Towards a Decision Support System for Disorders of the

Cardiovascular System

Diagnosing and Evaluation of the Treatment Efficiency

Anton Dolganov and Vladimir Kublanov

Ural Federal University, Mira 19, 620002, Yekaterinburg, Russian Federation

Keywords: Decision Support, Machine Learning, Feature Selection, Arterial Hypertension, Heart Rate Variability.

Abstract: The study describes a preliminary stage of the decision support system development for cardiovascular

system disorders. As the clinical model of the disorders, the arterial hypertension was used. The study

consisted of two steps: diagnosing of the arterial hypertension and an evaluation of the treatment efficiency

during the neuro-electrostimulation application. For the diagnosing part, a clinical study was conducted

involving heart rate variability signals recording while performing tilt-test functional load. Performance of

different machine learning techniques and feature selection strategies in task of binary classification

(healthy volunteers and patients suffering from arterial hypertension) were compared. The genetic

programming feature selection and quadratic discriminant analysis classifier reached the highest

classification accuracy. Best feature combinations were used to evaluate a treatment efficiency. The results

indicate the potential of the proposed decision support system.

1 INTRODUCTION

When human organism is normal, the cardiovascular

system function as coherent whole. One can

highlight etiological factors among variety of

reasons causing disorders of the cardiovascular

system normal functioning. These factors mainly

affect the vascular wall, changing its structure and

causing disorder of the vascular tone. The vascular

tone is essential for organism adaptation to the

constantly changing environment(Mohr et al., 2011).

Disorders of the cardiovascular system can be

subtle for a rather long time. However, abruptly they

can lead to the acute impairments. According to the

World Health Organization data, the heart failure

and insult remain the leading causes of death in the

world (Mendis et al., 2011). Therefore, the task of

the early pre-clinical express diagnosing is relevant.

Among the challenges in the diagnosing of the

cardiovascular system disorders are the situational

changes, like stress.

Thus it is apropriate to develop a objective

decision support system. In present work arterial

hypertension is used as a clinical model for

cardiovascular system disorders. The following tasks

are considered: application of the machine learning

techniques for diagnosing of the arterial

hypertension and evaluation of the neuro-

electrostimulation treatment efficiency.

2 CLINICAL MODEL OF THE

STUDY

As noted in World Health Organization (WHO,

2013) arterial hypertension is among the most

frequent cardio-vascular pathologies and occurs in

around 15-20% of the elderly population. The

hypertension is considered to be among the most

prominent factors of the heart failure, insult and

coronary failure. Therefore arterial hypertension is

suitable clinical model of the cardiovascular

disorder.

The nature of the arterial hypertension is

multifactorial. The hypertension arises as the

disorder of the vascular wall tone. The vascular

tone’s most important feature is the arterial pressure.

This is, in turn controlled by the regulatory

mechanisms within the autonomic nervous system

(ANS) (Kseneva et al., 2016).

The heart rate variability (HRV) is one of the

indirect means of the ANS monitoring. The heart

Dolganov A. and Kublanov V.

Towards a Decision Support System for Disorders of the Cardiovascular System - Diagnosing and Evaluation of the Treatment Efﬁciency.

DOI: 10.5220/0006753407270733

In Proceedings of the 11th International Joint Conference on Biomedical Engineering Systems and Technologies (HEALTHINF 2018), pages 727-733

ISBN: 978-989-758-281-3

rate varies from beat to beat. This is caused by the

constant adaptation processes, which are launched

by the ANS to keep balance of the cardiovascular

system. The HRV reflects the functioning of the

cardiovascular system and regulatory mechanisms of

the organism, as well as the ratio between

sympathetic and parasympathetic departments of the

ANS. Changes of the HRV features can be

preemptive indicators of the health disorders

(Ahmad et al., 2009).

