Information Technology for Classification of Donosological and Pathological States Using the Ensemble of Data Mining Methods
The purpose of the paper is to develop information technology for the classification of human health states using an set of Data Mining methods and to carry out its validation on examples of a operators` functional state and patient's disease severity. Results. The developed IT unites several s...
Збережено в:
| Дата: | 2021 |
|---|---|
| Автори: | , |
| Формат: | Стаття |
| Мова: | English |
| Опубліковано: |
Міжнародний науково-навчальний центр інформаційних технологій і систем НАН України та МОН України
2021
|
| Назва видання: | Cybernetics and computer engineering |
| Теми: | |
| Онлайн доступ: | https://nasplib.isofts.kiev.ua/handle/123456789/181418 |
| Теги: |
Додати тег
Немає тегів, Будьте першим, хто поставить тег для цього запису!
|
| Назва журналу: | Digital Library of Periodicals of National Academy of Sciences of Ukraine |
| Цитувати: | Information Technology for Classification of Donosological and Pathological States Using the Ensemble of Data Mining Methods / O.A. Kryvova, L.M. Kozak // Cybernetics and computer engineering. — 2021. — № 1 (203). — С. 77-96. — Бібліогр.: 44 назв. — англ. |
Репозитарії
Digital Library of Periodicals of National Academy of Sciences of Ukraine| id |
nasplib_isofts_kiev_ua-123456789-181418 |
|---|---|
| record_format |
dspace |
| spelling |
nasplib_isofts_kiev_ua-123456789-1814182025-02-09T22:22:29Z Information Technology for Classification of Donosological and Pathological States Using the Ensemble of Data Mining Methods Інформаційна технологія класифікації донозологічних та патологічних станів здоров’я з використанням ансамблю методів Data Mining Kryvova, O.A. Kozak, L.M. Medical and Biological Cybernetics The purpose of the paper is to develop information technology for the classification of human health states using an set of Data Mining methods and to carry out its validation on examples of a operators` functional state and patient's disease severity. Results. The developed IT unites several stages: I — data pre-processing; II — clustering, selecting the homogeneous groups (data segmentation); III — predictors` identification; IV — classifying the studied states, development of predictive models using machine learning algorithms (Decision trees, Support vector machines, neural networks) and the method crossvalidation. The proposed IT was used to classify the operators` functional statе and the patients` severity in case of disease progression. Метою роботи є розроблення інформаційної технології класифікації стану здоров'я людини за допомогою комплексу методів Data Mining за об'єктивними та експертними характеристиками. Результати. Розроблена інформаційна технологія об'єднує кілька етапів: I — попереднє оброблення даних; II — кластеризація, вибір однорідних груп (сегментація даних); III — ідентифікація предикторів; IV — класифікація досліджуваних станів, розроблення прогнозних моделей за допомогою алгоритмів машинного навчання (дерев рішень (Decision trees, опорних векторних машин Support vector machine, нейронних мереж) та методу перевірки навчальної вибірки (cross-validation). Запропоновану ІТ використано для дослідження функційного стану операторів та класифікації тяжкості стану пацієнтів у разі прогресування захворювання. 2021 Article Information Technology for Classification of Donosological and Pathological States Using the Ensemble of Data Mining Methods / O.A. Kryvova, L.M. Kozak // Cybernetics and computer engineering. — 2021. — № 1 (203). — С. 77-96. — Бібліогр.: 44 назв. — англ. 2663-2578 DOI: https://doi.org/10.15407/kvt203.01.077 https://nasplib.isofts.kiev.ua/handle/123456789/181418 004.75+004.932.2:616 en Cybernetics and computer engineering application/pdf Міжнародний науково-навчальний центр інформаційних технологій і систем НАН України та МОН України |
| institution |
Digital Library of Periodicals of National Academy of Sciences of Ukraine |
| collection |
DSpace DC |
| language |
English |
| topic |
Medical and Biological Cybernetics Medical and Biological Cybernetics |
| spellingShingle |
Medical and Biological Cybernetics Medical and Biological Cybernetics Kryvova, O.A. Kozak, L.M. Information Technology for Classification of Donosological and Pathological States Using the Ensemble of Data Mining Methods Cybernetics and computer engineering |
| description |
The purpose of the paper is to develop information technology for the classification of human health states using an set of Data Mining methods and to carry out its validation on examples of a operators` functional state and patient's disease severity. Results. The developed IT unites several stages: I — data pre-processing; II — clustering, selecting the homogeneous groups (data segmentation); III — predictors` identification; IV — classifying the studied states, development of predictive models using machine learning algorithms (Decision trees, Support vector machines, neural networks) and the method crossvalidation. The proposed IT was used to classify the operators` functional statе and the patients` severity in case of disease progression. |
| format |
Article |
| author |
Kryvova, O.A. Kozak, L.M. |
| author_facet |
Kryvova, O.A. Kozak, L.M. |
| author_sort |
Kryvova, O.A. |
| title |
