![Bulletin of Electrical Engineering and Informatics
Vol. 8, No. 1, March 2019, pp. 261~268
ISSN: 2302-9285, DOI: 10.11591/eei.v8i1.1400 261
Journal homepage: http://beei.org/index.php/EEI
A general framework for improving electrocardiography
monitoring system with machine learning
A. M. Khairuddin, KU Azir K. N. F., P. Eh Kan
Embedded, Networks and Advanced Computing Research Cluster (ENAC),
School of Computer and Communication Engineering, Universiti Malaysia Perlis,
Kampus Pauh, 02600 Ulu Pauh, Perlis, Malaysia
Article Info ABSTRACT
Article history:
Received Oct 28, 2018
Revised Nov 30, 2018
Accepted Dec 26, 2018
As one of the most important health monitoring systems, electrocardiography
(ECG) is used to obtain information about the structure and functions of the
human heart for detecting and preventing cardiovascular disease. Given its
important role, it is vital that the ECG monitoring system provides relevant
and accurate information about the heart. Over the years, numerous attempts
were made to design and develop more effective ECG monitoring system.
Nonetheless, the literature reveals not only several limitations in
conventional ECG monitoring system but also emphasizes on the need to
adopt new technology such as machine learning to improve the monitoring
system as well as its medical applications. This paper reviews previous
works on machine learning to explain its key features, capabilities as well as
presents a general framework for improving ECG monitoring system.
Keywords:
Electrocardiography
Framework
Limitations
Machine learning
Medical applications Copyright © 2019 Institute of Advanced Engineering and Science.
All rights reserved.
Corresponding Author:
A. M. Khairuddin,
Embedded, Networks and Advanced Computing Research Cluster (ENAC),
School of Computer and Communication Engineering, Universiti Malaysia Perlis,
Kampus Pauh, 02600 Ulu Pauh, Perlis, Malaysia.
Email: adamkhairuddin@studentmail.unimap.edu.my
1. INTRODUCTION
Cardiovascular disease has been established as a major threat to human lives and public health. The
statistics concerning cardiovascular disease are alarming. It has been reported that the disease kills as many
as six million people every year. Most of these deaths resulted from heart attacks. About half of all heart
attacks are fatal and a third of the deaths occur before reaching hospitals. Nonetheless, heart disease is
potentially avoidable and preventable. The disease has been primarily and extensively diagnosed by using the
electrocardiograph (ECG) monitoring system. With its capabilities to detect, monitor and provide useful
information about the heart, the ECG monitoring system has been able to reduce the number of premature
deaths, improved healthcare services and quality of life as well as lowered healthcare costs [1-7].
Electrocardiograph monitoring system has become the standard procedure for diagnosing the heart
since the 1940s. However, the earliest ECG monitoring system was basically developed with electrically
primitive analog electronics. Since the first electrocardiography device was introduced, there has been
numerous attempts to introduce better systems. Over the years, the ECG system has continued to evolve as
well as improve. The advances in new technologies have allowed researchers to design and develop more
effective ECG monitoring system. For instance, contemporary ECG monitoring system adopt analog-to-
digital converter to convert the signal convey by the system to digital signal that can be manipulated with
digital electronics.
Throughout the years, improvement in ECG monitoring system and devices was driven by the
advances, capabilities and potentials of new technologies such as pervasive computing, wireless mesh
networks, 3D printing, machine learning, cloud computing and big data analytics. These technologies have](https://image.slidesharecdn.com/341400-210625061029/85/A-general-framework-for-improving-electrocardiography-monitoring-system-with-machine-learning-1-320.jpg)
![ ISSN: 2302-9285
Bulletin of Electr Eng and Inf, Vol. 8, No. 1, March 2019 : 261– 268
262
shape and reshape important aspects of health care and learning. For instance, in health care, machine
learning promises a dramatic shift in diagnosis and treatment of heart disease. Rather than wait for
troublesome symptoms of the health disease to appear, through machine learning doctors now can
continuously monitor the vital signs of patients from a much earlier stage. Despite the capabilities and
potentials of machine learning, the literature reveals not only limited research but also not much information
on the adoption of machine learning in contemporary ECG monitoring system. This paper attempts to
provide some insights into the adoption of machine learning in improving ECG monitoring system. The first
part of the paper reviews previous works on ECG monitoring system to highlight its limitations. Following
this, the second part examines the literature on machine learning as well as presents a general framework for
improving ECG monitoring system.
2. LIMITATIONS OF ECG MONITORING SYSTEM
Electrocardiography (ECG) monitoring system is used to obtain information about the electrical
activity of the human heart and to detect cardiovascular disease. The literature indicates that over the years,
different types of ECG monitoring system have been developed by using new technologies. The review of
the past studies on ECG monitoring system however reveals a great deal of variance in terms of their
capabilities and performance. More specifically, the review of the literature reveals several limitations. The
limitations as identified in previous research are presented below.
2.1. Lack of learning capabilities
Despite the new technologies and efforts to develop intelligent ECG monitoring system, the review
of the literature and past studies indicates that existing system has limited capabilities. For instance, existing
ECG monitoring system appears to lack the ability to learn without being programmed. The learning ability
of the monitoring system can be improved by making it more responsive and interactive through machine
learning. However, the review of past research on ECG indicates very few studies have attempted to adopt
this technology. The adoption of machine learning in ECG monitoring system is useful because it can not
only help the monitoring system to learn but also be able to provide better detection and prediction of
cardiac anomalies [8].
2.2. Limited real time monitoring
Real time monitoring capability of existing ECG monitoring system seems to be not only limited but
also not emphasized. Intelligent ECG monitoring system can not only monitor the ECG signal of patients in
real time but also detect their different physical activity as well. Equipped with machine learning, the
monitoring system can automatically collect information on the physiological as well as physical behavior of
patients. Machine learning helps to process, analyze and wirelessly send the information directly to the
cardiologist for further evaluation. This will save the time needed to visit and examine patients’ vital signs on
a regular basis. Furthermore, the monitoring system can be used to gather medical information, monitor
patients in remote places, high altitude as well as underwater [9-16].
2.3. Inadequate focus on vital signs
The review of the previous studies on ECG further suggests that existing monitoring system has
primarily emphasized on heart rate as the main human vital sign for monitoring cardiovascular disease.
However, the literature on heart disease indicates that there are other important vital signs that are also useful
for detecting the disease. The other important vital signs included; respiration, body temperature and the
blood pressure of the patients. Moreover, several previous studies have also stressed on the need to focus on
these vital signs when developing intelligent ECG monitoring system [17, 18].
2.4. Minimal adoption of intelligent sensor
More recently, intelligent sensor is increasingly being recognized as an important technology for
improving ECG monitoring system. Findings of research on intelligent sensor have shown that it can make
the monitoring system more autonomous, easier to use as well as capable of providing constant monitoring of
patients. Despite the various advantages of adopting intelligent sensor technology, the review of past research
on ECG appears to indicate that there are not many studies that have focused and exploited the potentials of
this technology [19-21].
