One-Class Subject Authentication Using Feature Extraction by Grammatical Evolution on Accelerometer Data

Stefano Mauceri, James Sweeney, James McDermott

Research output: Chapter in Book/Report/Conference proceedingChapterpeer-review

Abstract

In this study Grammatical Evolution (GE) is used to extract features from accelerometer time series in order to increase the performance of a Kernel Density Estimation (KDE) classifier. Time series are collected through nine wrist-worn accelerometers assigned to as many subjects. The goal is to distinguish each subject from all the others in a one-class classification framework. GE-evolved solutions, referred to as feature extractors, are thoroughly analyzed. Each solution is a function able to target a specific sub-sequence of a time series and reduce it to a single scalar. In this way a long time series can be summarized to an arbitrary number of features. Results show that the proposed evolutionary algorithm outperforms two strong baselines.

Original languageEnglish
Title of host publicationStudies in Computational Intelligence
PublisherSpringer Science and Business Media Deutschland GmbH
Pages393-407
Number of pages15
DOIs
Publication statusPublished - 2021

Publication series

NameStudies in Computational Intelligence
Volume906
ISSN (Print)1860-949X
ISSN (Electronic)1860-9503

Fingerprint

Dive into the research topics of 'One-Class Subject Authentication Using Feature Extraction by Grammatical Evolution on Accelerometer Data'. Together they form a unique fingerprint.

Cite this