Information Technology for Classification of Donosological and Pathological States Using the Ensemble of Data Mining Methods |
| title_short |
Information Technology for Classification of Donosological and Pathological States Using the Ensemble of Data Mining Methods |
| title_full |
Information Technology for Classification of Donosological and Pathological States Using the Ensemble of Data Mining Methods |
| title_fullStr |
Information Technology for Classification of Donosological and Pathological States Using the Ensemble of Data Mining Methods |
| title_full_unstemmed |
Information Technology for Classification of Donosological and Pathological States Using the Ensemble of Data Mining Methods |
| title_sort |
information technology for classification of donosological and pathological states using the ensemble of data mining methods |
| publisher |
Міжнародний науково-навчальний центр інформаційних технологій і систем НАН України та МОН України |
| publishDate |
2021 |
| topic_facet |
Medical and Biological Cybernetics |
| url |
https://nasplib.isofts.kiev.ua/handle/123456789/181418 |
| citation_txt |
Information Technology for Classification of Donosological and Pathological States Using the Ensemble of Data Mining Methods / O.A. Kryvova, L.M. Kozak // Cybernetics and computer engineering. — 2021. — № 1 (203). — С. 77-96. — Бібліогр.: 44 назв. — англ. |
| series |
Cybernetics and computer engineering |
| work_keys_str_mv |
AT kryvovaoa informationtechnologyforclassificationofdonosologicalandpathologicalstatesusingtheensembleofdataminingmethods AT kozaklm informationtechnologyforclassificationofdonosologicalandpathologicalstatesusingtheensembleofdataminingmethods AT kryvovaoa ínformacíinatehnologíâklasifíkacíídonozologíčnihtapatologíčnihstanívzdorovâzvikoristannâmansamblûmetodívdatamining AT kozaklm ínformacíinatehnologíâklasifíkacíídonozologíčnihtapatologíčnihstanívzdorovâzvikoristannâmansamblûmetodívdatamining |
| first_indexed |
2025-12-01T09:38:29Z |
| last_indexed |
2025-12-01T09:38:29Z |
| _version_ |
1850298242932146176 |
| fulltext |
ISSN 2663-2586 (Online), ISSN 2663-2578 (Print). Cyb. and Comp. Eng. 2021. № 1 (203)
DOI: https://doi.org/10.15407/kvt203.01.077
УДК 004.75+004.932.2:616
KRYVOVA O.A., Researcher,
Medical Information Systems Department
ORCID: 0000-0002-4407-5990
e-mail: ol.kryvova@gmail.com
KOZAK L.M., DSc (Biology), Senior Researcher,
Leading Researcher of the Medical Information Systems Department
ORCID: 0000-0002-7412-3041
e-mail: lmkozak52@gmail.com
International Research and Training Center for Information Technologies
and Systems of the National Academy of Sciences of Ukraine
and Ministry of Education and Science of Ukraine,
40, Acad. Glushkov av., Kyiv, 03187, Ukraine
INFORMATION TECHNOLOGY FOR CLASSIFICATION
OF DONOSOLOGICAL AND PATHOLOGICAL STATES USING
THE ENSEMBLE OF DATA MINING METHODS
Introduction. The digital technologies implementation provides registration of large amounts
of bio-medical data (ECG, EEG, electronic medical records) as a basis for assessing and
predicting the patients` condition. Data Mining methods allow to identify the most informa-
tive indicators and typological groups, to classify the person` functional state and the pa-
tients` disease stages to predict their changes.
The purpose of the paper is to develop information technology for the classification of
human health states using an set of Data Mining methods and to carry out its validation on
examples of a operators` functional state and patient's disease severity.
Results. The developed IT unites several stages: I — data pre-processing; II — cluster-
ing, selecting the homogeneous groups (data segmentation); III — predictors` identification;
IV — classifying the studied states, development of predictive models using machine learning
algorithms (Decision trees, Support vector machines, neural networks) and the method cross-
validation. The proposed IT was used to classify the operators` functional statе and the pa-
tients` severity in case of disease progression.
Conclusions. The IT use to assess the operators` activity successes made it possible to
identify the most informative HRV indicators, changes in which can predict the operators`
reliability, taking into account the type of vegetative regulation. Assessing the disease activity
of children with dysplasia with IT use made it possible to identify diagnostic markers of CCC
and develop diagnostic rules for determining the stages of the disease by ECG parameters
(T wave symmetry, an integral indicator of the ST_T segment shape).
Keywords: information technology, Data Mining, machine learning models, severity of the patient.
© KRYVOVA O.A., KOZAK L.M., 2021
77
Kryvova O.A., Kozak L.M.
ISSN 2663-2586 (Online), ISSN 2663-2578 (Print). Cyb. and Comp. Eng. 2021. № 1 (203) 78
INTRODUCTION
At the current stage of digital medicine development, accompanied by the use of
multifunctional monitoring systems, individual mobile health monitoring de-
vices, there is a problem of interpretation of untreated primary arrays of hetero-
geneous medical data. One approach to solving it is to develop and apply infor-
mation technology using Data Mining methods. The basic definition of Data
Mining is the process of identifying patterns in data arrays (previously unknown)
and using them to predict health states and make decisions [1].