2.5. Limited ability to detect physical activity
As mentioned earlier, contemporary ECG monitoring system has primarily emphasized on heart rate
as the main vital sign for detecting and monitoring heart disease. Given the fact that heart attacks can also](https://image.slidesharecdn.com/341400-210625061029/85/A-general-framework-for-improving-electrocardiography-monitoring-system-with-machine-learning-2-320.jpg)
![Bulletin of Electr Eng and Inf ISSN: 2302-9285
A general framework for improving electrocardiography monitoring system with… (A. M. Khairuddin)
263
occurred in different circumstances (while eating, driving, physical exercising, swimming, etc.), several
studies have recommended the need for ECG monitoring system to detect information on physical activity
such as walking, running, swimming [22, 23].
As presented above, the review of the literature and past studies on ECG appears to indicate that
contemporary ECG system has weaknesses and that they still lack learning capabilities. In view of the
limitations of existing ECG devices as identified in previous studies and given the important role of the
monitoring system in detecting, preventing, controlling and treating health disease, there is a further need to
improve and develop a more capable ECG monitoring system with machine learning. According, the
following section explains machine learning and presents a general framework for improving the ECG
monitoring system by using the technology.
3. MACHINE LEARNING IN HEALTH CARE
With its various capabilities and potentials, machine learning is fundamentally changing health care.
According to [24]–[27], machine learning is replacing the simple estimated score systems that adopt
traditional statistics. More recently, machine learning is being used in precision medicine to help physicians
to make better predictions. Since analyzing the heartbeat through traditional ECG monitoring system may be
prone to human error and is also time consuming, machine learning has been adopted to automate
classification of arrhythmia and improve the monitoring system. According to [28, 29], the ECG automatic
system for arrhythmia classification involves the following four important sequential steps:
3.1. Preprocessing
Preprocessing as the first step in automated ECG system is involved with the reduction of noise in
the ECG signals. Recursive digital filters of the finite impulse response (FIR) is best used for the attenuation
of the known frequency bands [30]. Adaptive filter can also be used to remove noise from ECG signals [31],
but it does not offer great advantages over the finite impulse response (FIR) digital filters [32].
3.2. Segmentation
Following the preprocessing step, the heartbeat segmentation is concerned with the detection of the
R-peak or the QRS complex [33], [34]. In this step, the sensitivity and positive predictivity are used as the
two important measures to evaluate the accuracy of the heartbeat segmentation. Over the years, more
complex methods such as genetic algorithms [35], wavelet transform [36], filter banks [37], and quad level
vector [38] have also been adopted in heartbeat segmentation.
3.3. Feature extraction
The third step in the ECG automatic system for arrhythmia classification involves feature extraction.
One of the common features extracted is the RR interval, which is the time between the R-peak of a heartbeat
with another heartbeat. A study by [39] has demonstrated the use of normalized RR-interval to improve the
classification results. Other feature such as the distances between the fiducial points of a heartbeat (ECG
segments) can also be used in feature extraction. In order to determine the fiducial points, an algorithm has
been proposed by [40].
3.4. Classification
The fourth step involves classification of the different abnormal heartbeats. As far as classification
is concerned, researchers have been using different types of learning algorithms such as support vector
machines (SVM), artificial neural networks (ANN), linear discriminant (LD), and reservoir computing with
logistic regression (RC). There are however variations proposed for the SVM such as the study by [41] which
combined fuzzy theory to refine SVM classification as well as genetic algorithms combined with restricted
fuzzy SVM [42]. Having discussed machine learning and its capabilities, the following section presents the
general framework for using the technology to improve ECG monitoring system.
4. A GENERAL FRAMEWORK FOR IMPROVING ECG MONITORING SYSTEM
As indicated earlier, machine learning has at least three types of learning capabilities that include;
supervised learning, unsupervised learning and reinforcement learning. These learning capabilities can be
used to develop the classifier in the ECG monitoring system to enable it to classify arrhythmia (abnormal
heartbeat) and different human physical activity. By enabling the ECG monitoring system to recognize as
well as classify arrhythmia and physical activity, the system can continuously monitor the abnormal heartbeat
of patients and their physical activity at a much earlier stage. This in turn enables preventive measures and](https://image.slidesharecdn.com/341400-210625061029/85/A-general-framework-for-improving-electrocardiography-monitoring-system-with-machine-learning-3-320.jpg)
![ ISSN: 2302-9285
Bulletin of Electr Eng and Inf, Vol. 8, No. 1, March 2019 : 261– 268
264
more timely interventions. The section below explains briefly the general framework for improving the ECG
monitoring system.
4.1. Machine learning and ECG Arrhythmia classification
Using machine learning in the construction and optimization of the classifier for categorizing ECG
arrhythmia involves several important steps. The steps include; feature extraction, feature normalization,
classifier training, and classifier evaluation [43]. Figure 1 presents the steps in ECG arrhythmia classification.
Figure 1. Steps in ECG arrhythmia classification [43]
4.1.1. ECG signal
The data of ECG signals are acquired from the MIT-BIH arrhythmia database [44]. They are used as
the training set and the testing set for ECG arrhythmia classification. ECG wave form is the representation of
the heartbeat, consisting of five different peaks (P, Q, R, S, T). Their variances, distances and other
characteristics can help to identify the properties of the heart-beat [45]. These peaks are represented by their
respective values of amplitude and time intervals.
4.1.2. Preprocessing of ECG signal
The raw ECG signals collected by the electrodes can be affected by various artifacts (noises) such as
motion, baseline wandering, power-line interference. These artifacts can degrade the accuracy in the ECG
classification. Therefore, the denoising or filtering of these artifacts is important to remove the disturbance in
the fiducial point of the ECG waveform. Several filtering techniques such as high-pass filter, low-pass filter
and adaptive filtering are being used.
4.1.3. Feature extraction
The feature extraction is concerned with the finding of relevant features of ECG data that can
influence the accuracy performance of the ECG classification. If all known features are to be considered, it
will be a challenge that requires heavy computational resources. Examples of the features are wavelet, RR-
interval, R-peak, and QRS amplitude.
4.1.4. Feature selection
Feature selection is concerned with optimizing the features extracted by including only the
important attributes or properties of the ECG signal. This is important to ensure that the ECG classification
will be more accurate.
Train
Evaluation/
Validation
Classifier
Feature
Normalization
ECG Signal
Training set
Feature
Extraction
Testing set
Feature
Extraction
Feature
Normalization
Train](https://image.slidesharecdn.com/341400-210625061029/85/A-general-framework-for-improving-electrocardiography-monitoring-system-with-machine-learning-4-320.jpg)
![Bulletin of Electr Eng and Inf ISSN: 2302-9285
A general framework for improving electrocardiography monitoring system with… (A. M. Khairuddin)
265
4.1.5. Classification
In the classification stage, the ECG signals that represents the heart beats are classified as normal
beats or abnormal beats. The different machine learning classifiers used for ECG heartbeats include support
vector machine (SVM), artificial neural networks (ANN), and k-nearest neighbor (kNN).
4.1.6. Classifier evaluation
Classifiers selected for the arrhythmia classification are then evaluated by using performance
measures such as error rate, sensitivity, specificity, accuracy, and confusion matrix.
