An Efficient Feature Selection Method for Activity Classification

Shumei Zhang, Paul McCullagh, Vic Callaghan

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

6 Citations (Scopus)

Abstract

Abstract—Feature selection is a key step for activityclassification applications. Feature selection selects the mostrelevant features and considers how to use each of theselected features in the most suitable format. This paperproposes an efficient feature selection method that organizesmultiple subsets of features in a multilayer, rather thanutilizing all selected features together as one large feature set.The proposed method was evaluated by 13 subjects (agedfrom 23 to 50) in a lab environment. The experimentalresults illustrate that the large number of features (3 vs. 7features) are not associated with high classification accuracyusing a single Support Vector Machine (SVM) model (61.3%vs. 44.7%). However, the accuracy was improvedsignificantly (83.1% vs. 44.7%), when the selected 7 featureswere organized as 3 subsets and used to classify 10 postures(9 motionless with 1 motion) in 3 layers via a hierarchicalalgorithm, which combined a rule-based algorithm with 3independent SVM models.
LanguageEnglish
Title of host publicationUnknown Host Publication
Pages16-22
Number of pages7
DOIs
Publication statusPublished - 30 Jun 2014
Event2014 International Conference on Intelligent Environments - Shanghai, China
Duration: 30 Jun 2014 → …

Conference

Conference2014 International Conference on Intelligent Environments
Period30/06/14 → …

Fingerprint

Feature extraction
Support vector machines
Multilayers

Cite this

Zhang, S., McCullagh, P., & Callaghan, V. (2014). An Efficient Feature Selection Method for Activity Classification. In Unknown Host Publication (pp. 16-22) https://doi.org/10.1109/IE.2014.10
Zhang, Shumei ; McCullagh, Paul ; Callaghan, Vic. / An Efficient Feature Selection Method for Activity Classification. Unknown Host Publication. 2014. pp. 16-22
@inproceedings{56e435e70823462a93850b5b20e748f0,
title = "An Efficient Feature Selection Method for Activity Classification",
abstract = "Abstract—Feature selection is a key step for activityclassification applications. Feature selection selects the mostrelevant features and considers how to use each of theselected features in the most suitable format. This paperproposes an efficient feature selection method that organizesmultiple subsets of features in a multilayer, rather thanutilizing all selected features together as one large feature set.The proposed method was evaluated by 13 subjects (agedfrom 23 to 50) in a lab environment. The experimentalresults illustrate that the large number of features (3 vs. 7features) are not associated with high classification accuracyusing a single Support Vector Machine (SVM) model (61.3{\%}vs. 44.7{\%}). However, the accuracy was improvedsignificantly (83.1{\%} vs. 44.7{\%}), when the selected 7 featureswere organized as 3 subsets and used to classify 10 postures(9 motionless with 1 motion) in 3 layers via a hierarchicalalgorithm, which combined a rule-based algorithm with 3independent SVM models.",
author = "Shumei Zhang and Paul McCullagh and Vic Callaghan",
year = "2014",
month = "6",
day = "30",
doi = "10.1109/IE.2014.10",
language = "English",
pages = "16--22",
booktitle = "Unknown Host Publication",

}

Zhang, S, McCullagh, P & Callaghan, V 2014, An Efficient Feature Selection Method for Activity Classification. in Unknown Host Publication. pp. 16-22, 2014 International Conference on Intelligent Environments, 30/06/14. https://doi.org/10.1109/IE.2014.10

An Efficient Feature Selection Method for Activity Classification. / Zhang, Shumei; McCullagh, Paul; Callaghan, Vic.

Unknown Host Publication. 2014. p. 16-22.

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

TY - GEN

T1 - An Efficient Feature Selection Method for Activity Classification

AU - Zhang, Shumei

AU - McCullagh, Paul

AU - Callaghan, Vic

PY - 2014/6/30

Y1 - 2014/6/30

N2 - Abstract—Feature selection is a key step for activityclassification applications. Feature selection selects the mostrelevant features and considers how to use each of theselected features in the most suitable format. This paperproposes an efficient feature selection method that organizesmultiple subsets of features in a multilayer, rather thanutilizing all selected features together as one large feature set.The proposed method was evaluated by 13 subjects (agedfrom 23 to 50) in a lab environment. The experimentalresults illustrate that the large number of features (3 vs. 7features) are not associated with high classification accuracyusing a single Support Vector Machine (SVM) model (61.3%vs. 44.7%). However, the accuracy was improvedsignificantly (83.1% vs. 44.7%), when the selected 7 featureswere organized as 3 subsets and used to classify 10 postures(9 motionless with 1 motion) in 3 layers via a hierarchicalalgorithm, which combined a rule-based algorithm with 3independent SVM models.

AB - Abstract—Feature selection is a key step for activityclassification applications. Feature selection selects the mostrelevant features and considers how to use each of theselected features in the most suitable format. This paperproposes an efficient feature selection method that organizesmultiple subsets of features in a multilayer, rather thanutilizing all selected features together as one large feature set.The proposed method was evaluated by 13 subjects (agedfrom 23 to 50) in a lab environment. The experimentalresults illustrate that the large number of features (3 vs. 7features) are not associated with high classification accuracyusing a single Support Vector Machine (SVM) model (61.3%vs. 44.7%). However, the accuracy was improvedsignificantly (83.1% vs. 44.7%), when the selected 7 featureswere organized as 3 subsets and used to classify 10 postures(9 motionless with 1 motion) in 3 layers via a hierarchicalalgorithm, which combined a rule-based algorithm with 3independent SVM models.

U2 - 10.1109/IE.2014.10

DO - 10.1109/IE.2014.10

M3 - Conference contribution

SP - 16

EP - 22

BT - Unknown Host Publication

ER -