In recent years, more and more researches have been done to improve patients'
health. Multilevel schemes are developed, which use different types and methods of
adaptive learning and combine various sources of clinical information (EHR, labora-
tory data, monitors, medical images). In recent decades, researchers have noted that
the direction of Data Mining application and machine learning methods, namely the
patients’ classification into risk groups to predict treatment outcomes, mortality,
disease stages etc., was formed [2–10]. Analysis of the literature data for 2008–2019
leads to the conclusion that in terms of accuracy and clarity of the results the intel-
lectual analysis methods, which integrate hybrid methods and previous models of
clinical risk stratification, should be prefered [11].
PROBLEM STATEMENT
In the early 2000s, examples of successful use of Data Mining for biomedical
data analysis appeared [2]. The research was mainly aimed at improving the
diagnostic accuracy of the diseases identification by medical databases [3] and at
developing the decision support systems [4] and diferent studies by medical and
biological information [8].
Different types of machine learning methods are used to develop classifica-
tion diagnostic models: logistic regression, decision tree methods, random forest
(RF), support vector machine (SVM) or ensembles of classification models,
genetic algorithms, artificial neural or deep learning networks [4–7].
Among the growing number of works on the application of Data Mining and
machine learning technologies, the trend of the clinical direction of predictive
models, which use new multi-sensory, multi-resource and multiprocessor infor-
mation merging schemes, stands out. The architecture of such systems consists
of hybrid multilevel schemes, combines uncontrolled and controlled teaching
methods and methods of features selection. This approach makes it possible to
identify clinically significant patterns using data of monitoring, clinical meas-
ures, tools and treatment outcomes [9–11].
For almost 30 years, more than twenty classical tools (systems) for assessing
and forecasting the patients’ condition have been developed and updated [12–16].
Among them are severity scores, which quantitatively or qualitatively determine the
severity of the patient's condition and classify him into specific risk group based on
the analysis of deviations of anatomical, physiological, biochemical parameters.
Determining the severity of the condition the decision to hospitalize the patient in
the intensive care unit can be made. For example, in intensive care units in the
United States and the EC use scoring systems to assess the patients’ condition.
These are several scales: Simplified Acute Physiology Score (SAPS II) [14], Acute
Information Technology for Classification of Donosological and Pathological States Using
ISSN 2663-2586 (Online), ISSN 2663-2578 (Print). Cyb. and Comp. Eng. 2021. № 1 (203) 79
Physiology and Chronic Health Evaluations (APACHE II and III) [15], Mortality
Probability Models (MPM II-24) [16]. In addition to standardized scales designed
for the general population, a number of specialized scales have been developed to
assess the activity (stages, severity) of individual diseases.
A number of studies on the stratification of patients into risk groups accord-
ing to clinical data and treatment outcomes have demonstrated the superiority of
models developed by Data Mining methods over classical scales [17–19].
Much attention is paid to the choice of informative characteristics for the
analysis of prenosological and pathological human conditions. Many researchers
have determined that the cardiovascular system (CVC) is one of the main indica-
tors of adaptive capacity and responses of the whole organism [20, 21].
One of the most common methods of studying the mechanisms of regulation
of the cardiovascular system is the analysis of heart rate variability, which has
become a reliable and powerful tool for research in cardiology, assessment of
human functional status (FS) [21]. It is proved that the adaptive reactions of the
heart to constantly changing physiological conditions are reflected in changes in
heart rate variability, which provides information about the systemic reactions of
the body during deteriorating health and under the influence of external stress. It
is the CVS functioning level that can determine the boundary between the pre-
nosological state (health) and the disease, as well as affect the disease severity.
Methods of HRV analysis are being actively developed [21–29], the technology
of analysis is being improved, mathematical approaches to the analysis of nonlinear
dynamics of heart rhythm are involved, which has expanded the list of informative
indicators for assessing the human condition. Currently, studies of this condition
(norm and pathology) are carried out using estimates of irregularity and chaotic
rhythm, such as fractal dimension, entropy parameters [27]. The following approaches
are proposed for use: wavelet transform [28], the method of multispectral analysis of
CVC [29], analysis in the phase plane [23, 24]. Thus, one of the common and effective
approaches to detecting changes in human health is to assess the relationship of this
condition with the CVS state, which allows to determine functional changes in physio-
logical systems, identify the boundary between prenosological state (health) and dis-
ease, as well as affect the disease severity.
The purpouse of the paper is to develop information technology for the
classification of human health using sets of Data Mining methods by objective
and expert characteristics.
DEVELOPMENT OF INFORMATION TECHNOLOGY FOR CLASSIFICATION
OF FUNCTIONAL CONDITION AND HEALTH STATE
Large amounts of information, the need for it adequate analysis with the possibility of
further forecasting and planning of appropriate activities necessitate the development
and application of new technologies for assessing the current state of both individual
health and population health of Ukraine on objective and expert indicators.
Note the effectiveness of the use of Data Mining methods to determine the
risk groups according to clinical data, to assign patients to the appropriate group
by health markers with further prediction of its changes and evaluation of the
effectiveness of treatment. We have developed a method for detecting markers
of the cardiovascular system state, which is based on Data Mining models,
Kryvova O.A., Kozak L.M.