4.2. Machine learning and human activity recognition
Similarly, the learning capabilities of machine can also facilitate the ECG monitoring system in
terms of recognizing different human activity. In the case of human activity recognition, the adoption of
machine learning involves four specific phases. The four phases include; data collection, feature extraction,
classification and evaluation [46, 47]. Figure 2 shows the process of human activity recognition (HAR) using
machine learning based wearable sensors.
Figure 2. The Process of HAR based on Wearable Sensors [74]
4.2.1. Data collection
In human activity recognition, the first phase involves data collection by measuring the following
four types of attributes; location, physiological, environmental, and acceleration. For location data, the global
positioning system (GPS) is used to collect real-time navigation data. The physiological signals involve
collecting data on human vital signs such as heart rate, respiration, and skin temperature. Environmental
sensors are combined with accelerometers and other sensors to provide sufficient data on individual
performing physical activities under diverse conditions [48]. The acceleration signal is collected through the
triaxial accelerometer to recognize physical activities such as walking, running, or lying.
4.2.2. Feature extraction
In the feature extraction phase, the raw data collected is being processed. Human activity needs to
be recognized in a time window basis and not in a sample basis because these activities are performed during
long periods of time. Therefore, each time window needs to be applied with feature extraction to filter
relevant information and obtain quantitative measures for signal comparison. The two techniques that can be
used to extract features from time series data are statistical and structural.
Learning and
Inference
Structural
features
Feature extraction
Location data Physiological signals
Acceleration signals Environmental signals
Statistical
features
Recognition
model
Recognized
activity
Data Collection](https://image.slidesharecdn.com/341400-210625061029/85/A-general-framework-for-improving-electrocardiography-monitoring-system-with-machine-learning-5-320.jpg)
![ ISSN: 2302-9285
Bulletin of Electr Eng and Inf, Vol. 8, No. 1, March 2019 : 261– 268
266
4.2.3. Classification
The classifier needs to be trained with the training set (input set) to classify different types of
physical activities. These training set may or may not be labeled. Several important classifiers in human
activity recognition are decision trees, Bayesian methods, support vector machine (SVM), C4.5, and artificial
neural networks (ANN). The classifier selected needs to be evaluated by using different performance metrics
such as accuracy, precision, recall and F-measure.
4.2.4. Evaluation
The selected classifiers for human activity recognition need to be assessed by using different
performance metrics that include accuracy, precision, recall and F-measure.
5. CONCLUSION
This paper reviews the limitations of ECG monitoring system as well as presents a general
framework for improving the monitoring system with machine learning. Based on the review, the paper
identified at least five limitations of past works on ECG monitoring system. The limitations suggest that there
is a need to further improve the ability of ECG monitoring system to classify abnormal heartbeat as well as
physical activity by adopting the capabilities of machine learning. In addition, the paper surveys various key
features and capabilities of machine learning and presents a general framework for using the technology to
improve the ECG monitoring system, particularly in terms of detecting different types of heartbeat and
physical activity. By being able to recognize different types of heartbeat and physical activity, the ECG
monitoring system promises a significant shift in diagnosis and treatment of heart disease.
ACKNOWLEDGEMENTS
The authors acknowledge the financial support provided by Embedded, Networks and Advanced Computing
Research Cluster (ENAC) at Universiti Malaysia Perlis, Malaysia.
REFERENCES
[1] M. Hadjem, O. Salem, and F. Naït-abdesselam, “An ECG Monitoring System For Prediction Of Cardiac Anomalies
Using WBAN,” 2014 IEEE 16th Int. Conf. e-Health Networking, Appl. Serv., pp. 441–446, 2014.
[2] Bernama, “73% of Msians die of hypertension, diabetes, heart disease: MOH,” New Straits Times, 2017. [Online].
Available: https://www.nst.com.my/news/2017/04/228008/73-msians-die-hypertension-diabetes-heart-disease-moh.
[Accessed: 04-Jan-2018].
[3] T. C. Shin, “Malaysians get heart attack at younger age than others,” The Star Malaysia, 2016. [Online]. Available:
http://www.pressreader.com/malaysia/the-star-malaysia/20160807/281608124818240. [Accessed: 04-Jan-2018].
[4] A. Foo, “Younger Malaysians getting heart disease as country tackles obesity epidemic,” MIMS Today, 2016.
[Online]. Available: https://today.mims.com/younger-malaysians-getting-heart-disease-as-country-tackles-obesity-
epidemic. [Accessed: 04-Jan-2018].
[5] A. Tang, “The heart of health,” The Star Malaysia, 2016. [Online]. Available:
https://www.thestar.com.my/news/nation/2017/10/25/the-heart-of-health-too-much-stress-bad-food-habits-
sedentary-lifestyles-these-are-the-familiar-refr/. [Accessed: 04-Jan-2018].
[6] Z. Hashim and C. Y. Siew, “Here are 7 tips to keep your heart healthy,” The Star Malaysia, 2017. [Online].
Available: https://www.star2.com/health/wellness/2017/04/16/7-tips-healthy-heart/. [Accessed: 04-Jan-2018].
[7] J. Lynas, Cooking for a Healthy Heart. Octopus Publishing Group Ltd, 2002.
[8] S. Gayathri, M. Suchetha, and V. Latha, “ECG arrhythmia detection and classification using relevance vector
machine,” Procedia Eng., vol. 38, pp. 1333–1339, 2012.
[9] H. Yang and J. Chai, “The study and design of a wireless ECG monitoring system.,” Biomed. Instrum. Technol.,
vol. 46, no. 5, pp. 395–9, 2012.
[10] T. Zhu, M. Osipov, P. T., J. Oster, D. A. Clifton, and G. D. Clifford, “An Intelligent Cardiac Health Monitoring and
Review System,” Appropr. Healthc. Technol. Low Resour. Settings - AHT2014. 8th Int. Conf. - Promot. access to
Healthc. through Technol., pp. 25–25, 2014.
[11] E. Spanò, S. Di Pascoli, and G. Iannaccone, “Low-Power Wearable ECG Monitoring System for Multiple-Patient
Remote Monitoring,” IEEE Sens. J., vol. 16, no. 13, pp. 5452–5462, 2016.
[12] C. Worringham, A. Rojek, and I. Stewart, “Development and feasibility of a smartphone, ECG and GPS based
system for remotely monitoring exercise in cardiac rehabilitation,” PLoS One, vol. 6, no. 2, 2011.
[13] Alivecor.com, “Kardia Mobile,” 2017. [Online]. Available: https://store.alivecor.com/. [Accessed: 27-May-2017].
[14] Qardio, “QardioCore,” 2018. [Online]. Available: https://www.getqardio.com/qardiocore-wearable-ecg-ekg-
monitor-iphone/.
[15] J. S. Windsor, G. W. Rodway, and H. E. Montgomery, “A Review of Electrocardiography in the High Altitude
Environment,” High Alt. Med. Biol., vol. 11, no. 1, pp. 51–60, 2010.](https://image.slidesharecdn.com/341400-210625061029/85/A-general-framework-for-improving-electrocardiography-monitoring-system-with-machine-learning-6-320.jpg)
![Bulletin of Electr Eng and Inf ISSN: 2302-9285
A general framework for improving electrocardiography monitoring system with… (A. M. Khairuddin)
267
[16] B. A. Reyes et al., “Novel electrodes for underwater ECG monitoring,” IEEE Trans. Biomed. Eng., vol. 61, no. 6,
pp. 1863–1876, 2014.