It is worthy to stress out evaluation of the HRV

features during the functional loads. This has a range

of advantages as it allows minimizing the personal

differences and estimating the direction of the

changes. As an example of the functional load one

can consider the tilt-test(Ducla-Soares et al., 2007).

The tilt-test is an experimental way to test organism

reaction on the crossover from the horizontal

position to the vertical one (head up). Usually the

rotating table is used for this load. Among the

benefits of the rotating table application is increased

sensitivity, improved reproducibility of conditions

and results, safety and possibility to control the

angle of the rotation as well as its speed. Finally it

reduces the noise of the biomedical signals

registration, as it prevents active body movement

(Turk et al., 2010).

The organism reaction to the tilt-test is well

studied for both, healthy persons and in case of the

some pathologies. As studied are the systems that

are activated as the response of the cardiovascular

system. Therefore one can compare changes of the

HRV during the tilt-test with the well-known

physiological reactions of the cardiovascular system.

3 CLINICAL STUDY

DESCRIPTION

The pilot clinical study was approved by the local

Ethics Committee in Ural State Medical University,

Yekaterinburg, Russian Federation (protocol № 8

from 15 October 2015). 68 people have participated

in the study: 28 healthy people and 40 patients. All

the patients were diagnosed with the II/III degree of

arterial hypertension. All participants were

volunteers, listened to the detailed explanation of the

study stages and had signed the written participation

consent. The clinical study itself took place at the

Sverdlovsk Clinical Hospital of Mental Diseases for

Military Veterans (Yekaterinburg, Russian

Federation).

The following exclusion criteria were considered

for the patients: liver, respiratory or kidney failure,

diabetes of I type, diffusion collagen disease, heart

failure of III-IV class (by the NYHA classification),

acute impairment of cerebral circulation (6 month

prior to the study), unstable angina or myocardial

infarction (6 month prior to the study), permanent

atrial fibrillation. Women during pregnancy and

lactation period were also not included in the

participation group.

Patients suffering from arterial were taking

standard pharmacological therapy –angiotensin-

converting-enzyme inhibitor, calcium channel

blockers, and diuretics in the medium therapeutic

doses. As an addition to the pharmacological therapy

the methods of the physiotherapy can apply which

are directed to the normalization of the autonomic

regulation and improvement of the cardiovascular

system functioning. The SYMPATHOCOR-01

neuro-electrostimulator (registration certificate №

FCR 2007/00757) is one of such devices. The neuro-

electrostimulator can control the vascular tone by

means of the sympathetic nervous system correction

(Petrenko et al., 2017).

During the study the electrocardiography (ECG)

signals were recorded by the electroencephalograph-

analyzer “Encephalan-131-03” (“Medicom-MTD”,

Taganrog, Russian Federation) in the first limb lead

(Kleiger et al. 2005). Afterwards the ECG signal

recording the “Encephalan-131-03” software

automatically derives the HRV signals.

The clinical data was recorded in three functional

states involving the rotating table Lojer (Vammalan

Konepaja DY, Finland). During the first state the

participants were calmly lying on the exanimating

table (state F). At the second state the tilt-test was

performed – the head end of the table is lifted up to

from the horizontal position (state O). At the

final state the participant returns to the horizontal

position (state K). The duration of the signal record

in each state was 300 seconds. The whole study was

supervised by a physician.

4 METHODS OF DATA

PROCESSING

4.1 Heart Rate Variability Features

That list of 64 HRV features in this study consisted

of time-domain and frequency-domain features

established by the European Society of Cardiology

(Malik, 1996; Tarvainen et al., 2014) as well as

relevant non-linear features (Sivanantham and

Shenbaga Devi, 2014). In our study, in addition to

commonly used features, the wavelet transform

features were used (Egorova et al., 2014). The

detailed list of the features was described in our

previous works on this topic (Vladimir Kublanov et

al., 2017).

Even though getting as much features from the

signal is vital, one should also consider proper

feature selection to avoid using redundant data

(Thangavel and Pethalakshmi, 2009). In this work

HRV features in three functional states were used

for the diagnosing of the arterial hypertension. In

particular the classification task was solved by

means of different machine learning techniques.