ISSN 2663-2586 (Online), ISSN 2663-2578 (Print). Cyb. and Comp. Eng. 2021. № 1 (203) 80
which are based on the analysis of heart rate variability (HRV) [30]. The devel-
opment of the method takes into account the experience of using the constructed
Data Mining models to determine population health clusters that are homogene-
ous in terms of medical and demographic indicators [31].
We have formed an ensemble of Data Mining methods, developed a research
scheme that uses a combination of filtering methods, cluster analysis algorithms (k-
means, EM) and classification (Decision Trees, Neural networks, SVM) using infor-
mative ECG features. The application of these methods makes it possible to combine
the possibilities of solving specific tasks at the stages of analysis: reducing the sample
size, selecting criteria / markers of the appropriate health level and classification of a
particular subject / patient to the appropriate group according to his health.
Let consider in more detail an ensemble of used Data Mining methods.
Selection of informative parameters (filtering). The choice of variables
follows from two tasks: 1) to find informative variables strongly connected with
the target feature, 2) to define a small parameters subset, keeping enough infor-
mation on initial indicators.
A peсuliarity of the initial data in our study was a large number of indicators —
ECG parameters. The multilevel system of indicators, calculated by automated ECG
analysis, had a total of 240 features. Such a large amount of primary data is characteris-
tic of many tasks in various fields of medical research. The correlation matrix was cal-
culated among the predictors to avoid the problem of multicollinearity. The correlation
coefficient (R > 0,7) is used as criteria for deciding whether variable may be excluded
from the analysis because another input variable contains the same information.
As you know, there are several reasons for the negative impact of a large
number of non-informative parameters on the learning algorithm quality, three
of which are considered basic [32]. One important reason is that as the parame-
ters number increases, more learning objects are needed for reliable classifica-
tion. In addition, with increasing parameters number decreases the statistical
reliability of the algorithm on the control data. The advantage of selecting in-
formative features is the increase in the accuracy of the classification algorithm,
generalization ability, achieving the possibility for the best interpretation of data.
Usually a preliminary selection of parameters is carried out before the start
of machine learning algorithms. Statistical criteria for correlation of each of the
primary features with the target feature and ordering (for example, by the size
and significance of Chi-Square Pearson, F — Fisher) are used. Further selection
of a set (combinations) of informative parameters is performed using classifica-
tion algorithms for greater accuracy [33].
Clustering. One of the effective methods of data processing is their seg-
mentation using cluster analysis methods (unsupervised learning). The clustering
process divides the data set into cluster groups or subclasses [3]. Clustering
(subgroups) allows you to use all available information to build multiple models,
and then make more accurate predictions for the model.
We used two most popular algorithms, namely k-means, EM, which are im-
plemented in the module Data Miner STATISTICA 10 [35, 36].
K–means method. The patients were divided into groups using the general-
ized k–means method. This method makes it possible to distribute observations
(from space Xn) into k clusters according to the following criteria.
Information Technology for Classification of Donosological and Pathological States Using
ISSN 2663-2586 (Online), ISSN 2663-2578 (Print). Cyb. and Comp. Eng. 2021. № 1 (203) 81
The first criterion for recalculating cluster centers is the minimization of the
objective function (F1) by the sum of the squares of the distances between each
object xi and the center of the cluster µn, to which it belonged at each iteration:
2
1
1
min
n
k
i i
n x X
F x μ
= ∈
= − →∑∑ , (1)
where xi is the set of n observations, n is the number of objects to be divided into
k groups (clusters), µn — cluster centers.
And the second criterion (F2) determines that the distances sum between the
clusters should be as large as possible:
2
2
, 1
max
k
i n
n i
F μ μ
=
= − →∑ . (2)
Additionally, in contrast to the classical method k–means, a cross-check is
performed on N random samples, which allows to minimize the error and to
select the optimal number of clusters. If the error function (average distance
between cluster centers) for a solution k+1 clusters is not 5 % better than the
solution for k clusters, then the solution with k clusters will be final (optimal).
Fuzzy clustering algorithm (EM). The expectation-maximization (EM) algo-
rithm assumes that the data correspond to a linear combination of distributions
(normal, lognormal, binomial):
1 1
( ) ( ), 1, 0,
k k
i i i i
i i
P x w p x w w
= =
= ⋅ = ≥∑ ∑ (3)
where k is a number of components in a mixture of distributions P(x), wi — is
weights of components, pi(x) — distribution density of components.
At each step of the iterative process, the expectation parameters are esti-
mated and the likelihood function is calculated until the maximum of logarith-
mic likelihood is reached. The k-fold cross-validation use with the error function
evaluation (loglikelihood) helps to determine the final number of clusters.
One of the accepted methods of estimating the required number of clusters
is the Cluster Validity Indices method [37].
Classification and Regression Trees (CART). Decision trees have become
the most common approach to solving the problem of assessing the patient's
condition [17], to detect ischemia of the heart [38], to classify the stages of heart
disease [39], as well as to identify changes in human functional states [30].