[17] C. D. Katsis, Y. Goletsis, G. Rigas, and D. I. Fotiadis, “A wearable system for the affective monitoring of car
racing drivers during simulated conditions,” Transp. Res. Part C Emerg. Technol., vol. 19, no. 3, pp. 541–551,
2011.
[18] P. S. Pandian et al., “Smart Vest: Wearable multi-parameter remote physiological monitoring system,” Med. Eng.
Phys., vol. 30, no. 4, pp. 466–477, 2008.
[19] G. Baxendale, “Health Wearables,” Itnow, vol. 58, no. 3, pp. 42–43, 2016.
[20] Y. Sun and X. B. Yu, “Capacitive Biopotential Measurement for Electrophysiological Signal Acquisition: A
Review,” IEEE Sens. J., vol. 16, no. 9, pp. 2832–2853, 2016.
[21] T. Chaichana, Y. Pititeeraphab, M. Sangworasil, and T. Masuura, “Implementation of Wireless Electrocardiogram
Monitoring System,” vol. 4, no. 3, pp. 248–252, 2016.
[22] J. Boyle, M. Karunanithi, T. Wark, W. Chan, and C. Colavitti, “Quantifying functional mobility progress for
chronic disease management,” Annu. Int. Conf. IEEE Eng. Med. Biol. - Proc., pp. 5916–5919, 2006.
[23] M. N. Nyan, F. E. H. Tay, A. W. Y. Tan, and K. H. W. Seah, “Distinguishing fall activities from normal activities
by angular rate characteristics and high-speed camera characterization,” Med. Eng. Phys., vol. 28, no. 8, pp. 842–
849, 2006.
[24] C. Krittanawong, H. Zhang, Z. Wang, M. Aydar, and T. Kitai, “Artificial Intelligence in Precision Cardiovascular
Medicine,” J. Am. Coll. Cardiol., vol. 69, no. 21, pp. 2657–2664, 2017.
[25] N. J. Nilsson, Introduction to Machine Learning : Second Edition, vol. 56. 2010.
[26] P. Domingos, The Master Algorithm: How the Quest for the Ultimate Learning Machine Will Remake Our World.
2015.
[27] S. Ben-David and S. Shalev-Shwartz, Understanding Machine Learning: From Theory to Algorithms. 2014.
[28] E. J. da S. Luz, W. R. Schwartz, G. Camara-Chavez, and D. Menotti, “ECG-based heartbeat classification for
arrhythmia detection: A survey,” Comput. Methods Programs Biomed., vol. 127, pp. 144–164, 2016.
[29] S. H. Jambukia, V. K. Dabhi, and H. B. Prajapati, “Classification of ECG signals using machine learning
techniques: A survey,” 2015 Int. Conf. Adv. Comput. Eng. Appl., pp. 714–721, 2015.
[30] P. A. Lynn, “Recursive digital filters for biological signals,” Med. Biol. Eng., vol. 9, no. 1, pp. 37–43, 1971.
[31] E. R. Ferrara and B. Widrow, “Fetal Electrocardiogram Enhancement by Time-Sequenced Adaptive Filtering,”
IEEE Trans. Biomed. Eng., vol. BME-29, no. 6, pp. 458–460, 1982.
[32] N. V. Thakor and Y. S. Zhu, “Applications of Adaptive Filtering to ECG Analysis: Noise Cancellation and
Arrhythmia Detection,” IEEE Trans. Biomed. Eng., vol. 38, no. 8, pp. 785–794, 1991.
[33] Y. C. Yeh and W. J. Wang, “QRS complexes detection for ECG signal: The Difference Operation Method,”
Comput. Methods Programs Biomed., vol. 91, no. 3, pp. 245–254, 2008.
[34] O. Sayadi and M. B. Shamsollahi, “A model-based Bayesian framework for ECG beat segmentation,” Physiol.
Meas., vol. 30, no. 3, pp. 335–352, 2009.
[35] R. Poli, S. Cagnoni, and G. Valli, “Genetic Design of Optimum Linear and Nonlinear QRS Detectors,” IEEE
Trans. Biomed. Eng., vol. 42, no. 11, pp. 1137–1141, 1995.
[36] S. Kadambe, R. Murray, and G. Paye Boudreaux-Bartels, “Wavelet transform-based QRS complex detector,” IEEE
Trans. Biomed. Eng., vol. 46, no. 7, pp. 838–848, 1999.
[37] V. X. Afonso, W. J. Tompkins, T. Q. Nguyen, and S. Luo, “ECG beat detection using filter banks,” IEEE Trans.
Biomed. Eng., vol. 46, no. 2, pp. 192–202, 1999.
[38] H. Kim, R. F. Yazicioglu, P. Merken, C. Van Hoof, and H. J. Yoo, “ECG signal compression and classification
algorithm with quad level vector for ECG holter system,” IEEE Trans. Inf. Technol. Biomed., vol. 14, no. 1, pp. 93–
100, 2010.
[39] C. C. Lin and C. M. Yang, “Heartbeat classification using normalized RR-intervals and wavelet features,” Proc. -
2014 Int. Symp. Comput. Consum. Control. IS3C 2014, no. Ldc, pp. 650–653, 2014.
[40] P. Laguna, “Automatic detection of wave boundaries in multilead ECG signals: validation with the CSE database,”
Comput. Biomed. Res., vol. 27, pp. 45–60, 1994.
[41] N. O. Ozcan and F. Gurgen, “Fuzzy support vector machines for ECG arrhythmia detection,” Proc. Int. Conf.
Pattern Recognit., pp. 2973–2976, 2010.
[42] J. A. Nasiri, M. Naghibzadeh, H. S. Yazdi, and B. Naghibzadeh, “ECG arrhythmia classification with support
vector machines and genetic algorithm,” EMS 2009 - UKSim 3rd Eur. Model. Symp. Comput. Model. Simul., pp.
187–192, 2009.
[43] D. Paiva and M. Batista, “Automatic Arrhythmia Classification : A Pattern Recognition Approach Biomedical
Engineering,” 2014.
[44] A. L. Goldberger et al., “PhysioBank, PhysioToolkit, and PhysioNet,” 2014.
[45] E. H. Houssein and M. Kilany, “ECG signals classification : a review,” no. January, 2017.
[46] I. Gyllensten, “Physical activity recognition in daily life using a triaxial accelerometer,” p. 65, 2010.
[47] O. D. Lara and M. A. Labrador, “A Survey on Human Activity Recognition using Wearable Sensors,” IEEE
Commun. Surv. Tutorials, vol. 15, no. 3, pp. 1192–1209, 2013.
