Exploring physiological indicators of pregnancy outcomes using machine learning and longitudinal patient data
Promotor(s) : Geurts, Pierre
Date of defense : 25-Jun-2018/26-Jun-2018 • Permalink :
|Exploring physiological indicators of pregnancy outcomes using machine learning and longitudinal patient data
|Date of defense :
|Committee's member(s) :
Van Droogenbroeck, Marc
|[en] Sleep monitoring
|Engineering, computing & technology > Electrical & electronics engineering
|Université de Liège, Liège, Belgique
|Master en ingénieur civil électricien, à finalité spécialisée en "electrical engineering"
|Master thesis of the Faculté des Sciences appliquées
[en] In the context of pregnancy monitoring, sleep detection and classification based on the electrophysiological and accelerometer signals recorded by the sensor developed by the society Bloomlife are investigated.
The first classification problem consists in differentiating between the sleep and the wakefulness
states while the second one involves the classification between the five main sleep stages, which are the light sleep (N1), the moderate sleep (N2), the deep sleep (N3), the rapid-eye-movements sleep (REM) and the wakefulness state.
The extraction of the main sleep physiological indicators that can be found in the recorded signals
is addressed. Among the different feature categories that were highlighted, the spectral analysis of the heart rate variability, which is highly related to the autonomous neural system activity, seems to be the most informative as much for the sleep detection as for the sleep stages classification. As a huge amount of characteristics has been computed to track the most representative as possible physiological information in the recorded signals, a feature selection is performed.
And finally, the predictions obtained with the three learning algorithms that have been selected,
a K-nearest neighbors, a random forest and a conditional random field algorithms, are evaluated. Among
these three classifiers, the conditional random field algorithm is the most promising algorithm for performing sleep monitoring since it takes into account the cycling structure of the sleep architecture. A bigger data set has however to be collected to improve its overall performances.
Size: 3.68 MB
Format: Adobe PDF
Size: 116.92 kB
Format: Adobe PDF
Cite this master thesis
The University of Liège does not guarantee the scientific quality of these students' works or the accuracy of all the information they contain.