The advantage of the decision tree method is that there are no requirements for
data distribution, their type. This approach facilitates the interpretation of the results,
the model is displayed as a tree, the structure of which is determined by logical rules
(IF — THAT). Its purpose is to predict the target variable based on other features
known as predictors, which makes it possible to detect complex interactions.
We used the CART algorithm, a recursive method that allows us to develop
classification and regression models. According to the CART algorithm, the data
set is distributed across all variables sequentially into segments. The purpose of
sequential segmentation is to obtain uniformity of data on the selected attribute,
reducing uncertainty in the partition node.
Kryvova O.A., Kozak L.M.
ISSN 2663-2586 (Online), ISSN 2663-2578 (Print). Cyb. and Comp. Eng. 2021. № 1 (203) 82
In the CART algorithm for predictor selection and division into two nodes,
the index as a measure of uncertainty (Gini) is used:
( ) 1 ( )( )
k
d d
i j
i j
Gini d P P
≠
= −∑ , (4)
where Pi is the probability of classification in node d as i or j.
In each node, the reduction of the impurity is maximized.
To summarize the result, the optimal size tree is selected by cutting branches in
combination with the method of estimating the error of cross-checking (algorithms
minimal cost-complexity tree pruning, V-fold cross-validation).
The method of Support Vector Machine is based on vector space model,
which aims to find such a surface distribution between classes, which is the most
remote from all points of the learning set any of the classes. If we denote the
learning data set D = {Xi, yi}, where X is the vector of the i-point and yi is the
corresponding class label, then the linear classifier has the form:
( ) ( ),T
if x sign W X b= + (5)
where WT is weight vector and b is constant.
The optimization problem is solved, namely, the task of achieving the
maximum gap between the reference points:
1 min
2
TW W → . (6)
For all (Xi, yi) € D is satisfied when
( ) 1T
i iy W X b+ ≥ . (7)
This method is implemented in STATISTICA Data Mining module, there is
a possibility of transition to a nonlinear model using other core functions.
ANN Neural network is a mathematical apparatus that simulates the work of a
network of brain neurons. The components of the neural network consist of inputs xi,
which are fed to the neurons synapses that are connected by axons in several hidden
layers and the final outputs yi. The neuron state is described by a function
n
i i
i
S x w=∑ , (8)
where n is a number of inputs, wi — weights i — synapse. The output value of
the axon is
( ),Y f S= (9)
where f(S) is the activation function.
When learning the network, the task is to minimize the objective error func-
tion by the method of least squares:
Information Technology for Classification of Donosological and Pathological States Using
ISSN 2663-2586 (Online), ISSN 2663-2578 (Print). Cyb. and Comp. Eng. 2021. № 1 (203) 83
2
1
1( ) ( ) ,
2
k
j j
j
E w y d
=
= −∑ (10)
where yj is the value of the jth output of the neural network, dj — target value of
the jth output, k — is a number of neurons in the output layer.
Classification quality indicators. To evaluate the performance of the pro-
posed model the sensitivity, specificity, accuracy, and F-score are calculated.
The sensitivity is the proportion of positive instances that are correctly classified as
positive. The specificity is the proportion of negative instances that are correctly classi-
fied as negative. The accuracy is the proportion of instances that are correctly classified.
( ) TPSensitivity Recall
TP FN
=
+
, (11)
TPPrecision
TP FP
=
+
, (12)
TNSpecificity
FP TN
=
+
, (13)
TP TNpredicive Accuracy
TP TN FP FN
+
=
+ + +
, (14)
2 ,Recall PrecisionF score
Recall Precision
× ×
=
+
(15)
where — TP, TN, FP and FN are the numbers of true positives, true negatives,
false positives and false negatives respectively.
For multiclass case, these measures can be obtained from the confusion ma-
trix by comparing numbers of instances for each class in the matrix against in-
stances of all the other classes. F-score, since it combines precision and recall
into a single number evaluating the whole system performance [40].
To solve the tasks for a particular subject of analysis, the formation of an
appropriate ensemble of the considered set of methods was done. The initial data
were indicators of heart rate variability, objective indicators of the studied
physiological systems and expert assessments of the human health state.
The proposed information technology for the classification of functional
states and human health consists of four main stages (Fig. 1).
Stage 1. Data pre-processing. At stage 1, the input data is pre-processed,
checked for completeness, the presence of emissions, type compliance, reformat-
ting. The target feature is determined by the specific task of the analysis: infor-
mation about the response of body systems to external influences, expert data on
the severity of the condition (disease activity) of patients and so on. At this
stage, the number of primary HRV indicators was reduced and the most informa-
tive ones were selected regarding the target feature using filtration methods.
Kryvova O.A., Kozak L.M.
ISSN 2663-2586 (Online), ISSN 2663-2578 (Print). Cyb. and Comp. Eng. 2021. № 1 (203) 84
Fig. 1. Stages of information technology for the classification of functional states and
human health
At this stage, the primary selection methods of informative parameters are
used with statistical correlation criteria (Pearson’s Chi-Square, F), that results in
a reduction in the study volume to further determine the classification groups for
the gradation of studied state changes.
