A regression subset-selection strategy for fat-structure data

Christian Gatu; Marko Sysi-Aho; Matej Orešič

doi:10.1007/978-3-7908-2084-3_29

A regression subset-selection strategy for fat-structure data

Christian Gatu^*, Marko Sysi-Aho, Matej Orešič

^*Corresponding author for this work

VTT Technical Research Centre of Finland

Research output: Chapter in Book/Report/Conference proceeding › Conference article in proceedings › Scientific › peer-review

Abstract

A strategy is proposed for finding the most significant linear regression submodel for fat-structure data, that is when the number of variables n exceeds the number of available observations m. The method consists of two stages. First, a heuristic is employed to preselect a number of variables n_S such that n_S ≤m. The second stage performs an exhaustive search on the reduced list of variables. It employs a regression tree structure that generates all possible subset models. Non-optimal subtrees are pruned using a branch-and-bound device. Cross validation experiments on a real biomedical dataset are presented and analyzed.

Original language	English
Title of host publication	COMPSTAT 2008
Subtitle of host publication	Proceedings in Computational Statistics
Editors	Paula Brito
Place of Publication	Heidelberg
Publisher	Physica Verlag
Pages	349-358
ISBN (Electronic)	978-3-7908-2084-3
ISBN (Print)	978-3-7908-2083-6
DOIs	https://doi.org/10.1007/978-3-7908-2084-3_29
Publication status	Published - 2008
MoE publication type	A4 Article in a conference publication
Event	18th Symposium on Computational Statistics - Porto, Portugal Duration: 24 Aug 2008 → 29 Aug 2008

Conference

Conference	18th Symposium on Computational Statistics
Country/Territory	Portugal
City	Porto
Period	24/08/08 → 29/08/08

Keywords

regression tree
branch-and-bound
model selection
fat-structure data

Access to Document

10.1007/978-3-7908-2084-3_29

Cite this

@inproceedings{611a113de2a94678882c244f41bf8334,

title = "A regression subset-selection strategy for fat-structure data",

abstract = "A strategy is proposed for finding the most significant linear regression submodel for fat-structure data, that is when the number of variables n exceeds the number of available observations m. The method consists of two stages. First, a heuristic is employed to preselect a number of variables nS such that nS ≤m. The second stage performs an exhaustive search on the reduced list of variables. It employs a regression tree structure that generates all possible subset models. Non-optimal subtrees are pruned using a branch-and-bound device. Cross validation experiments on a real biomedical dataset are presented and analyzed.",

keywords = "regression tree, branch-and-bound, model selection, fat-structure data",

author = "Christian Gatu and Marko Sysi-Aho and Matej Ore{\v s}i{\v c}",

year = "2008",

doi = "10.1007/978-3-7908-2084-3\_29",

language = "English",

isbn = "978-3-7908-2083-6",

pages = "349--358",

editor = "Paula Brito",

booktitle = "COMPSTAT 2008",

publisher = "Physica Verlag",

address = "Germany",

note = "18th Symposium on Computational Statistics ; Conference date: 24-08-2008 Through 29-08-2008",

}

TY - GEN

T1 - A regression subset-selection strategy for fat-structure data

AU - Gatu, Christian

AU - Sysi-Aho, Marko

AU - Orešič, Matej

PY - 2008

Y1 - 2008

N2 - A strategy is proposed for finding the most significant linear regression submodel for fat-structure data, that is when the number of variables n exceeds the number of available observations m. The method consists of two stages. First, a heuristic is employed to preselect a number of variables nS such that nS ≤m. The second stage performs an exhaustive search on the reduced list of variables. It employs a regression tree structure that generates all possible subset models. Non-optimal subtrees are pruned using a branch-and-bound device. Cross validation experiments on a real biomedical dataset are presented and analyzed.

AB - A strategy is proposed for finding the most significant linear regression submodel for fat-structure data, that is when the number of variables n exceeds the number of available observations m. The method consists of two stages. First, a heuristic is employed to preselect a number of variables nS such that nS ≤m. The second stage performs an exhaustive search on the reduced list of variables. It employs a regression tree structure that generates all possible subset models. Non-optimal subtrees are pruned using a branch-and-bound device. Cross validation experiments on a real biomedical dataset are presented and analyzed.

KW - regression tree

KW - branch-and-bound

KW - model selection

KW - fat-structure data

U2 - 10.1007/978-3-7908-2084-3_29

DO - 10.1007/978-3-7908-2084-3_29

M3 - Conference article in proceedings

SN - 978-3-7908-2083-6

SP - 349

EP - 358

BT - COMPSTAT 2008

A2 - Brito, Paula

PB - Physica Verlag

CY - Heidelberg

T2 - 18th Symposium on Computational Statistics

Y2 - 24 August 2008 through 29 August 2008

ER -

A regression subset-selection strategy for fat-structure data

Abstract

Conference

Keywords

Access to Document

Fingerprint

Cite this