accuracy achieved at the end of Co-training.

KW - Co-training

KW - Multiple classifier systems

KW - Random subspace methods

KW - RASCO

KW - Relevant subspace method

KW - Semi-supervised learning

UR - http://www.scopus.com/inward/record.url?scp=77952554171&partnerID=8YFLogxK

U2 - 10.1016/j.neucom.2010.01.018

DO - 10.1016/j.neucom.2010.01.018

M3 - Article

AN - SCOPUS:77952554171

SN - 0925-2312

VL - 73

SP - 1652

EP - 1661

JO - Neurocomputing

JF - Neurocomputing

IS - 10-12

ER -

Co-training with relevant random subspaces

Abstract

Funding

Keywords

Access to Document

Other files and links

Fingerprint

Cite this