Arrythmia Classiﬁcation Using MATLAB

Classiﬁcation Learner App

Cesar N. Silva, Fernanda F. Lopes, Jefferson A. Matos

and Maria Claudia F. Castro

Electrical Engineering Department, Centro Universitario FEI, S

ao Bernardo do Campo, Brazil

Keywords:

Electrocardiogram (ECG), Cardiac Arrhythmia, Classiﬁcation, Machine-Learning.

Abstract:

Vital sign monitoring is becoming a part of our daily lives, emerging as a trend of smart wearable devices used

to manage health. Cardiac arrhythmia is any variation in the normal heartbeat rhythm, causing the heart to beat

improperly. This work presents a study on the classiﬁcation of cardiac arrhythmias in 4 classes, Normal (N),

Supraventricular Ectopic (SVE), Ventricular Ectopic (LV), and Fusion of Normal and Ventricular (F). Using

the MIT-BIH Arrhythmia Database and the Classiﬁcation Learner App from MATLAB

for training, it was

possible to investigate 24 models, where the Subspace KNN Ensemble obtained the best accuracy (74.4%)

and was later used for implementation in the suggested user interface application.

1 INTRODUCTION

According to the World Health Organization (WHO,

2021), cardiovascular diseases (CVDs) are the lead-

ing global cause of death, taking an estimated 17.9

million lives each year, with more than 75% occur-

ring in low- and middle-income countries (LMICs).

While intensive global efforts to prevent cardiovascu-

lar disease are underway, cardiac arrhythmias remain

neglected, especially in LMICs (Mkoko et al., 2020).

Cardiac arrhythmia refers to any variation in the

normal heartbeat rhythm, causing the heart to beat too

fast (Tachycardia), too slowly (Bradycardia), or errat-

ically. An arrhythmia occurs when the sinus node,

known as the natural pacemaker, develops an abnor-

mal rhythm, the normal conduction pathway is inter-

rupted, or when another part of the heart takes over

as the pacemaker (Humphreys et al., 2011; Ameri-

can Heart Association (AHA), 2016). When the heart

does not beat properly, it can not effectively pump

blood, and the organs, such as the brain, lungs, and

even the heart may be damaged or shut down. Thus,

arrhythmias should be diagnosed and treated as early

as possible to reduce the risk of sudden death.

Currently, vital sign monitoring is becoming a part

of our daily lives, emerging as a trend of smart wear-

able devices used to manage health. Their adoption

has further accelerated with the growth of telehealth

during the COVID-19 pandemic. The most widely

https://orcid.org/0000-0002-8800-713X

https://orcid.org/0000-0002-2751-0014

used tool for monitoring and diagnosing heart func-

tion, such as arrhythmia, is the Electrocardiogram

(ECG), a graphical representation of the heart’s elec-

trical activity. For an early diagnosis, an efﬁcient, in-

telligent, and robust automated arrhythmia classiﬁca-

tion system must be incorporated into smart wearable

devices (Bayoumy et al., 2021).

To cope with such challenges, several works

have been carried out on arrhythmia classiﬁcation.

Machine-learning-oriented techniques are adopted,

requiring at least ﬁve steps: ECG signal condition-

ing such as ampliﬁcation and denoising, feature ex-

traction, feature selection, classiﬁcation, and perfor-

mance analysis (Mohebbanaaz et al., 2020).

ECG signal features mainly depend on time in-

terval, amplitude, and segment duration. The most

common are morphological information such as am-

plitudes and intervals identiﬁcation of peaks P, R, T,

and QRS complex, as well as information about the

RR range/interspace, which is the distance between

peaks of two successive R waves in the ECG sig-

nal (de Albuquerque et al., 2018; Celin and Vasanth,

2018; Kuila et al., 2020; Mohebbanaaz et al., 2022).

