Authors:
Monika Simjanoska
1
;
Martin Gjoreski
2
;
Ana Madevska Bogdanova
1
;
Bojana Koteska
1
;
Matjaž Gams
2
and
Jurij Tasič
3
Affiliations:
1
Ss. Cyril and Methodius University, Macedonia, The Former Yugoslav Republic of
;
2
Jozef Stefan Institute and International Postgraduate School, Slovenia
;
3
University of Ljubljana, Slovenia
Keyword(s):
Blood Pressure, ECG-derived, Complexity Analysis, Machine Learning, Stacking, Classification.
Related
Ontology
Subjects/Areas/Topics:
Artificial Intelligence
;
Biomedical Engineering
;
Biomedical Signal Processing
;
Data Mining
;
Databases and Information Systems Integration
;
Devices
;
Enterprise Information Systems
;
Health Information Systems
;
Human-Computer Interaction
;
Pattern Recognition and Machine Learning
;
Physiological Computing Systems
;
Physiological Modeling
;
Sensor Networks
;
Signal Processing
;
Soft Computing
;
Wearable Sensors and Systems
Abstract:
The recent advancement on wearable physiological sensors supports the development of real-time diagnosis
in preventive medicine that demands various signal processing techniques to enable the extraction of the vital
signs (e.g., blood pressure). Blood pressure estimation from physiological sensors data is challenging task
that usually is solved by a combination of multiple signals. In this paper we present a novel complexity
analysis-based machine-learning perspective on the problem of blood pressure class estimation only from
ECG signals. We show that high classification accuracy of 96.68% can be achieved by extracting information
via complexity analysis on the ECG signal followed by applying a stack of machine-learning classifiers. In
addition, the proposed stacking approach is compared to a traditional machine-learning approaches and feature
analysis is performed to determine the influence of the different features on the classification accuracy. The
experimental data was gathered
by daily monitoring of 20 subjects with two different ECG sensors.
(More)