In stage 2 — clustering to form groups, the number and composition of ty-
pological groups were determined by sets of informative indicators. At this
stage, cluster analysis methods (k –means, EM) are used.
Stage 3. Predictors` identification. The transition to step 3 is carried out if there is
a need for analysis of the repeated measurements. That is, when the purpose of the
study is to identify changes in the informative indicators associated with changes in
the factor (e.g., response to exercise). Methods of repeated analysis of variance (Re-
pANOVA) are used, which allows to determine differences in informative indicators
changes in certain subgroups, as well as to provide a statistical assessment of the factor
influence. The result obtained at this stage will be a set of informative features that are
statistically significantly related to the factor.
Stage 4. Classification of the human condition severity. In this stage, infor-
mative indicators set were tested, which are predictors of the CVS state as at-
tributes of the state classification model. This step is performed if the initial data
contains the target attribute (class label) provided by the experts.
Information Technology for Classification of Donosological and Pathological States Using
ISSN 2663-2586 (Online), ISSN 2663-2578 (Print). Cyb. and Comp. Eng. 2021. № 1 (203) 85
Algorithms CART, ANN, and SVM were used. Comparisons of classifica-
tion features sets selected by different models and general classification accu-
racy of different models were performed. The efficiency indicators of the models
for each class were calculated (sensitivity, specificity, accuracy).
For samples of small number, cross-validation (10 -fold) was used to opti-
mize the complexity of the model. The model was chosen according to two crite-
ria: high enough accuracy and optimal complexity.
Calculated according to the CART algorithm, the decision tree with the op-
timal size allows to formulate classification rules for health of each severity
(logical conditions for the values of a small set of ECG parameters).
If the classification quality is unsatisfactory, it is possible to return to the previous
stages 1, 2, 3 using other selection methods of features subsets (or model parameters).
In the presence of a test sample, the quality of classification models is checked on it.
The end result is the classification rules according to the informative set of
ECG parameters, which determine the patient's condition severity.
During the development, the procedures of the Data Miner module of the
STATISTICA 10 package were used. Note that the Data Mining algorithms are
implanted in the Weka, RapidMiner, SAS Enterprise Miner software and in the
modern design tool Python, R.
STUDY OF FUNCTIONAL STATE AND HUMAN HEALTH WITH
THE USE OF DEVELOPED INFORMATION TECHNOLOGY
The proposed IT is used to solve problems aimed at studying the operators`
functional state (prenozological state) and to classify the patients` severity in
case of disease progression.
Determination of specific changes in operators` HRV indicators (pre-
nosological state). Verification of the developed IT was carried out according to
the experimental study of the reliability of operator activity under information
load, which was performed by employees of the Research Institute of Military
Medicine of the Armed Forces of Ukraine [41].
The condition of CVS regulatory mechanisms was studied by ECG recording
(for 2 min) using Cardio Sens AIC (KHAI Medica, Kharkiv). The analysis was per-
formed on the main HRV indicators, which belong to the generally accepted informa-
tive characteristic set of human functional state (statistical characteristics, spectral
analysis, spectral components in the ranges ULF, VLF, LF, HF).
The professionally important qualities of military operators and their reli-
ability of activities were assessed by tests consisted of information-intensive
tasks: the dynamic memorization quality test (DMQ); the test of determining the
speed and accuracy of the reaction to a moving object (RMO); the attention con-
centration and short-term memory test (ACSM). Factors influencing the opera-
tors` FS were determined according to the training process stages: 1 — rest state;
2 — QDM; 3 — RMO; 4 — ACSM; 5 — recovery state [42].
At the preparatory stage of each test, the individual optimal load level (τlim),
which the operator can still perform without errors, was determined. At the train-
ing stage, the tasks complexity increased by 10 % of the determined individual
optimal level. The percentage of errors made was used as an indicator of the
operator activity reliability at different levels of test task complexity. The tech-
nique of the training cycle is described in detail in the works [41, 42].
Kryvova O.A., Kozak L.M.
ISSN 2663-2586 (Online), ISSN 2663-2578 (Print). Cyb. and Comp. Eng. 2021. № 1 (203) 86
m
RR
SD
NN
LF
/H
F
AM
o,
%
(V
LF
+L
F)
/H
F
‐1.0
‐0.8
‐0.6
‐0.4
‐0.2
0.0
0.2
0.4
0.6
0.8
1.0
S
V
Fig. 2. Graf of means of indicators in two groups (S — sympathonics,
V — vagotonics)
At the IT first stage the complex of the HRV main indicators on groups is
defined:
— statistical characteristics: mode of RR-intervals (M0RR), amplitude mode
(AMo, %), standard deviation of RR-intervals (SDNN), stress index (SI);
— spectral indicators: total spectral power of the TP spectrum (0.003–0.4 Hz),
spectral components in the bands ULF (< 0.015 Hz), VLF (0.015–0.04 Hz),
LF (0.04–0,15 Hz), HF (0.15–0.4 Hz), the activation indices of subcortical centers
VLF/HF, index centralization IC = (VLF + LF)/HF.