Recently, a great interest has been in the applica-

tion of classiﬁcation algorithms based on Deep Learn-

ing (Zhang et al., 2020; Hassan et al., 2022; Irfan

et al., 2022) with accuracies up to 99.35%. How-

ever, other techniques have also been used to clas-

sify arrhythmias, such as Decision Trees (Moheb-

banaaz et al., 2022), Random Forest (AbdElMoneem

et al., 2020), K-Nearest Neighbor (Mustaqeem et al.,

2018; AbdElMoneem et al., 2020), Ensemble Clas-

220

Silva, C., Lopes, F., Matos, J. and Castro, M.

Arrythmia Classiﬁcation Using MATLAB

 Classiﬁcation Learner App.

DOI: 10.5220/0011666300003414

In Proceedings of the 16th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2023) - Volume 4: BIOSIGNALS, pages 220-225

ISBN: 978-989-758-631-6; ISSN: 2184-4305

 2023 by SCITEPRESS – Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)

siﬁers (Shalini et al., 2019; AbdElMoneem et al.,

2020), Support Vector Machines (de Albuquerque

et al., 2018; Mustaqeem et al., 2018; AbdElMoneem

et al., 2020), and others (de Albuquerque et al., 2018).

In these cases, accuracies range from 70 to 98%.

Most works comprise multiclass classiﬁcation to

differentiate among up to 16 types of arrhythmias.

However, given that atrial ﬁbrillation (AF) is the most

common heart arrhythmia, its detection has received

speciﬁc attention, either in a simple recognition sys-

tem or to classify it into subtypes (Celin and Vasanth,

2018; Horoba et al., 2019; Ganapathy et al., 2021;

Ramesh et al., 2021; Sager et al., 2021; da Silva et al.,

2021; Fuadah and Lim, 2022).

In this context, this work presents a study on the

detection and classiﬁcation of cardiac arrhythmias us-

ing the MATLAB

Classiﬁcation Learner applica-

tion. Four classes were deﬁned according to the As-

sociation for the Advancement of Medical Instrumen-

tation (AAMI): Normal (N), Ectopic Supraventricular

(SVE), Ventricular Ectopic (VE), and Fusion of Nor-

mal and Ventricular (F). As a suggestion, a low-cost

device for ECG acquisition and a user interface for

communication with the health professional is also

presented.

2 MATERIALS AND METHODS

Figure 1: Data processing ﬂowchart.

The ﬂowchart for conducting the research is shown in

Figure 1. Firstly, the MIT-BIH dataset was prepared

for model training by performing signal extraction,

using the WFDB package, class preparation, data bal-

ance, and segmentation. Then ECG signals were par-

titioned into two sets of records in order to separate

the patients into training/validation (80%) and testing

(20%) groups. The structured data was exported to

the MATLAB

Classiﬁcation Learner to the training

process and investigation of the best two models ex-

ported back to the MATLAB

algorithm to the test

phase and ﬁnal metric analysis.

2.1 Dataset

The MIT-BIH Arrhythmia Database, which is pub-

licly available online at physionet.org (Goldberger

et al., 2000), is a well-known and worldwide used

standard dataset for arrhythmia detectors evaluation

(de Albuquerque et al., 2018; Celin and Vasanth,

2018; Kuila et al., 2020; Hassan et al., 2022; Ir-

fan et al., 2022; Mohebbanaaz et al., 2022). It was

collected by Boston’s Beth Israel Hospital (BIH) Ar-

rhythmia Laboratory between 1975 and 1979 (Moody

and Mark, 2001). The dataset contained 48 half-hour

excerpts of two-channel ambulatory ECG recordings,

obtained from 47 subjects, both male and female (25

and 22, respectively) of different age groups (between

23 and 89 years). The analog records were digi-

tized according to a sampling rate of 360 Hz, ﬁl-

tered using a 0.1–100 Hz bandpass ﬁlter, the heart-

beats were marked and manually classiﬁed by experts

in 16 classes of arrhythmia.

Following the recommendation of ANSI/AAMI

standard EC57 (ANSI/AAMI, 2020), the 15 arrhyth-

mia classes reported in the database’s annotations

were grouped into 5 classes, as depicted in Table 1.

The 5th class (D), with an Unknown or with a pace-

maker, was discarded.