4.2 Machine Learning Techniques

The aim of the study was to evaluate variety of

machine learning techniques which are based on the

different core principles. In particular, the

considered machine learning techniques included:

 Linear and Quadratic Discriminant Analysis

(LDA and QDA) – are based on the finding the

separating hyperplane, either linear or

quadratic(Cacoullos, 2014);

 k-Nearest Neighbors (k-NN) – analyzing the

similarity between the closet objects in the

training subset (Peterson, 2009);

 Decision Trees (DT) – representation of the

classification rule as an hierarchical sequence of

Boolean blocks «if…then…» (Rokach and

Maimon, 2008)

 Naïve Bayes classifier (NB) – taking into

account posterior probabilities of the dataset,

with additional assumptions of the independency

of the features (Ng and Jordan, 2002).

As different machine learning techniques are

involved one have to use similar metric to compare

classification efficacy. For this, the leave-one-out

cross-validation (LOOCV) metric was used.

LOOCV implies using one of the observations in the

dataset as the test set, while the remaining data is

used as the training set. This procedure is repeated

for all the observations in the dataset. Application of

LOOCV procedure tends to minimize overfitting as

the efficacy is estimated on the external data,

previously not shown to the classifier (Zhang and

Yang, 2015).

4.3 Feature Selection Strategies

The principal component analysis (PCA) procedure

is commonly used to reveal internal structure of the

data by converting original features into set of

principal components – linearly uncorrelated

features combinations. This is a dimension reduction

technique, as it allows to project original features

dataset into a smaller number of principal

components. The first principal component is

selected in a way that it has the highest possible

variance of the dataset (Jolliffe, 2002).

The exhaustive search is guaranteed to find the

most optimal solution, in our case combination of

features with the highest classification accuracy.

However there is obvious limitation – time to

evaluate all the possible combinations. As example

the number of 5-combinations in 64 features set

extends 7 million. At the same time number of 5-

combinations in 128 features set (features of two

different states used simultaneously) is over 250

million. In order to reduce the search space it was

suggested to consider only such combinations that

are formed by features with low correlation

coefficient (less than 0.25). It is acceptable

restriction for HRV features, as part of them are

duplicates in the mathematic aspects as well as in the

biological interpretation (V. S. Kublanov et al.,

2017).

The greedy search algorithm core principle is to

take on each iteration the most optimal decision

(Ruiz and Stützle, 2007). In our study at the first

step, classification efficiency of all features is

evaluated separately. The feature with the highest

classification accuracy is selected. On the second

step, combinations of the selected feature with the

remaining ones are evaluated. The combination with

the highest classification accuracy is selected. The

algorithm is repeated until all features are picked on

the classification accuracy decreases.

As alternative to the exhaustive search the

genetic programming (GP) can be used. This

heuristic approach - involves application of the

Darwin’s evolutionary strategies for improvement of

the task’s solution (Koza, 1992). In this work the

binary encoding was picked. The ratio between main

genetic operations (copy, crossover, mutation) was

1:2:7. The initial population was 100 randomly

picked 3-combinations of the non-correlated

features. Maximal number of generations – 20.

Overall the evolution was repeated for 50 times (V.

Kublanov et al., 2017).

5 RESULTS

5.1 Arterial Hypertension Diagnosing

In the study the classification accuracy of two

groups – healthy and patients suffering from arterial

hypertension was evaluated for each of the machine

learning technique. Table 1 presents maximal

classification accuracy, reached by each of the

classifiers for different feature selection strategies. It

is worthy to point out, that different datasets were

considered: features of the single functional state (F,

O or K), features in two functional states (F-O, F-K,

O-K) and features in three functional states (F-O-K).

Table 1: Maximal Classification Accuracy, %.

