elib
DLR-Header
DLR-Logo -> http://www.dlr.de
DLR Portal Home | Imprint | Privacy Policy | Contact | Deutsch
Fontsize: [-] Text [+]

Classification of Not-Worn Phases and Inactivity Phases in Accelerometry Data using Machine Learning

Mayat, Nils (2021) Classification of Not-Worn Phases and Inactivity Phases in Accelerometry Data using Machine Learning. Bachelor's, University of Applied Sciences, RheinAhrCampus, Koblenz, Germany.

[img] PDF - Only accessible within DLR
1MB

Abstract

During the study “Bone loss and recovery after space exposure” (EDOS-2) conducted by the Institute of Aerospace Medicine at the German Aerospace Center (Deutsches Zentrum für Luft- und Raumfahrt, DLR) accelerometry data of 10 Russian cosmonauts was sampled before and after their stay on the International Space Station (ISS). This was done in order to examine the changes in their daily physical activity (PA) during the recovery phase. The tri axial AM20 3D USB accelerometer and activity meter by Traxmeet was used to sample the accelerometry data. The accelerometer shuts off during phases of low acceleration which causes breaks in the data. These breaks can be cause by inactivity of the subject (IAB) or by not wearing the device (NWB). In order to examine the PA of the cosmonauts these breaks need to be classified. The amount of 407072 breaks that needed to be classified, lead to machine learning as first choice for the classification task. A decision tree, a decision tree with linear discriminant analysis (LDA), a decision tree with principal component analysis (PCA) and a LDA classifier were trained with data of 33 subjects who followed a 20 minute protocol with everyday life activities. Each model was trained with the length parameters and without them resulting in 8 different models. During the testing phase 11 subject carried out a randomised protocol which was then used to evaluate the models. Among the models with length parameters the decision tree in combination with a LDA performed best with a F0.5-Score of 0.815 and an AUC of 0.88. The LDA classifier had the best results among the models without the length parameters with a F0.5-Score of 0.725 and an AUC of 0.76. For the classification of the cosmonaut data an additional threshold was set. All breaks which lasted longer than three hours were classified as NWBs. The LDA decision tree classified 356374 (87.55%) breaks as IABs and 50698 (12.45%) as NWBs. The LDA classifier classified 256666 (63.05%) breaks as IABs and 150406 (36.95%) breaks as NWBs.

Item URL in elib:https://elib.dlr.de/143170/
Document Type:Thesis (Bachelor's)
Title:Classification of Not-Worn Phases and Inactivity Phases in Accelerometry Data using Machine Learning
Authors:
AuthorsInstitution or Email of AuthorsAuthor's ORCID iD
Mayat, NilsHochschule KoblenzUNSPECIFIED
Date:2021
Refereed publication:Yes
Open Access:No
Gold Open Access:No
In SCOPUS:No
In ISI Web of Science:No
Number of Pages:40
Status:Published
Keywords:Machine Learning, Actimetry
Institution:University of Applied Sciences, RheinAhrCampus, Koblenz, Germany
Department:Hochschule Koblenz
HGF - Research field:Aeronautics, Space and Transport
HGF - Program:Space
HGF - Program Themes:Research under Space Conditions
DLR - Research area:Raumfahrt
DLR - Program:R FR - Research under Space Conditions
DLR - Research theme (Project):R - Muscle Mechanics and Metabolism
Location: Köln-Porz
Institutes and Institutions:Institute of Aerospace Medicine > Muscle and Bone Metabolism
Deposited By: Schönenberg, Sandra
Deposited On:05 Aug 2021 15:03
Last Modified:10 Aug 2021 13:48

Repository Staff Only: item control page

Browse
Search
Help & Contact
Information
electronic library is running on EPrints 3.3.12
Copyright © 2008-2017 German Aerospace Center (DLR). All rights reserved.