[48] J. Yin, Q. Yang, and J. J. Pan, “Sensor-based abnormal human-activity detection,” IEEE Trans. Knowl. …, vol. 20,
no. 8, pp. 1082–1090, 2008.](https://image.slidesharecdn.com/341400-210625061029/85/A-general-framework-for-improving-electrocardiography-monitoring-system-with-machine-learning-7-320.jpg)
A general framework for improving electrocardiography monitoring system with machine learning
- 1. Bulletin of Electrical Engineering and Informatics Vol. 8, No. 1, March 2019, pp. 261~268 ISSN: 2302-9285, DOI: 10.11591/eei.v8i1.1400 261 Journal homepage: http://beei.org/index.php/EEI A general framework for improving electrocardiography monitoring system with machine learning A. M. Khairuddin, KU Azir K. N. F., P. Eh Kan Embedded, Networks and Advanced Computing Research Cluster (ENAC), School of Computer and Communication Engineering, Universiti Malaysia Perlis, Kampus Pauh, 02600 Ulu Pauh, Perlis, Malaysia Article Info ABSTRACT Article history: Received Oct 28, 2018 Revised Nov 30, 2018 Accepted Dec 26, 2018 As one of the most important health monitoring systems, electrocardiography (ECG) is used to obtain information about the structure and functions of the human heart for detecting and preventing cardiovascular disease. Given its important role, it is vital that the ECG monitoring system provides relevant and accurate information about the heart. Over the years, numerous attempts were made to design and develop more effective ECG monitoring system. Nonetheless, the literature reveals not only several limitations in conventional ECG monitoring system but also emphasizes on the need to adopt new technology such as machine learning to improve the monitoring system as well as its medical applications. This paper reviews previous works on machine learning to explain its key features, capabilities as well as presents a general framework for improving ECG monitoring system. Keywords: Electrocardiography Framework Limitations Machine learning Medical applications Copyright © 2019 Institute of Advanced Engineering and Science. All rights reserved. Corresponding Author: A. M. Khairuddin, Embedded, Networks and Advanced Computing Research Cluster (ENAC), School of Computer and Communication Engineering, Universiti Malaysia Perlis, Kampus Pauh, 02600 Ulu Pauh, Perlis, Malaysia. Email: adamkhairuddin@studentmail.unimap.edu.my 1. INTRODUCTION Cardiovascular disease has been established as a major threat to human lives and public health. The statistics concerning cardiovascular disease are alarming. It has been reported that the disease kills as many as six million people every year. Most of these deaths resulted from heart attacks. About half of all heart attacks are fatal and a third of the deaths occur before reaching hospitals. Nonetheless, heart disease is potentially avoidable and preventable. The disease has been primarily and extensively diagnosed by using the electrocardiograph (ECG) monitoring system. With its capabilities to detect, monitor and provide useful information about the heart, the ECG monitoring system has been able to reduce the number of premature deaths, improved healthcare services and quality of life as well as lowered healthcare costs [1-7]. Electrocardiograph monitoring system has become the standard procedure for diagnosing the heart since the 1940s. However, the earliest ECG monitoring system was basically developed with electrically primitive analog electronics. Since the first electrocardiography device was introduced, there has been numerous attempts to introduce better systems. Over the years, the ECG system has continued to evolve as well as improve. The advances in new technologies have allowed researchers to design and develop more effective ECG monitoring system. For instance, contemporary ECG monitoring system adopt analog-to- digital converter to convert the signal convey by the system to digital signal that can be manipulated with digital electronics. Throughout the years, improvement in ECG monitoring system and devices was driven by the advances, capabilities and potentials of new technologies such as pervasive computing, wireless mesh networks, 3D printing, machine learning, cloud computing and big data analytics. These technologies have
- 2. ISSN: 2302-9285 Bulletin of Electr Eng and Inf, Vol. 8, No. 1, March 2019 : 261– 268 262 shape and reshape important aspects of health care and learning. For instance, in health care, machine learning promises a dramatic shift in diagnosis and treatment of heart disease. Rather than wait for troublesome symptoms of the health disease to appear, through machine learning doctors now can continuously monitor the vital signs of patients from a much earlier stage. Despite the capabilities and potentials of machine learning, the literature reveals not only limited research but also not much information on the adoption of machine learning in contemporary ECG monitoring system. This paper attempts to provide some insights into the adoption of machine learning in improving ECG monitoring system. The first part of the paper reviews previous works on ECG monitoring system to highlight its limitations. Following this, the second part examines the literature on machine learning as well as presents a general framework for improving ECG monitoring system. 2. LIMITATIONS OF ECG MONITORING SYSTEM Electrocardiography (ECG) monitoring system is used to obtain information about the electrical activity of the human heart and to detect cardiovascular disease. The literature indicates that over the years, different types of ECG monitoring system have been developed by using new technologies. The review of the past studies on ECG monitoring system however reveals a great deal of variance in terms of their capabilities and performance. More specifically, the review of the literature reveals several limitations. The limitations as identified in previous research are presented below. 2.1. Lack of learning capabilities Despite the new technologies and efforts to develop intelligent ECG monitoring system, the review of the literature and past studies indicates that existing system has limited capabilities. For instance, existing ECG monitoring system appears to lack the ability to learn without being programmed. The learning ability of the monitoring system can be improved by making it more responsive and interactive through machine learning. However, the review of past research on ECG indicates very few studies have attempted to adopt this technology. The adoption of machine learning in ECG monitoring system is useful because it can not only help the monitoring system to learn but also be able to provide better detection and prediction of cardiac anomalies [8]. 2.2. Limited real time monitoring Real time monitoring capability of existing ECG monitoring system seems to be not only limited but also not emphasized. Intelligent ECG monitoring system can not only monitor the ECG signal of patients in real time but also detect their different physical activity as well. Equipped with machine learning, the monitoring system can automatically collect information on the physiological as well as physical behavior of patients. Machine learning helps to process, analyze and wirelessly send the information directly to the cardiologist for further evaluation. This will save the time needed to visit and examine patients’ vital signs on a regular basis. Furthermore, the monitoring system can be used to gather medical information, monitor patients in remote places, high altitude as well as underwater [9-16]. 2.3. Inadequate focus on vital signs The review of the previous studies on ECG further suggests that existing monitoring system has primarily emphasized on heart rate as the main human vital sign for monitoring cardiovascular disease. However, the literature on heart disease indicates that there are other important vital signs that are also useful for detecting the disease. The other important vital signs included; respiration, body temperature and the blood pressure of the patients. Moreover, several previous studies have also stressed on the need to focus on these vital signs when developing intelligent ECG monitoring system [17, 18]. 2.4. Minimal adoption of intelligent sensor More recently, intelligent sensor is increasingly being recognized as an important technology for improving ECG monitoring system. Findings of research on intelligent sensor have shown that it can make the monitoring system more autonomous, easier to use as well as capable of providing constant monitoring of patients. Despite the various advantages of adopting intelligent sensor technology, the review of past research on ECG appears to indicate that there are not many studies that have focused and exploited the potentials of this technology [19-21]. 2.5. Limited ability to detect physical activity As mentioned earlier, contemporary ECG monitoring system has primarily emphasized on heart rate as the main vital sign for detecting and monitoring heart disease. Given the fact that heart attacks can also