At the second stage of IT with the help of cluster analysis the group at rest
state (1) was determined by the vegetative regulation type: 1) predominance of
sympathetic division (S, sympathonics), 2) predominance of parasympathetic
division (V, vagotonics). In fig. 2 standardized average values of indicators for
which clustering was performed are provided. The group of sympathonics (S)
includes 42, vagotonics (V) 28 operators.
Heart rate parameters in groups with different types of vegatative regula-
tion, which differed significantly at rest stage (1), undergo significant changes
during the training cycle, and at the stage of recovery (5) there is no significant
difference between the two typological groups (Fig. 3).
If at the initial stage (1) the spectrum was dominated by components of the
activity of the autonomous control loop (HF, LF), then after performing tests in
both groups there is a redistribution of power spectrum against the background
of decreasing mode of RR-intervals. The power of the high-frequency compo-
nent (HF) decreases, the low-frequency component of the spectrum (VLF) in-
creases, and the LF (first-order slow-wave power) increases, which reflects the
activity of the vasomotor center.
At the same time, it was determined that the test loads of dynamic memory
(DMQ) and rapid response (RMO) cause greater changes in HRV than the acti-
vation of attention concentration and memory (ACSM). Characteristically, after
performing all test loads (step 5), the components of the heart rate spectrum
return to values at rest state (1), except for the spectrum total power (TP) due to
an increase in the LF component in the sympathonics` group.
Information Technology for Classification of Donosological and Pathological States Using
ISSN 2663-2586 (Online), ISSN 2663-2578 (Print). Cyb. and Comp. Eng. 2021. № 1 (203) 87
R1*CL2; LS Means
Current effect: F(4, 260)=15,789, p=,00000
Effective hypothesis decomposition
Vertical bars denote 0,95 confidence intervals
S
V
mRR 1
mRR 2
mRR 3
mRR 4
mRR 5
R1
600
700
800
900
1000
D
V
_1
R1*CL2; LS Means
Current effect: F(4, 260)=8,0596, p=,00000
Effective hypothesis decomposition
Vertical bars denote 0,95 confidence intervals
S
V
AMo 1
AMo 2
AMo 3
AMo 4
AMo 5
R1
25
35
45
55
D
V_
1
a b
S
V
HF % 1
HF % 2
HF % 3
HF % 4
HF % 5
10
20
30
40
D
V
_1
S
V
TP 1 TP 2 TP 3 TP 4 TP 5
R1
0
2000
4000
6000
D
V
_1
c d
S
V
VLF% 1
VLF % 2
VLF % 3
VLF % 4
VLF % 5
R1
10
20
30
40
50
D
V
_1
S
V
IC 1 IC 2 IC 3 IC 4 IC 5
R1
0
2
4
6
8
10
DV
_1
e f
Fig. 3. Changes in spectral components at the stages of training (1, 2, 3, 4, 5) in sym-
pathonics and vagotonics: a) mode; b) the RR mode amplitude; c) HF - high frequency
component; d) TP — total power; e) VLF — power in the region of very low frequencies;
f) (VLF + LF) / HF — centralization index
At the same time, it was determined that the test loads of dynamic memory
(DMQ) and rapid response (RMO) cause greater changes in HRV than the acti-
vation of attention concentration and memory (ACSM). Characteristically, after
performing all test loads (step 5), the components of the heart rate spectrum
return to values at rest state (1), except for the spectrum total power (TP) due to
an increase in the LF component in the sympathonics` group.
According to the literature, it is known that in the bases of mechanisms of
formation of the low-frequency component (VLF) are stressors that activate the
renin-angiotensin-aldosterone system and increase the catecholamines concen-
Kryvova O.A., Kozak L.M.
ISSN 2663-2586 (Online), ISSN 2663-2578 (Print). Cyb. and Comp. Eng. 2021. № 1 (203) 88
tration in plasma. The VLF component power is associated with the activity of
suprasegmental (hypothalamic) centers of vegetative regulation, which are
transmitted through the sympathetic part of the VNS [21].
Thus, after performing NPS and PPO tests changes in heart rate regulation
occur, namely: acceleration of heart rate (decrease in MoRR), increase in low-
frequency (VLF) and high-frequency (HF) heart rate fluctuations, increase in the
centralization index (influence of the central control loop), indicating the stress-
ful nature of these loads and significant psycho-emotional stress of the operators.
Development of a classification model of disease activity in children
with dysplasia. Connective tissue dysplasia (CTD) is a systemic disease that
arises at an early age, has many manifestations in the cardiovascular system,
musculoskeletal system and other organs. To predict the disease development, it
is important to know the diagnostic criteria that characterize the stages of disease
activity. The purpose of the study is to determine these criteria according to the
system of ECG indicators.
The classification of the severity of the condition of children with CTD was
developed according to the arrays of ECG indicators, as well as indicators of the
severity of the condition of patients determined by expert physicians. The final
indicator of CVS state is the final assessment (FA), which is formed from com-
plex assessments of lower level: health rate regulation, myocardial status and
additional features (quantitative and qualitative assessments of different coding
systems, arrhythmias, risk of sudden cardiac events etc.). Complex indicators are
calculated in points (0–100).