Feature Selection LDA QDA 3-NN 4-NN 5-NN DT NB

ll features 73.5 66.2 76.5 79.4 79.4 76.5 70.6

PCA 86.8 83.8 83.8 82.4 86.8 77.9 76.5

Non-correlated

pace

89.7 91.2 91.2 89.7 91.2 95.6 92.6

Greedy algorithm 94.1 95.6 91.2 91.2 92.6 94.1 94.1

Genetic

rogramming

95.6 98.5 91.2 91.2 92.6 97.0 97.1

In case when all features were used the maximal

accuracy was achieved by NN classifiers when F-O

features were used. It is worthy to point out, that all

highest results were obtained in those cases, when

features of state O were included. However, the

highest classification accuracy is capped at 79,4 in

this case. Therefore it is appropriate to optimize

inclusion of the features.

When PCA was used the classification accuracy

was evaluated using combinations of first 10

principal components. The explained variance was

not less than 0.80. Similarly, high classification

accuracy was reached, when features of state O were

involved. The highest results were obtained by the

LDA classifier, using 9 first principles components

of features O-K; by the 5-NN classifier, using first 8

components of state O.

The semi-optimal search on the non-correlated

space was limited due to number of combinations.

Therefore, only combinations including up to 5

features were considered. Again, the highest

classification accuracy was associated with the state

O (around 90 %). When features of states F and K

were used separately, the results were comparable to

those of PCA. Overall, the DT classifier with the

following combination F EnInterp, O HFn f, K

fVLFmax, K SDLF, reached the highest

classification accuracy. Here EnInterp is the

Shannon Entropy of interpolated time-series, HFn f

– normalized spectral power of High Frequency

band, evaluated by means of the Fast-Fourier

transform, VLFmax – maximum of the Very Low

Frequency spectral power, SDLF – standard

deviation of the Low Frequency time-series derived

by the wavelet transform. Even though, the accuracy

improved the question was open – are these the best

possible results or they can be improved by adding

additional features?

As can be noted from the results in table 1,

application of the greedy algorithm allows

improving results of the LDA, QDA and NB, due to

increase of the used features. For example, the QDA

classification accuracy improved up to the 95.6%

when using 6 features of the F-O. However, this

approach does not guarantee the optimal solution, as

results of the DT classifiers have become slightly

worse. Because of that, the genetic programming

paradigm was used.

Data in last row in table 1 show that application

of the genetic programming allowed to improve

classification accuracy for all machine learning

techniques but nearest neighbors. The highest

accuracy – 98.5% – was reached by the QDA

classifier. Actually, four combinations reached such

accuracy. They are presented in table 2. These

combinations were selected for evaluation of the

treatment process efficiency.

Table 2: Best Features Combinations for Arterial

Hypertension Diagnosing.

Combination id Features

QDA-1

F SI F EnInterp O kurtosis O ZCR O LF/HF f

O RF O f(LFmax) O f(VLFmax) O LFn wt K HR K

f(LFmax) K VLFmax K EnHF K EnVLF

QDA-2

F SI F EnInterp O HR O kurtosis O LF/HF f O RF O

f(VLFmax) O LFn wt O EnVLF

f(LFmax)

VLFmax

HF wt

EnHF

EnVLF

QDA-3

F SI F EnInterp O kurtosis O ZCR O LF/HF f

O RF O f(LFmax) O f(VLFmax) O LFn wt

O EnVLF K f(LFmax) K VLFmax K EnHF K

EnVLF

QDA-4

F SI F EnInterp O kurtosis O ZCR O LF/HF f

O RF O f(LFmax) O f(VLFmax) O LFn wt

O EnVLF K HR K f(LFmax) K VLFmax K EnHF K

EnVLF

In table 2, SI, is the Stress Index; ZCR is the

zero-crossing rate (in relation to the mean of the R-R

time series); LF/HF – ratio of the Low and High

Frequency spectral bands; f(LFmax) and f(VLFmax)

are the frequencies, corresponding to the maximums

of the Low and Very Low Frequency spectral power

respectively; LFn normalized spectral power of High

Frequency band; HR is Heart Rate, EnHF and

EnVLF are the entropies of the High and Very Low

Frequency time-series derived by the wavelet

transform; RF is the respiration frequency

(frequency, corresponding to the maximum of the

High Frequency spectral power). The f affix denotes

features evaluated by means of the Fast-Fourier

transform; wt affix denotes features evaluated by

means of the wavelet transform.