Table 1: MIT-BIH database classes grouped according to

AAMI Standard.

AAMI

Group

MIT-

BIH

Class

MIT-BIH Class (Description)

N Normal beat

L Left bundle branch block beat

R Right bundle branch block beat

e Atrial escape beat

j Nodal (junctional) escape beat

SVE

A Atrial premature beat

a Aberrated atrial premature beat

J Nodal (junctional) premature

beat

S Supraventricular premature beat

V Premature ventricular contrac-

tion

E Ventricular escape beat

F F Fusion of ventricular and nor-

mal beat

f Fusion of paced and normal beat

/ Paced beat

Q Unclassiﬁable beat

However, as the database is unbalanced, with the

most typical classes having much more examples, the

results could be biased. Data balancing was per-

Arrythmia Classiﬁcation Using MATLAB

 Classiﬁcation Learner App

221

formed through the resampling method with down-

sampling for the majority class. Each one of the con-

sidered classes had 2296 samples. Based on that, data

records were partitioned into two groups for training

and testing according to Table 2.

Before training, the annotated R wave peaks (Fig-

ure 2) were taken into consideration, a window of 300

samples around the peaks (P-149 to P+150 samples)

was segmented. No further pre-processing was done,

nor feature was extracted, other than the data window

(Singh et al., 2019).

Table 2: Data records partitionnig.

Training Testing

101 106 108 109 111 100 103

112 114 115 116 118 105 113

119 121 122 123 124 117 200

201 202 203 205 207 209 212

208 210 214 215 219 213

220 221 222 223 228

230 231 232 233 234

Figure 2: ECG example data.

2.2 Classiﬁcation Learner App

The classiﬁcation learner application provided by

MATLAB

is a toolbox that allows interactive data

analysis training classiﬁers with several machine-

learning models, such as Decision Trees, Discrim-

inant Analysis, Support Vector Machines (SVM),

Nearest Neighbor Classiﬁers (KNN), and Ensemble

Classiﬁers. The app provides classiﬁer performance

metrics, such as validation accuracy, confusion ma-

trix, receiver operating characteristic curve (ROC),

and the area under the ROC curve (AUC), among

other resources.

In this research, a set of 24 classiﬁers were

adopted: Fine, Medium, and Coarse Decision Trees;

Linear and Quadratic Discriminant Analysis; Gaus-

sian and Kernel Naive Bayes; Linear, Quadratic, Cu-

bic, and Gaussian SVM; Fine Medium, Coarse, Co-

sine, Cubic, and Weighted KNN; Boosted Trees,

Bagged Trees, Subspace Discriminant, Subspace

KNN, and RUSBoost Trees Ensemble Classiﬁers.

Default classiﬁer parameters, and the k-fold cross-

validation method, using k=10, were applied. Above

mentioned criteria were used to evaluate the perfor-

mances of different classiﬁers to assess the two classi-

ﬁers that had the highest validation accuracies. Those

models were exported for testing and ﬁnal evaluation.

3 RESULTS

Table 3 shows validation accuracies for all classiﬁers.

Most SVM and KNN classiﬁers reached accuracies

above 90%. Cubic SVM and Subspace KNN Ensem-

ble Classiﬁers reached accuracies of 94%, being the

best, while Naive Bayes Classiﬁers presented had the

worst results. Figures 3 and 4 show the confusion ma-

trices of the best two classiﬁers, from which it can be

noted that the major misclassiﬁcation occurred for the

F class.

These two trained models were exported back to

the MATLAB

algorithm for testing. Figures 5 and 6

show the resulting confusion matrices, with mean ac-

curacies of 67.1% for the Cubic SVM and 74.4% for

the Subspace KNN Ensemble. In both cases, ma-

jor misclassiﬁcation occurred between the SVE e N

classes.

Table 3: Validation Accuracies.

