On The Stability of Feature Selection in the Presence of Feature Correlations

Konstantinos Sechidis; Konstantinos Papangelou; Sarah Nogueira; James Weatherall; Gavin Brown

On The Stability of Feature Selection in the Presence of Feature Correlations

Konstantinos Sechidis, Konstantinos Papangelou, Sarah Nogueira, James Weatherall, Gavin Brown

Machine Learning and Robotics

Research output: Chapter in Book/Conference proceeding › Conference contribution › peer-review

Abstract

Feature selection is central to modern data science. The `stability' of a feature selection algorithm refers to the sensitivity of its choices to small changes in training data. This is, in effect, the robustness of the chosen features. This paper considers the estimation of stability when we expect strong pairwise correlations, otherwise known as feature redundancy. We demonstrate that existing measures are inappropriate here, as they systematically underestimate the true stability, giving an overly pessimistic view of a feature set. We propose a new statistical measure which overcomes this issue, and generalises previous work.

Original language	English
Title of host publication	European Conference on Machine Learning
Publication status	Accepted/In press - 7 Jun 2019
Event	European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases - Wurzburg, Germany Duration: 16 Sept 2019 → 20 Sept 2019

Conference

Conference	European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases
Abbreviated title	ECMLPKDD
Country/Territory	Germany
City	Wurzburg
Period	16/09/19 → 20/09/19

Embargoed Document

ECML2019_stability (5)
Accepted author manuscript, 365 KB
Embargo ends: 1/01/29

Cite this

@inproceedings{0e64c9098e604d82aa27524114491f88,

title = "On The Stability of Feature Selection in the Presence of Feature Correlations",

abstract = "Feature selection is central to modern data science. The `stability' of a feature selection algorithm refers to the sensitivity of its choices to small changes in training data. This is, in effect, the robustness of the chosen features. This paper considers the estimation of stability when we expect strong pairwise correlations, otherwise known as feature redundancy. We demonstrate that existing measures are inappropriate here, as they systematically underestimate the true stability, giving an overly pessimistic view of a feature set. We propose a new statistical measure which overcomes this issue, and generalises previous work.",

author = "Konstantinos Sechidis and Konstantinos Papangelou and Sarah Nogueira and James Weatherall and Gavin Brown",

year = "2019",

month = jun,

day = "7",

language = "English",

booktitle = "European Conference on Machine Learning",

note = "European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases, ECMLPKDD ; Conference date: 16-09-2019 Through 20-09-2019",

}

TY - GEN

T1 - On The Stability of Feature Selection in the Presence of Feature Correlations

AU - Sechidis, Konstantinos

AU - Papangelou, Konstantinos

AU - Nogueira, Sarah

AU - Weatherall, James

AU - Brown, Gavin

PY - 2019/6/7

Y1 - 2019/6/7

N2 - Feature selection is central to modern data science. The `stability' of a feature selection algorithm refers to the sensitivity of its choices to small changes in training data. This is, in effect, the robustness of the chosen features. This paper considers the estimation of stability when we expect strong pairwise correlations, otherwise known as feature redundancy. We demonstrate that existing measures are inappropriate here, as they systematically underestimate the true stability, giving an overly pessimistic view of a feature set. We propose a new statistical measure which overcomes this issue, and generalises previous work.

AB - Feature selection is central to modern data science. The `stability' of a feature selection algorithm refers to the sensitivity of its choices to small changes in training data. This is, in effect, the robustness of the chosen features. This paper considers the estimation of stability when we expect strong pairwise correlations, otherwise known as feature redundancy. We demonstrate that existing measures are inappropriate here, as they systematically underestimate the true stability, giving an overly pessimistic view of a feature set. We propose a new statistical measure which overcomes this issue, and generalises previous work.

M3 - Conference contribution

BT - European Conference on Machine Learning

T2 - European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases

Y2 - 16 September 2019 through 20 September 2019

ER -

On The Stability of Feature Selection in the Presence of Feature Correlations

Abstract

Conference

Embargoed Document

Fingerprint

Cite this