Results - Details
Search command : Author="Τσακαλίδης"
And Author="Παναγιώτης"
Current Record: 23 of 66
|
Identifier |
000425811 |
Title |
Automated dynamical analysis of multivariate physiological data for personalized healthcare |
Alternative Title |
Αυτοματοποιημένη δυναμική ανάλυση πολυπαραγοντικών φυσιολογικών δεδομένων για εξατομικευμένη περίθαλψη υγείας |
Author
|
Ζερβού, Μιχαέλα-Αρετή Σ.
|
Thesis advisor
|
Τσακαλίδης, Παναγιώτης
|
Reviewer
|
Παπαδούλη, Μαρία
Τζαγκαράκης, Γεώργιος
|
Abstract |
Wearable body sensors along with data analytics have been recently employed in the
domain of personalized healthcare, since wearable technology can monitor physical
activity, collect data and deliver real-time feedback to individuals or experts. However,
there are several challenges associated with the monitoring of patient performance using
body sensor data. Specifically, the captured sensor data might be too patient-specific and
inadequate for the differentiation of normal and abnormal behavior, which is formulated
as a classification problem. Thus, a generic approach is required to analyze patient activity
in an unsupervised or semi-supervised manner for identifying patterns and also being
scalable across multiple heterogeneous sensors capturing different data.
Multivariate time series analysis is considered as a difficult problem due to the complexity
of the data types. The main challenges of processing time series data involve the high
dimensionality, the presence of noise and redundancy in the data. The majority of existing
diagnosis techniques employ features extracted from symbolic or frequency-domain
representations of the associated data, whilst ignoring completely the behavior of the
underlying data generating the dynamical system. Moreover, those techniques lack the
capability of concurrently processing multiple dimensions. To address this problem, this
thesis proposes a novel self-tuned architecture for feature extraction, by modeling
directly the inherent dynamics of multidimensional wearable sensor time series data in
higher-dimensional phasespaces, which encode state recurrences considering a specific
task. In essence, our work extends the state-of-the-art multidimensional recurrence
quantification analysis (RQA),that is based on state vectors, to a more generic state
matrix-based architecture. State matrices are considered more appropriate for describing
multi-dimensional signals from a mathematical perspective, enabling them to model the
correlations not only within a signal but also between different signals. It is worth
mentioning that in contrast to the state-of-the-art, the herein proposed framework is
directly applicable to several data structures such as images, hyper-spectral data or video
streams.
Furthermore, this study employs machine learning techniques to design an efficient
feature extraction scheme for the discovery of information-rich patterns that best capture
the underlying data dynamics of multidimensional time series data for dyslexia detection.
Dyslexia is a developmental learning disorder of single word reading accuracy and/or
fluency, with compelling research directed towards understanding the contributions of
the visual system. While dyslexia is not an oculomotor disease, readers with dyslexia have
shown different eye movements than typically developing students during text reading.
Experimental evaluation on real data of eye-tracking for diagnosing dyslexia
demonstrates an improved performance of our method in terms of classification
accuracy, F-score and robustness in the presence of noise when compared against a stateof-the-art unidimensional and vector-based multidimensional RQA.
|
Language |
English |
Subject |
Multivariate timeseries |
|
Αυτόματη δυναμική ανάλυση |
|
Πολυμορφικές χρονοσειρές |
Issue date |
2019-07-26 |
Collection
|
School/Department--School of Sciences and Engineering--Department of Computer Science--Post-graduate theses
|
|
Type of Work--Post-graduate theses
|
Permanent Link |
https://elocus.lib.uoc.gr//dlib/5/0/8/metadata-dlib-1573576328-869533-14050.tkl
|
Views |
505 |