Classiﬁer Accuracy (%)

Cubic SVM 94.1

Subspace KNN Ensemble 94.0

Quadratic SVM 93.4

Fine KNN 93.2

Weighted KNN 92.2

Medium Gaussian SVM 91.8

Cosine KNN 91.7

Bagged Trees Ensemble 91.3

Fine Gaussian SVM 91.0

Medium KNN 90.8

Cubic KNN 90.8

Linear SVM 82.9

Boosted Trees Ensemble 81.6

Fine Tree 81.4

RUSBoost Trees Ensemble 81.0

Coarse KNN 78.5

Quadratic Discriminant 78.4

Coarse Gaussian SVM 78.3

Subspace Discriminant Ensemble 78.0

Linear Discriminant 77.6

Medium Tree 75.3

Coarse Tree 65.0

Kernel Naive Bayes 61.8

Gaussian Naive Bayes 55.9

BIOSIGNALS 2023 - 16th International Conference on Bio-inspired Systems and Signal Processing

222

Figure 3: Validation Confusion Matrix for Cubic SVM.

Figure 4: Validation Confusion Matrix for Subspace KNN

Ensemble.

3.1 Hardware and User Interface

From a practical point of view, we propose the use of a

low-cost device for capturing the ECG signal, consist-

ing of an AD8233 module connected to an Arduino,

and a simple interface developed on the MATLAB

Appdesigner (Figure 7). After the acquisition, and

classiﬁcation, the interface will inform, through an e-

mail, the detection of possible arrhythmia to a doctor

or accredited person.

4 DISCUSSION

The monitoring of vital signs by wearable devices

can contribute to the decentralization of health care,

allowing self-management and anticipation of emer-

gency care. Therefore, even if the ﬁnal diagnosis is

the healthcare professional’s responsibility, machine-

learning techniques can automatically recognize and

classify speciﬁc patterns in these signals, indicating to

Figure 5: Test Confusion Matrix for Cubic SVM.

Figure 6: Test Confusion Matrix for Subspace KNN En-

semble.

the user. Several works in the literature are dedicated

to applying machine-learning techniques to recognize

cardiac arrhythmias, mostly with accuracies between

70% and 98%.

This work presented a study with 24 classiﬁers

using the Classiﬁcation Learner application from

MATLAB

and the MIT-BIH Arrhythmia Database,

which is one of the most used databases. However,

although the validation results were promising, show-

ing accuracies of 94% for Cubic SVM and Subspace

KNN Ensemble, the test phase results showed lower

accuracies (74%), with most misclassiﬁcations be-

tween SVE e N classes.

The ECG data were digitized according to a sam-

pling rate of 360 Hz and ﬁltered using a 0.1–100 Hz

bandpass ﬁlter. Hence, noises like power line inter-

ference, baseline drifts, motion artifacts, and elec-

tromyography noise can be added, and thus the lack

of pre-processing can be affected classiﬁcation per-

formance. From a practical point of view, despite the

need for an acquisition system to implement some ﬁl-

tering in the signal to eliminate noise, the literature

consulted did not clearly show this type of process-

ing.

Another fact was using the raw QRS complex in-

Arrythmia Classiﬁcation Using MATLAB

 Classiﬁcation Learner App

223

Figure 7: Hardware Design.

stead of a set of extracted features. Results showed

that the investigated models could not deal with it.

The literature showed that the only model able to deal

with a raw signal are those based on Deep Learning

due to a proper structure (Zhang et al., 2020; Has-

san et al., 2022; Irfan et al., 2022). Other models

need a set of features able to discriminate different

classes (de Albuquerque et al., 2018; Celin and Vas-

anth, 2018; Kuila et al., 2020; Mohebbanaaz et al.,

2022). Furthermore, despite reducing the number

of classes, arrhythmias groups may interfere in this

discrimination, especially without speciﬁc features as

applied in this study.

Moreover, datasets are usually unbalanced. The

results could be biased because the class with non-

ectopic data has much more samples than the others.