5.2 Evaluation of the Treatment

Efficiency

In order to evaluate efficiency of the treatment

process HRV features of eight patients were

analyzed. Data of these patients was not previously

used for the evaluations in chapter 5.1. The HRV

signals were registered in accordance with the

description, presented in chapter 3. Biomedical

signals were registered several times: the initial

registration; after a single procedure of the

SYMPATHOCOR-01 device neuro-

electrostimulation; after five procedures of the

SYMPATHOCOR-01 device neuro-

electrostimulation. The neuro-electrostimulation

procedures were performed in accordance with the

dynamic correction of the sympathetic nervous

system methodology (Petrenko et al., 2015).

First of all, it is worthy to point out that all eight

patients were classified by the combinations

QDA-(1-4) as ones, suffering from arterial

hypertension. This was in line with the physician

diagnosing. After that dynamics of the distance, until

the separating hyperplane, was analyzed. The

estimates were compared with data of the arterial

pressure changes.

In order to evaluate arterial pressure not less than

two measurements on each arm was done. Interval

between each measurement was not less than 2

minutes. If there was difference more than 5 mmHg

than additional, measurement took place. The lowest

measurement was taken as the final one (Mancia et

al., 2013).

Table 3 presents correlation coefficients of the

systolic and diastolic arterial pressure (Ads and

ADd), with the distance, until the separating

hyperplane, evaluated using features combinations

QDA-(1-4).

Table 3: Correlation of the HRV Features Prediction with

the Arterial Pressure Dynamic.

QDA-1 QDA -2 QDA -3 QDA -4

ADs 0,714 0,700 0,708 0,707

ADd 0,798 0,779 0,795 0,795

Presented in table 3 data has high degree of

significance (p-value <0,0001). The obtained results

highlights agreement between HRV features and

arterial pressure. The actual dynamic of the

evaluations shows a tendency toward the healthy

people. This was in line with physician comments

and subjective feeling of the patients.

6 DISCUSSION

Development of the decision support system is the

emerging area. There are works that use

anthropometric data to classify arterial hypertension.

However, despite having relatively high accuracy

there are some limitations. As we think the most

significant is limited possibility to quantify

dynamics of the physiological changes during the

treatment (Pytel et al., 2015).

Systems, which are based only on the data of

arterial pressure, also have limitations. The single

measurement of the arterial pressure does not always

correctly reflect current functional state. Features of

the arterial pressure can be distorted by the stress

situations (including the white coat hypertension) or

circadian rhythms.

Heart rate variability features are less susceptible

to such changes, as they are, essentially averaged

over a relatively long time interval. In addition,

application of the functional load allows to ‘direct’

functional changes. Lastly the proposed procedure

allows preventing stress situations, as it is in fact, 15

minutes of lying.

Within the study, it was multiple times noted that

features of the state O (tilt-test), have the higher

classification possibilities. This fact confirms

feasibility of such functional load application.

In present works, pilot results of the neuro-

electrostimulation device SYMPATHOCOR-01

application were obtained. The device can be used as

the physiotherapeutic method for treatment of the

arterial hypertension. However, in order to

accurately evaluate applicability of the device one

has to conduct full-scale clinical double-blind

studies. This is not within the tasks of present study.