- 3. Bulletin of Electr Eng and Inf ISSN: 2302-9285 A general framework for improving electrocardiography monitoring system with… (A. M. Khairuddin) 263 occurred in different circumstances (while eating, driving, physical exercising, swimming, etc.), several studies have recommended the need for ECG monitoring system to detect information on physical activity such as walking, running, swimming [22, 23]. As presented above, the review of the literature and past studies on ECG appears to indicate that contemporary ECG system has weaknesses and that they still lack learning capabilities. In view of the limitations of existing ECG devices as identified in previous studies and given the important role of the monitoring system in detecting, preventing, controlling and treating health disease, there is a further need to improve and develop a more capable ECG monitoring system with machine learning. According, the following section explains machine learning and presents a general framework for improving the ECG monitoring system by using the technology. 3. MACHINE LEARNING IN HEALTH CARE With its various capabilities and potentials, machine learning is fundamentally changing health care. According to [24]–[27], machine learning is replacing the simple estimated score systems that adopt traditional statistics. More recently, machine learning is being used in precision medicine to help physicians to make better predictions. Since analyzing the heartbeat through traditional ECG monitoring system may be prone to human error and is also time consuming, machine learning has been adopted to automate classification of arrhythmia and improve the monitoring system. According to [28, 29], the ECG automatic system for arrhythmia classification involves the following four important sequential steps: 3.1. Preprocessing Preprocessing as the first step in automated ECG system is involved with the reduction of noise in the ECG signals. Recursive digital filters of the finite impulse response (FIR) is best used for the attenuation of the known frequency bands [30]. Adaptive filter can also be used to remove noise from ECG signals [31], but it does not offer great advantages over the finite impulse response (FIR) digital filters [32]. 3.2. Segmentation Following the preprocessing step, the heartbeat segmentation is concerned with the detection of the R-peak or the QRS complex [33], [34]. In this step, the sensitivity and positive predictivity are used as the two important measures to evaluate the accuracy of the heartbeat segmentation. Over the years, more complex methods such as genetic algorithms [35], wavelet transform [36], filter banks [37], and quad level vector [38] have also been adopted in heartbeat segmentation. 3.3. Feature extraction The third step in the ECG automatic system for arrhythmia classification involves feature extraction. One of the common features extracted is the RR interval, which is the time between the R-peak of a heartbeat with another heartbeat. A study by [39] has demonstrated the use of normalized RR-interval to improve the classification results. Other feature such as the distances between the fiducial points of a heartbeat (ECG segments) can also be used in feature extraction. In order to determine the fiducial points, an algorithm has been proposed by [40]. 3.4. Classification The fourth step involves classification of the different abnormal heartbeats. As far as classification is concerned, researchers have been using different types of learning algorithms such as support vector machines (SVM), artificial neural networks (ANN), linear discriminant (LD), and reservoir computing with logistic regression (RC). There are however variations proposed for the SVM such as the study by [41] which combined fuzzy theory to refine SVM classification as well as genetic algorithms combined with restricted fuzzy SVM [42]. Having discussed machine learning and its capabilities, the following section presents the general framework for using the technology to improve ECG monitoring system. 4. A GENERAL FRAMEWORK FOR IMPROVING ECG MONITORING SYSTEM As indicated earlier, machine learning has at least three types of learning capabilities that include; supervised learning, unsupervised learning and reinforcement learning. These learning capabilities can be used to develop the classifier in the ECG monitoring system to enable it to classify arrhythmia (abnormal heartbeat) and different human physical activity. By enabling the ECG monitoring system to recognize as well as classify arrhythmia and physical activity, the system can continuously monitor the abnormal heartbeat of patients and their physical activity at a much earlier stage. This in turn enables preventive measures and
- 4. ISSN: 2302-9285 Bulletin of Electr Eng and Inf, Vol. 8, No. 1, March 2019 : 261– 268 264 more timely interventions. The section below explains briefly the general framework for improving the ECG monitoring system. 4.1. Machine learning and ECG Arrhythmia classification Using machine learning in the construction and optimization of the classifier for categorizing ECG arrhythmia involves several important steps. The steps include; feature extraction, feature normalization, classifier training, and classifier evaluation [43]. Figure 1 presents the steps in ECG arrhythmia classification. Figure 1. Steps in ECG arrhythmia classification [43] 4.1.1. ECG signal The data of ECG signals are acquired from the MIT-BIH arrhythmia database [44]. They are used as the training set and the testing set for ECG arrhythmia classification. ECG wave form is the representation of the heartbeat, consisting of five different peaks (P, Q, R, S, T). Their variances, distances and other characteristics can help to identify the properties of the heart-beat [45]. These peaks are represented by their respective values of amplitude and time intervals. 4.1.2. Preprocessing of ECG signal The raw ECG signals collected by the electrodes can be affected by various artifacts (noises) such as motion, baseline wandering, power-line interference. These artifacts can degrade the accuracy in the ECG classification. Therefore, the denoising or filtering of these artifacts is important to remove the disturbance in the fiducial point of the ECG waveform. Several filtering techniques such as high-pass filter, low-pass filter and adaptive filtering are being used. 4.1.3. Feature extraction The feature extraction is concerned with the finding of relevant features of ECG data that can influence the accuracy performance of the ECG classification. If all known features are to be considered, it will be a challenge that requires heavy computational resources. Examples of the features are wavelet, RR- interval, R-peak, and QRS amplitude. 4.1.4. Feature selection Feature selection is concerned with optimizing the features extracted by including only the important attributes or properties of the ECG signal. This is important to ensure that the ECG classification will be more accurate. Train Evaluation/ Validation Classifier Feature Normalization ECG Signal Training set Feature Extraction Testing set Feature Extraction Feature Normalization Train
