gms | German Medical Science

63. Jahrestagung der Deutschen Gesellschaft für Medizinische Informatik, Biometrie und Epidemiologie e. V. (GMDS)

Deutsche Gesellschaft für Medizinische Informatik, Biometrie und Epidemiologie

02. - 06.09.2018, Osnabrück

Analysis of time series data from wearable accelerometer devices with wavelet transform and neural networks

Meeting Abstract

Search Medline for

  • Konrad Neumann - Institute of Biometry and Clinical Epidemiology, Charité-Universitätsmedizin Berlin, Berlin Institute of Health (BIH), Berlin, Deutschland
  • Torsten Rackoll - Center for Stroke Research Berlin, Charité - Universitätsmedizin Berlin, Berlin, Deutschland
  • Ulrike Grittner - Institute of Medical Biometrics and Clinical Epidemiology, Charité - Universitätsmedizin Berlin, Berlin, Deutschland

Deutsche Gesellschaft für Medizinische Informatik, Biometrie und Epidemiologie. 63. Jahrestagung der Deutschen Gesellschaft für Medizinische Informatik, Biometrie und Epidemiologie e.V. (GMDS). Osnabrück, 02.-06.09.2018. Düsseldorf: German Medical Science GMS Publishing House; 2018. DocAbstr. 219

doi: 10.3205/18gmds107, urn:nbn:de:0183-18gmds1072

Published: August 27, 2018

© 2018 Neumann et al.
This is an Open Access article distributed under the terms of the Creative Commons Attribution 4.0 License. See license information at



Wearable devices, such as accelerometers, have become widely used tools to measure physical activity. Time series of acceleration magnitude are argued to reflect the circadian rhythm. Sleep-wake disturbances are often a prodromal feature preceding the onset of Alzheimer’s disease (AD) [1].

Different regression methods such as the function-on-scalar approach [2], functional mixed models [3] or two stage models [4] were proposed to analyze time-series from wearable devices. However, all regression methods have in common that they are only able to detect associations between covariates and activity levels but fail to identify activity patterns that could indicate a disturbed circadian rhythm.

Since time series from accelerometers are rather noisy, most regression models use a smooth of the data. We propose smoothing with wavelets [5] and compared this novel approach with different other smoothing techniques. Moreover, coefficients from wavelet transform are apt to train a neural network for pattern recognition. Wavelet coefficients are particularly suitable for pattern recognition in time series since they are localized weighted averages at various scales [5].

We demonstrate the different analysis approaches with simulated data and with real data from a study with 62 young, 66 old and 31 mildly cognitive impaired participants who wore an accelerator sensor (ActiGraph GT3X ™) for seven consecutive days. In particular, the simulation study demonstrated that neural networks are able to recognize various patterns that could indicate a disturbed sleep-wake rhythm, a possible precursor of AD.

The authors declare that they have no competing interests.

The authors declare that an ethics committee vote is not required.


Naismith SL, et al. Sleep disturbance in mild cognitive impairment: Differential effects of current and remitted depression. Acta Neuropsychiatr. 2011;23:167–172.
Goldsmith J, Liu X, Jacobson JS, Rundle A. New Insights into Activity Patterns in Children, Found Using Functional Data Analyses. Med Sci Sports Exerc. 2016;48:1723–1729.
Xiao L, et al. Quantifying the lifetime circadian rhythm of physical activity: A covariate-dependent functional approach. Biostatistics. 2015;16:352–367.
Bai J, Sun Y, Schrack JA, Crainiceanu CM, Wang MC. A two-stage model for wearable device data. Biometrics. 2018 Jun;74(2):744-752. DOI: 10.1111/biom.12781 External link
Percival DB, Walden AT. Wavelet Methods for Time Series Analysis. Cambridge: Cambridge University Press.