The limitation of the present study can be

relatively small clinical data sample. However,

preliminary results of the study on data, not used for

the classifiers training, demonstrate potential of the

proposed system. In future studies, our research

group is interested in conductance of the additional

studies on a bigger data sample. Moreover it is

possible to improve results by means of bagging,

boosting and stacking as it was shown in .

7 CONCLUSIONS

The present work described initial steps of the

decision support system development for

cardiovascular system disorders. The arterial

hypertension was used as the clinical model.

At first, the diagnosing accuracy was evaluated.

For that heart rate variability signal were registered

during the tilt-test functional load. The heart rate

variability is one of the indirect means to assess

functioning of the autonomic nervous system,

which, in turn, is essential in the pathogenesis of the

arterial hypertension

Possibilities of different machine learning

techniques were analyzed, in particular linear and

quadratic discriminant analysis, k-nearest neighbors,

decision trees and Naïve Bayes classifier. Various

feature selection techniques were tested: principal

component analysis, semi-optimal search on non-

correlated features space, greedy algorithm and

genetic programming. It was noted that the genetic

programming feature selection and quadratic

discriminant analysis classifier reached the highest

classification accuracy.

Best feature combinations were used to evaluate

treatment efficiency during the neuro-

electrostimulation by the SYMPATHOCOR-01

device. The results highlight significant agreement

of the heart rate variability with the arterial pressure

data.

The accumulated during the present study

groundwork will become a basis for a decision

support system for disorders of the cardiovascular

system.

ACKNOWLEDGEMENTS

The work was supported by Act 211 Government of

the Russian Federation, contract № 02.A03.21.0006.

REFERENCES

Ahmad, S., Tejuja, A., Newman, K.D., Zarychanski, R.,

Seely, A.J., 2009. Clinical review: A review and

analysis of heart rate variability and the diagnosis and

prognosis of infection. Crit. Care 13, 232.

Cacoullos, T., 2014. Discriminant analysis and

applications. Academic Press.

Ducla-Soares, J.L., Santos-Bento, M., Laranjo, S.,

Andrade, A., Ducla-Soares, E., Boto, J.P., Silva-

Carvalho, L., Rocha, I., 2007. Wavelet analysis of

autonomic outflow of normal subjects on head-up tilt,

cold pressor test, Valsalva manoeuvre and deep

breathing. Exp. Physiol. 92, 677–686.

Egorova, D.D., Kazakov, Y.E., Kublanov, V.S., 2014.

Principal Components Method for Heart Rate

Variability Analysis. Biomed. Eng. 48, 37–41.

Jolliffe, I., 2002. Principal component analysis. Wiley

Online Library.

Koza, J.R., 1992. Genetic programming: on the

programming of computers by means of natural

selection. MIT press.

Kseneva, S.I., Borodulina, E.V., Trifonova, O.Y., Udut,

V.V., 2016. Cardiac Autonomic Drive during Arterial

Hypertension and Metabolic Disturbances. Bull. Exp.

Biol. Med. 161, 237–240.

Kublanov, V., Dolganov, A., Gamboa, H., 2017. Genetic

programming application for features selection in task

of arterial hypertension classification. In: Proceedings.

Presented at the 2017 International Multi-Conference

on Engineering, Computer and Information Sciences

(SIBIRCON), IEEE, Novosibirsk Akademgorodok,

Russia, pp. 561–565.

Kublanov, V., Dolganov, A., Kazakov, Y., 2017.

Diagnostics of the Arterial Hypertension by Means of

the Discriminant Analysis Analysis of the Heart Rate

Variability Signals Features Combinations. Proc. 10th

Int. Jt. Conf. Biomed. Eng. Syst. Technol. Vol 4

Biosignals 291–298.

Kublanov, V.S., Dolganov, A.Y., Belo, D., Gamboa, H.,

2017. Comparison of Machine Learning Methods for

the Arterial Hypertension Diagnostics. Appl. Bionics

Biomech. 2017.

Malik, M., 1996. Heart rate variability: Standards of

measurement, physiological interpretation, and clinical

use. Circulation 93, 1043–1065.