It is critical to balance the dataset or approximate it to

ensure that each class receives the same priority (Ab-

dElMoneem et al., 2020; Hassan et al., 2022). How-

ever, most of the literature work did not mention any

data balance. The use of a resampling approach can

accomplish balancing. Otherwise, as this work imple-

mented data balance only using the down-sampling

technique for the majority class, the number of re-

sulting samples could be not enough for the machine-

learning approach, being responsible for the reached

accuracies. The use of the up-sampling technique in

combination with down-sampling would increase the

number of available samples, improving results.

Despite that, for practical use, some improve-

ments are expected, such as the accuracy increase

for the arrhythmia classiﬁcation, and for a smart de-

vice, the classiﬁer must be embedded instead of being

through a PC interface.

5 CONCLUSION

This work showed a study of 24 models for cardiac

arrhythmia classiﬁcation using the Classiﬁcation App

from MATLAB

and suggested a low-cost device

for capturing the ECG signal with a simple interface

developed on the MATLAB

Appdesigner, allowing

rapid health professional communication for practical

use. Results were promising; however, more attention

should be given to the extraction of features in order

to increase classiﬁcation accuracy and to the imple-

mentation of an embedded system.

ACKNOWLEDGEMENTS

The authors would like to thank Centro Universit

ario

FEI for project support.

REFERENCES

AbdElMoneem, S. S., Said, H. H., and Saad, A. A. (2020).

Arrhythmia disease classiﬁcation and mobile based

system design. Journal of Physics: Conference Se-

ries, 1447(1):012014.

American Heart Association (AHA) (2016).

What is an arrhythmia? Available on-

line: https://www.heart.org/en/health-

topics/arrhythmia/about-arrhythmia.

ANSI/AAMI (2020). ANSI/AAMI EC57:2012/(R)2020 -

Testing and reporting performance results of cardiac

rhythm and ST segment measurement algorithms.

Bayoumy, K., Gaber, M., Elshafeey, A., Mhaimeed, O.,

Dineen, E. H., Marvel, F. A., Martin, S. S., Muse,

E. D., Turakhia, M. P., Tarakji, K. G., and Elshazly,

M. B. (2021). Smart wearable devices in cardiovascu-

lar care: where we are and how to move forward. Nat

Rev Cardiol., 18:581–599.

Celin, S. and Vasanth, K. (2018). Ecg signal classiﬁcation

using various machine learning techniques. J Med

Syst, 42.

da Silva, L. F., Queiroz, J. A., Vanessa, C., Barros, A. K.,

Lopes, G. C., and Cabral, L. (2021). Separation

BIOSIGNALS 2023 - 16th International Conference on Bio-inspired Systems and Signal Processing

224

method of atrial ﬁbrillation classes with high order

statistics and classiﬁcation using machine learning. In

BIOSIGNALS.

de Albuquerque, V. H. C., Nunes, T. M., Pereira, D. R.,

da S. Luz, E. J., Menotti, D., Papa, J. P., and Tavares,

J. M. R. S. (2018). Robust automated cardiac arrhyth-

mia detection in ecg beat signals. Neural Comput &

Applic, 29:679–693.

Fuadah, Y. N. and Lim, K. M. (2022). Optimal classiﬁcation

of atrial ﬁbrillation and congestive heart failure using

machine learning. Frontiers in Physiology, 12.

Ganapathy, N., Baumg

artel, D., and Deserno, T. M. (2021).

Automatic detection of atrial ﬁbrillation in ecg us-

ing co-occurrence patterns of dynamic symbol assign-

ment and machine learning. Sensors, 21(10).

Goldberger, A., Amaral, L., Glass, L., JM, H., Ivanov,

P., Mark, R., Mietus, J., Moody, G., Peng, C.,

and Stanley, H. (2000). Physiobank, physiotoolkit,

and physionet: components of a new research re-

source for complex physiologic signals. Circulation.,

101(23):E215–20.

Hassan, S. U., Zahid, M. S. M., Abdullah, T. A., and

Husain, K. (2022). Classiﬁcation of cardiac ar-

rhythmia using a convolutional neural network and

bi-directional long short-term memory. DIGITAL

HEALTH, 8:20552076221102766.

