Reducing the intrusion of user-trained activity recognition systems

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

Abstract

Many supervised activity recognition systems require a fully labelled time-series for accurate classification. However, gathering these labels is a difficult and often unrealistic task, especially over long-time frames or outside of laboratory conditions. A potential solution is through diary studies, allowing for a user-trained activity recognition system. Requests will be presented on the user's smart device and while this approach will be significantly less intrusive than current methods, frequent or inappropriately timed requests could reduce user acceptance. This paper proposes to further reduce user intrusion by making a prediction about the next user request and analyzing the classifiers confidence in this prediction. Two methods are presented, and with careful selection of the confidence threshold, they resulted in up to a 35% reduction in user requests with a minimal reduction in accuracy.
LanguageEnglish
Title of host publication2018 29th Irish Signals and Systems Conference (ISSC)
Number of pages4
ISBN (Electronic)978-1-5386-6046-1
DOIs
Publication statusPublished - 28 Dec 2018
Event29th Irish signals and Systems Conference 2018 - Queens University, Belfast, Northern Ireland
Duration: 21 Jun 201822 Jun 2018
http://www.issc.ie/site/view/7/

Conference

Conference29th Irish signals and Systems Conference 2018
CountryNorthern Ireland
CityBelfast
Period21/06/1822/06/18
Internet address

Fingerprint

Labels
Time series
Classifiers

Keywords

  • activity recognition
  • Random Forest
  • ECOC-SVM
  • experience sampling

Cite this

@inproceedings{f897b228149a4e6ba42de9e1f2951640,
title = "Reducing the intrusion of user-trained activity recognition systems",
abstract = "Many supervised activity recognition systems require a fully labelled time-series for accurate classification. However, gathering these labels is a difficult and often unrealistic task, especially over long-time frames or outside of laboratory conditions. A potential solution is through diary studies, allowing for a user-trained activity recognition system. Requests will be presented on the user's smart device and while this approach will be significantly less intrusive than current methods, frequent or inappropriately timed requests could reduce user acceptance. This paper proposes to further reduce user intrusion by making a prediction about the next user request and analyzing the classifiers confidence in this prediction. Two methods are presented, and with careful selection of the confidence threshold, they resulted in up to a 35{\%} reduction in user requests with a minimal reduction in accuracy.",
keywords = "activity recognition, Random Forest, ECOC-SVM, experience sampling",
author = "William Duffy and Kevin Curran and Daniel Kelly and Tom Lunney",
year = "2018",
month = "12",
day = "28",
doi = "10.1109/ISSC.2018.8585343",
language = "English",
isbn = "978-1-5386-6047-8",
booktitle = "2018 29th Irish Signals and Systems Conference (ISSC)",

}

Duffy, W, Curran, K, Kelly, D & Lunney, T 2018, Reducing the intrusion of user-trained activity recognition systems. in 2018 29th Irish Signals and Systems Conference (ISSC). 29th Irish signals and Systems Conference 2018, Belfast, Northern Ireland, 21/06/18. https://doi.org/10.1109/ISSC.2018.8585343

Reducing the intrusion of user-trained activity recognition systems. / Duffy, William; Curran, Kevin; Kelly, Daniel; Lunney, Tom.

2018 29th Irish Signals and Systems Conference (ISSC). 2018.

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

TY - GEN

T1 - Reducing the intrusion of user-trained activity recognition systems

AU - Duffy, William

AU - Curran, Kevin

AU - Kelly, Daniel

AU - Lunney, Tom

PY - 2018/12/28

Y1 - 2018/12/28

N2 - Many supervised activity recognition systems require a fully labelled time-series for accurate classification. However, gathering these labels is a difficult and often unrealistic task, especially over long-time frames or outside of laboratory conditions. A potential solution is through diary studies, allowing for a user-trained activity recognition system. Requests will be presented on the user's smart device and while this approach will be significantly less intrusive than current methods, frequent or inappropriately timed requests could reduce user acceptance. This paper proposes to further reduce user intrusion by making a prediction about the next user request and analyzing the classifiers confidence in this prediction. Two methods are presented, and with careful selection of the confidence threshold, they resulted in up to a 35% reduction in user requests with a minimal reduction in accuracy.

AB - Many supervised activity recognition systems require a fully labelled time-series for accurate classification. However, gathering these labels is a difficult and often unrealistic task, especially over long-time frames or outside of laboratory conditions. A potential solution is through diary studies, allowing for a user-trained activity recognition system. Requests will be presented on the user's smart device and while this approach will be significantly less intrusive than current methods, frequent or inappropriately timed requests could reduce user acceptance. This paper proposes to further reduce user intrusion by making a prediction about the next user request and analyzing the classifiers confidence in this prediction. Two methods are presented, and with careful selection of the confidence threshold, they resulted in up to a 35% reduction in user requests with a minimal reduction in accuracy.

KW - activity recognition

KW - Random Forest

KW - ECOC-SVM

KW - experience sampling

U2 - 10.1109/ISSC.2018.8585343

DO - 10.1109/ISSC.2018.8585343

M3 - Conference contribution

SN - 978-1-5386-6047-8

BT - 2018 29th Irish Signals and Systems Conference (ISSC)

ER -