Mancia, G., Fagard, R., Narkiewicz, K., Redon, J.,

Zanchetti, A., Böhm, M., Christiaens, T., Cifkova, R.,

Backer, G.D., Dominiczak, A., Galderisi, M.,

Grobbee, D.E., Jaarsma, T., Kirchhof, P., Kjeldsen,

S.E., Laurent, S., Manolis, A.J., Nilsson, P.M.,

Ruilope, L.M., Schmieder, R.E., Sirnes, P.A., Sleight,

P., Viigimaa, M., Waeber, B., Zannad, F., 2013. 2013

ESH/ESC Guidelines for the management of arterial

hypertension. Eur. Heart J. 34, 2159–2219.

Mendis, S., Puska, P., Norrving, B., 2011. Global atlas on

cardiovascular disease prevention and control. World

Health Organization.

Mohr, J.P., Grotta, J.C., Wolf, P.A., Moskowitz, M.A.,

Mayberg, M.R., Von Kummer, R., 2011. Stroke E-

Book: Pathophysiology, Diagnosis, and Management.

Elsevier Health Sciences.

Ng, A.Y., Jordan, M.I., 2002. On discriminative vs.

generative classifiers: A comparison of logistic

regression and naive bayes. In: Advances in Neural

Information Processing Systems. pp. 841–848.

Peterson, L.E., 2009. K-nearest neighbor. Scholarpedia 4,

1883.

Petrenko, T., Kublanov, V., Retyunskiy, K., 2017. The

role of neuroplasticity in the treatment of cognitive

impairments by means multifactor neuro-

electrostimulation of the segmental level of the

autonomic nervous system. Eur. Psychiatry 41, S770.

Petrenko, T.S., Kublanov, V.S., Retiunskiy, K.Y., 2015.

Dynamic Correction of the Activity Sympathetic

Nervous System (Dcasns) to Restore Cognitive

Functions. Eur. Psychiatry 30, 843.

Pytel, K., Nawarycz, T., Ostrowska-Nawarycz, L., Drygas,

W., 2015. Anthropometric predictors and Artificial

Neural Networks in the diagnosis of hypertension. In:

2015 Federated Conference on Computer Science and

Information Systems (FedCSIS). Presented at the 2015

Federated Conference on Computer Science and

Information Systems (FedCSIS), pp. 287–290.

Rokach, L., Maimon, O., 2008. Data mining with decision

trees: theory and applications. World Scientific.

Ruiz, R., Stützle, T., 2007. A simple and effective iterated

greedy algorithm for the permutation flowshop

scheduling problem. Eur. J. Oper. Res. 177, 2033–

2049.

Sivanantham, A., Shenbaga Devi, S., 2014. Cardiac

arrhythmia detection using linear and non-linear

features of HRV signal. In: 2014 International

Conference on Advanced Communication Control and

Computing Technologies (ICACCCT). pp. 795–799.

Tarvainen, M.P., Niskanen, J.-P., Lipponen, J.A., Ranta-

Aho, P.O., Karjalainen, P.A., 2014. Kubios HRV–

heart rate variability analysis software. Comput.

Methods Programs Biomed. 113, 210–220.

Thangavel, K., Pethalakshmi, A., 2009. Dimensionality

reduction based on rough set theory: A review. Appl.

Soft Comput. 9, 1–12.

Turk, U., Alioglu, E., Kirilmaz, B., Duygu, H., Tuzun, N.,

Tengiz, I., Zoghi, M., Ercan, E., 2010. Prediction of

HeadUp Tilt Test Result: Is it Possible? Pacing Clin.

Electrophysiol. 33, 153–158.

WHO, 2013. A global brief on hypertension: silent killer,

global public health crisis: World Health Day 2013.

Zhang, Y., Yang, Y., 2015. Cross-validation for selecting

a model selection procedure. J. Econom. 187, 95–112.