- 5. Bulletin of Electr Eng and Inf ISSN: 2302-9285 A general framework for improving electrocardiography monitoring system with… (A. M. Khairuddin) 265 4.1.5. Classification In the classification stage, the ECG signals that represents the heart beats are classified as normal beats or abnormal beats. The different machine learning classifiers used for ECG heartbeats include support vector machine (SVM), artificial neural networks (ANN), and k-nearest neighbor (kNN). 4.1.6. Classifier evaluation Classifiers selected for the arrhythmia classification are then evaluated by using performance measures such as error rate, sensitivity, specificity, accuracy, and confusion matrix. 4.2. Machine learning and human activity recognition Similarly, the learning capabilities of machine can also facilitate the ECG monitoring system in terms of recognizing different human activity. In the case of human activity recognition, the adoption of machine learning involves four specific phases. The four phases include; data collection, feature extraction, classification and evaluation [46, 47]. Figure 2 shows the process of human activity recognition (HAR) using machine learning based wearable sensors. Figure 2. The Process of HAR based on Wearable Sensors [74] 4.2.1. Data collection In human activity recognition, the first phase involves data collection by measuring the following four types of attributes; location, physiological, environmental, and acceleration. For location data, the global positioning system (GPS) is used to collect real-time navigation data. The physiological signals involve collecting data on human vital signs such as heart rate, respiration, and skin temperature. Environmental sensors are combined with accelerometers and other sensors to provide sufficient data on individual performing physical activities under diverse conditions [48]. The acceleration signal is collected through the triaxial accelerometer to recognize physical activities such as walking, running, or lying. 4.2.2. Feature extraction In the feature extraction phase, the raw data collected is being processed. Human activity needs to be recognized in a time window basis and not in a sample basis because these activities are performed during long periods of time. Therefore, each time window needs to be applied with feature extraction to filter relevant information and obtain quantitative measures for signal comparison. The two techniques that can be used to extract features from time series data are statistical and structural. Learning and Inference Structural features Feature extraction Location data Physiological signals Acceleration signals Environmental signals Statistical features Recognition model Recognized activity Data Collection
- 6. ISSN: 2302-9285 Bulletin of Electr Eng and Inf, Vol. 8, No. 1, March 2019 : 261– 268 266 4.2.3. Classification The classifier needs to be trained with the training set (input set) to classify different types of physical activities. These training set may or may not be labeled. Several important classifiers in human activity recognition are decision trees, Bayesian methods, support vector machine (SVM), C4.5, and artificial neural networks (ANN). The classifier selected needs to be evaluated by using different performance metrics such as accuracy, precision, recall and F-measure. 4.2.4. Evaluation The selected classifiers for human activity recognition need to be assessed by using different performance metrics that include accuracy, precision, recall and F-measure. 5. CONCLUSION This paper reviews the limitations of ECG monitoring system as well as presents a general framework for improving the monitoring system with machine learning. Based on the review, the paper identified at least five limitations of past works on ECG monitoring system. The limitations suggest that there is a need to further improve the ability of ECG monitoring system to classify abnormal heartbeat as well as physical activity by adopting the capabilities of machine learning. In addition, the paper surveys various key features and capabilities of machine learning and presents a general framework for using the technology to improve the ECG monitoring system, particularly in terms of detecting different types of heartbeat and physical activity. By being able to recognize different types of heartbeat and physical activity, the ECG monitoring system promises a significant shift in diagnosis and treatment of heart disease. ACKNOWLEDGEMENTS The authors acknowledge the financial support provided by Embedded, Networks and Advanced Computing Research Cluster (ENAC) at Universiti Malaysia Perlis, Malaysia. REFERENCES [1] M. Hadjem, O. Salem, and F. Naït-abdesselam, “An ECG Monitoring System For Prediction Of Cardiac Anomalies Using WBAN,” 2014 IEEE 16th Int. Conf. e-Health Networking, Appl. Serv., pp. 441–446, 2014. [2] Bernama, “73% of Msians die of hypertension, diabetes, heart disease: MOH,” New Straits Times, 2017. [Online]. Available: https://www.nst.com.my/news/2017/04/228008/73-msians-die-hypertension-diabetes-heart-disease-moh. [Accessed: 04-Jan-2018]. [3] T. C. Shin, “Malaysians get heart attack at younger age than others,” The Star Malaysia, 2016. [Online]. Available: http://www.pressreader.com/malaysia/the-star-malaysia/20160807/281608124818240. [Accessed: 04-Jan-2018]. [4] A. Foo, “Younger Malaysians getting heart disease as country tackles obesity epidemic,” MIMS Today, 2016. [Online]. Available: https://today.mims.com/younger-malaysians-getting-heart-disease-as-country-tackles-obesity- epidemic. [Accessed: 04-Jan-2018]. [5] A. Tang, “The heart of health,” The Star Malaysia, 2016. [Online]. Available: https://www.thestar.com.my/news/nation/2017/10/25/the-heart-of-health-too-much-stress-bad-food-habits- sedentary-lifestyles-these-are-the-familiar-refr/. [Accessed: 04-Jan-2018]. [6] Z. Hashim and C. Y. Siew, “Here are 7 tips to keep your heart healthy,” The Star Malaysia, 2017. [Online]. Available: https://www.star2.com/health/wellness/2017/04/16/7-tips-healthy-heart/. [Accessed: 04-Jan-2018]. [7] J. Lynas, Cooking for a Healthy Heart. Octopus Publishing Group Ltd, 2002. [8] S. Gayathri, M. Suchetha, and V. Latha, “ECG arrhythmia detection and classification using relevance vector machine,” Procedia Eng., vol. 38, pp. 1333–1339, 2012. [9] H. Yang and J. Chai, “The study and design of a wireless ECG monitoring system.,” Biomed. Instrum. Technol., vol. 46, no. 5, pp. 395–9, 2012. [10] T. Zhu, M. Osipov, P. T., J. Oster, D. A. Clifton, and G. D. Clifford, “An Intelligent Cardiac Health Monitoring and Review System,” Appropr. Healthc. Technol. Low Resour. Settings - AHT2014. 8th Int. Conf. - Promot. access to Healthc. through Technol., pp. 25–25, 2014. [11] E. Spanò, S. Di Pascoli, and G. Iannaccone, “Low-Power Wearable ECG Monitoring System for Multiple-Patient Remote Monitoring,” IEEE Sens. J., vol. 16, no. 13, pp. 5452–5462, 2016. [12] C. Worringham, A. Rojek, and I. Stewart, “Development and feasibility of a smartphone, ECG and GPS based system for remotely monitoring exercise in cardiac rehabilitation,” PLoS One, vol. 6, no. 2, 2011. [13] Alivecor.com, “Kardia Mobile,” 2017. [Online]. Available: https://store.alivecor.com/. [Accessed: 27-May-2017]. [14] Qardio, “QardioCore,” 2018. [Online]. Available: https://www.getqardio.com/qardiocore-wearable-ecg-ekg- monitor-iphone/. [15] J. S. Windsor, G. W. Rodway, and H. E. Montgomery, “A Review of Electrocardiography in the High Altitude Environment,” High Alt. Med. Biol., vol. 11, no. 1, pp. 51–60, 2010.
