A Stability Evaluation of Feature Ranking Algorithms on Breast Cancer Data Analysis

Shaode Yu, Bingjie Li, Boji Liu, Mingxue Jin, Junjie Wu, Hang Yu

Research output: Chapter in Book/Report/Conference proceedingConference contribution

1 Scopus citations

Abstract

Stability of feature preference is a most vital yet rarely explored characteristics of feature ranking algorithms. In this study, 23 feature rankers are evaluated on 4 breast cancer datasets (BCDR-F03, WDBC, GSE10810 and GSE15852) using an advanced stability estimator (S), and 3 rankers are identified showing good stability (S ≥ 0.55) consistently on the four datasets. It suggests that data sufficiency is crucial for the construction of feature importance measure, since more rankers are stable on medical imaging datsets (BCDR-F03 and WDBC) than on gene expression datasets (GSE10810 and GSE15852), and high-dimensional small-sample-size datasets are big challenges of stability estimation. In our future work, more attention should be paid to the topics of developing stable feature ranking algorithms and stability estimators to well tackle different sizes of medical datasets.

Original languageEnglish (US)
Title of host publicationProceedings of CECNet 2022 - 12th International Conference on Electronics, Communications and Networks, CECNet 2022
EditorsAntonio J. Tallon-Ballesteros
PublisherIOS Press BV
Pages606-613
Number of pages8
ISBN (Electronic)9781643683683
DOIs
StatePublished - Dec 13 2022
Event12th International Conference on Electronics, Communications and Networks, CECNet 2022 - Virtual, Online
Duration: Nov 4 2022Nov 7 2022

Publication series

NameFrontiers in Artificial Intelligence and Applications
Volume363
ISSN (Print)0922-6389
ISSN (Electronic)1879-8314

Conference

Conference12th International Conference on Electronics, Communications and Networks, CECNet 2022
CityVirtual, Online
Period11/4/2211/7/22

Keywords

  • Stability
  • breast cancer
  • data analysis
  • feature ranking
  • matFR

ASJC Scopus subject areas

  • Artificial Intelligence

Fingerprint

Dive into the research topics of 'A Stability Evaluation of Feature Ranking Algorithms on Breast Cancer Data Analysis'. Together they form a unique fingerprint.

Cite this