Horoba, K., Czabanski, R., Wrobel, J., Matonia, A., Mar-

tinek, R., Kupka, T., Kahankova, R., Leski, J. M., and

Graczyk, S. (2019). Recognition of atrial ﬁbrilation

episodes in heart rate variability signals using a ma-

chine learning approach. In 2019 MIXDES - 26th In-

ternational Conference ”Mixed Design of Integrated

Circuits and Systems”, pages 419–424.

Humphreys, M., Warlow, C., and McGowan, J. (2011). Ar-

rhythmias and their Management, chapter 10, pages

132–155. John Wiley & Sons, Ltd.

Irfan, S., Anjum, N., Althobaiti, T., Alotaibi, A. A., Sid-

diqui, A. B., and Ramzan, N. (2022). Heartbeat classi-

ﬁcation and arrhythmia detection using a multi-model

deep-learning technique. Sensors, 22(15).

Kuila, S., Dhanda, N., and Joardar, S. (2020). Fea-

ture extraction of electrocardiogram signal using ma-

chine learning classiﬁcation. International Journal

of Electrical and Computer Engineering (IJECE),

10(6):6598–6605.

Mkoko, P., Bahiru, E., Ajijola, O. A., Bonny, A., and

Chin, A. (2020). Cardiac arrhythmias in low- and

middle-income countries. Cardiovasc Diagn Ther.,

10(2):350–360.

Mohebbanaaz, Kumari, L. V. R., and Sai, Y. P. (2022). Clas-

siﬁcation of ecg beats using optimized decision tree

and adaptive boosted optimized decision tree. Signal,

Image and Video Processing, 16:695–703.

Mohebbanaaz, Sai, Y. P., and kumari, L. R. (2020). A

review on arrhythmia classiﬁcation using ecg sig-

nals. In 2020 IEEE International Students’ Confer-

ence on Electrical,Electronics and Computer Science

(SCEECS), pages 1–6.

Moody, G. B. and Mark, R. G. (2001). The impact of the

mit-bih arrhythmia database. IEEE Engineering in

Medicine and Biology Magazine., 20(3):45–50.

Mustaqeem, A., Anwar, S. M., and Majid, M. (2018). Mul-

ticlass classiﬁcation of cardiac arrhythmia using im-

proved feature selection and svm invariants. Com-

putational and Mathematical Methods in Medicine,

2018:7310496.

Ramesh, J., Solatidehkordi, Z., Aburukba, R., and Sagahy-

roon, A. (2021). Atrial ﬁbrillation classiﬁcation with

smart wearables using short-term heart rate variabil-

ity and deep convolutional neural networks. Sensors,

21(21).

Sager, S., Bernhardt, F., Kehrle, F., Merkert, M., Potschka,

A., Meder, B., Katus, H., and Scholz, E. (2021).

Expert-enhanced machine learning for cardiac ar-

rhythmia classiﬁcation. PLoS One, 16(12):e0261571.

Shalini, B., Nandini, V., Sandhya, M., and R, B. (2019).

Prediction and classiﬁcation of cardiac arrhythmia.

International Research Journal of Engineering and

Technology (IRJET), 6(6):572–576.

Singh, V., Tewary, S., Sardana, V., and Sardana, H. K.

(2019). Arrhythmia detection - a machine learning

based comparative analysis with mit-bih ecg data. In

2019 IEEE 5th International Conference for Conver-

gence in Technology (I2CT), pages 1–5.

WHO (2021). Cardiovascular diseases (cvds). fact sheets.

Available online. https://www.who.int/en/news-

room/fact-sheets/detail/cardiovascular-diseases-

(cvds). Accessed Oct/2022.

Zhang, J., Liu, A., Gao, M., Chen, X., Zhang, X., and

Chen, X. (2020). Ecg-based multi-class arrhythmia

detection using spatio-temporal attention-based con-

volutional recurrent neural network. Artiﬁcial Intelli-

gence in Medicine, 106:101856.

Arrythmia Classiﬁcation Using MATLAB

 Classiﬁcation Learner App

225