- 7. Bulletin of Electr Eng and Inf ISSN: 2302-9285 A general framework for improving electrocardiography monitoring system with… (A. M. Khairuddin) 267 [16] B. A. Reyes et al., “Novel electrodes for underwater ECG monitoring,” IEEE Trans. Biomed. Eng., vol. 61, no. 6, pp. 1863–1876, 2014. [17] C. D. Katsis, Y. Goletsis, G. Rigas, and D. I. Fotiadis, “A wearable system for the affective monitoring of car racing drivers during simulated conditions,” Transp. Res. Part C Emerg. Technol., vol. 19, no. 3, pp. 541–551, 2011. [18] P. S. Pandian et al., “Smart Vest: Wearable multi-parameter remote physiological monitoring system,” Med. Eng. Phys., vol. 30, no. 4, pp. 466–477, 2008. [19] G. Baxendale, “Health Wearables,” Itnow, vol. 58, no. 3, pp. 42–43, 2016. [20] Y. Sun and X. B. Yu, “Capacitive Biopotential Measurement for Electrophysiological Signal Acquisition: A Review,” IEEE Sens. J., vol. 16, no. 9, pp. 2832–2853, 2016. [21] T. Chaichana, Y. Pititeeraphab, M. Sangworasil, and T. Masuura, “Implementation of Wireless Electrocardiogram Monitoring System,” vol. 4, no. 3, pp. 248–252, 2016. [22] J. Boyle, M. Karunanithi, T. Wark, W. Chan, and C. Colavitti, “Quantifying functional mobility progress for chronic disease management,” Annu. Int. Conf. IEEE Eng. Med. Biol. - Proc., pp. 5916–5919, 2006. [23] M. N. Nyan, F. E. H. Tay, A. W. Y. Tan, and K. H. W. Seah, “Distinguishing fall activities from normal activities by angular rate characteristics and high-speed camera characterization,” Med. Eng. Phys., vol. 28, no. 8, pp. 842– 849, 2006. [24] C. Krittanawong, H. Zhang, Z. Wang, M. Aydar, and T. Kitai, “Artificial Intelligence in Precision Cardiovascular Medicine,” J. Am. Coll. Cardiol., vol. 69, no. 21, pp. 2657–2664, 2017. [25] N. J. Nilsson, Introduction to Machine Learning : Second Edition, vol. 56. 2010. [26] P. Domingos, The Master Algorithm: How the Quest for the Ultimate Learning Machine Will Remake Our World. 2015. [27] S. Ben-David and S. Shalev-Shwartz, Understanding Machine Learning: From Theory to Algorithms. 2014. [28] E. J. da S. Luz, W. R. Schwartz, G. Camara-Chavez, and D. Menotti, “ECG-based heartbeat classification for arrhythmia detection: A survey,” Comput. Methods Programs Biomed., vol. 127, pp. 144–164, 2016. [29] S. H. Jambukia, V. K. Dabhi, and H. B. Prajapati, “Classification of ECG signals using machine learning techniques: A survey,” 2015 Int. Conf. Adv. Comput. Eng. Appl., pp. 714–721, 2015. [30] P. A. Lynn, “Recursive digital filters for biological signals,” Med. Biol. Eng., vol. 9, no. 1, pp. 37–43, 1971. [31] E. R. Ferrara and B. Widrow, “Fetal Electrocardiogram Enhancement by Time-Sequenced Adaptive Filtering,” IEEE Trans. Biomed. Eng., vol. BME-29, no. 6, pp. 458–460, 1982. [32] N. V. Thakor and Y. S. Zhu, “Applications of Adaptive Filtering to ECG Analysis: Noise Cancellation and Arrhythmia Detection,” IEEE Trans. Biomed. Eng., vol. 38, no. 8, pp. 785–794, 1991. [33] Y. C. Yeh and W. J. Wang, “QRS complexes detection for ECG signal: The Difference Operation Method,” Comput. Methods Programs Biomed., vol. 91, no. 3, pp. 245–254, 2008. [34] O. Sayadi and M. B. Shamsollahi, “A model-based Bayesian framework for ECG beat segmentation,” Physiol. Meas., vol. 30, no. 3, pp. 335–352, 2009. [35] R. Poli, S. Cagnoni, and G. Valli, “Genetic Design of Optimum Linear and Nonlinear QRS Detectors,” IEEE Trans. Biomed. Eng., vol. 42, no. 11, pp. 1137–1141, 1995. [36] S. Kadambe, R. Murray, and G. Paye Boudreaux-Bartels, “Wavelet transform-based QRS complex detector,” IEEE Trans. Biomed. Eng., vol. 46, no. 7, pp. 838–848, 1999. [37] V. X. Afonso, W. J. Tompkins, T. Q. Nguyen, and S. Luo, “ECG beat detection using filter banks,” IEEE Trans. Biomed. Eng., vol. 46, no. 2, pp. 192–202, 1999. [38] H. Kim, R. F. Yazicioglu, P. Merken, C. Van Hoof, and H. J. Yoo, “ECG signal compression and classification algorithm with quad level vector for ECG holter system,” IEEE Trans. Inf. Technol. Biomed., vol. 14, no. 1, pp. 93– 100, 2010. [39] C. C. Lin and C. M. Yang, “Heartbeat classification using normalized RR-intervals and wavelet features,” Proc. - 2014 Int. Symp. Comput. Consum. Control. IS3C 2014, no. Ldc, pp. 650–653, 2014. [40] P. Laguna, “Automatic detection of wave boundaries in multilead ECG signals: validation with the CSE database,” Comput. Biomed. Res., vol. 27, pp. 45–60, 1994. [41] N. O. Ozcan and F. Gurgen, “Fuzzy support vector machines for ECG arrhythmia detection,” Proc. Int. Conf. Pattern Recognit., pp. 2973–2976, 2010. [42] J. A. Nasiri, M. Naghibzadeh, H. S. Yazdi, and B. Naghibzadeh, “ECG arrhythmia classification with support vector machines and genetic algorithm,” EMS 2009 - UKSim 3rd Eur. Model. Symp. Comput. Model. Simul., pp. 187–192, 2009. [43] D. Paiva and M. Batista, “Automatic Arrhythmia Classification : A Pattern Recognition Approach Biomedical Engineering,” 2014. [44] A. L. Goldberger et al., “PhysioBank, PhysioToolkit, and PhysioNet,” 2014. [45] E. H. Houssein and M. Kilany, “ECG signals classification : a review,” no. January, 2017. [46] I. Gyllensten, “Physical activity recognition in daily life using a triaxial accelerometer,” p. 65, 2010. [47] O. D. Lara and M. A. Labrador, “A Survey on Human Activity Recognition using Wearable Sensors,” IEEE Commun. Surv. Tutorials, vol. 15, no. 3, pp. 1192–1209, 2013. [48] J. Yin, Q. Yang, and J. J. Pan, “Sensor-based abnormal human-activity detection,” IEEE Trans. Knowl. …, vol. 20, no. 8, pp. 1082–1090, 2008.
- 8. ISSN: 2302-9285 Bulletin of Electr Eng and Inf, Vol. 8, No. 1, March 2019 : 261– 268 268 BIOGRAPHIES OF AUTHORS Adam Mohd Khairuddin is a Ph.D student at School of Computer and Communication Engineering (SCCE) since 2017. His research interests include feature extraction, machine learning and electrocardiography (ECG). Currently working as, a senior lecturer at School of Computer and Communication Engineering (SCCE) since 2014. She is an active member and principal researcher of Embedded, Network and Advanced Computing (ENAC). She also undertaking teaching, research and administrative duties within. Meanwhile she also supervising Ph.D/Master and degree students whereas her research focus on the ambient intelligence, human-computer interaction, embedded computer, and Internet-of-Things. She has currently published 11 national and international conference papers. Phaklen Ehkan received the BEng Electrical-Electronic Engineering (UTM), MSc. IT (UUM) and Ph.D in Computer Engineering (UniMAP-University of Birmingham, UK). He worked as an Engineer/Sr. Engineer in MNC- Electronic Industries for six years before joined the University Malaysia Perlis as a lecturer in 2003. Currently, he is an Associate Professor attached to School of Computer and Communication Engineering, UniMAP. His research interests include Reconfigurable Computing and FPGA, Digital Design and Embedded System, Digital and Image Processing, System on Chip (SoC), Smart System and IoT. He has published over 80 articles in International Journals and Proceedings Scopus indexed. Dr. Phaklen Ehkan is currently a Chartered Engineer (UK), Professional Technologist (MBOT), graduate member of BEM, member of IEEE, BCS and IACSIT.