The study was based on data from laboratory and clinical examination of 25
children with CTD manifestations. Disease activity was measured by the Juve-
nile Arthritis Activity Scale (JADAS) [43]. 6-channel ECG recording was per-
formed for 5–20 minutes using a Cardio Plus P device.
Cluster analysis methods (k-means with 10-fold cross-validation) allowed
identifying two typological groups for comprehensive assessments of the CVS
regulation, myocardium condition and its reserves:
- group 1 (16 children) had a low level of complex assessment:
FA1 = 58.3 ± 8.1;
- group 2 (9 children) - significantly differed by higher complex assessment:
FA2 = 68.3 ± 5,2. (I = 68,3 ± 5,2).
The optimal set of CVS state predictors is determined. The set of predictors
consists of the following primary ECG parameters: cardiac arrhythmia (Heart
rhythm disorders), T-wave amplitude (lead II), integrated indicator of the STT
form (lead II), QRS — alpha angle, T-wave symmetry ratio. The error of the
regression model (by CART algorithm) for a set of 5 parameters ECG is 18.8 %,
the correlation coefficient R = 0.88.
Classification models of disease activity stages were developed. CVS FA
predictors were tested as attributes on CART, Neural network, SVM models.
The target variable — disease activity was determined by 3 gradations provided
by experts (1 — the initial stage of activity, 2, 3 — subsequent stages of
inflammation increasing). The quality comparison of 3 models in the training
sample gave such training errors: CART — 0 %, Neural network — 24 %,
SVM — 44%. That is, for a small sample, the best result was for the C&RT
decision tree model — 100% classification accuracy.
Information Technology for Classification of Donosological and Pathological States Using
ISSN 2663-2586 (Online), ISSN 2663-2578 (Print). Cyb. and Comp. Eng. 2021. № 1 (203) 89
After 10-fold cross-validation of the CART models revealed 4 indicators, which
determine the disease activity with an overall classification accuracy of 88 %.
The most significant attributes of the disease activity model and their contribution to
the CART model (by rank) is shown (Tab. 1).
The optimal classification tree, which was determined after 10-fold cross-
validation is given (Fig.4). It should be noted that the distribution of the training
sample into groups with different activity was unbalanced.
The quality measures of the classification of these disease stages are shown
(Tab.2). The average F-score = 0,94.
The quality measures of the classification of these disease stages. Thus ac-
cording to the optimal model, the CTD stages are classified with high accuracy.
In accordance with the tree splitting conditions, logical rules for the severity classi-
fication of the condition are formulated, in particular the basic classification rules are:
• Low disease activity level D1:
if Ind STT > 49,5 and Ind STT ≤ 84,0 and Amp T(II) ≤ -160,5 then D1 = 1
if Ind STT > 49,5 and Ind STT ≤ 84,0 and Amp T(II) > -160,5 and
α-QRS ≤ 75,5 and SimmT(I) ≤ -0,57 then D1 = 1
• Middle disease activity level D2 =2:
if Ind STT > 49,5 and ≤ 84,0 and Amp T(II) > -160,5 and α-QRS > 75,5
then D2 = 2
if Ind STT > 49,5 and Ind STT ≤ 84,0 and Amp T(II) > -160,5 and
α-QRS ≤ 75,5 and SimmT(I) > 0,57 then D2 = 2
• High disease activity level D3 =3:
if Ind STT ≤ 49,5 then D3 = 3
if Ind STT > 49,5 and Ind STT > 84,0 then D3 = 3.
Table 1. Predictor importance for the classification of CTD activity stages
The best predictors Variable importance rank
T - wave symmet ratio (I) 100
Ampl. T (II) 86
Ind. form STT (II) 76
α QRS 68
Table 2. Classification results for CTD activity detection
Disease activity
Measures (%)
1 2 3
Sensitivity 100 81,8 100
Specificity 85,7 100 100
Predictive Accuracy 92 92 100
Kryvova O.A., Kozak L.M.
ISSN 2663-2586 (Online), ISSN 2663-2578 (Print). Cyb. and Comp. Eng. 2021. № 1 (203) 90
Tree 2 graph for disease activity
Num. of non-terminal nodes: 5, Num. of terminal nodes: 6
ID=1 N=251
ID=3 N=231
ID=4 N=221
ID=7 N=172
ID=10 N=111
ID=2 N=23
ID=6 N=51
ID=12 N=81 ID=13 N=32
ID=11 N=62
ID=5 N=13
Int ind form STT(II)
<= 49,5 > 49,5
Int ind form STT(II)
<= 84,0 > 84,0
Amp T(II)
<= -160,5 > -160,5
a QRS
<= 75,5 > 75,5
Simm Т (I) <= 0,57 > 0,57
1
2
3
Fig. 4. Classification tree of CTD activity
Thus, with the help of the developed information technology the ECG indicators
are determined, the changes of which can be markers of CVS disorders in the case of
inflammatory processes in children diagnosed with juvenile arthritis, rheumatic dis-
ease. Markers of the initial stages of activity were determined by the following ECG
parameters: α-QRS angle, chahge of the the T–wave (I) symmetry ratio.
Changes in the STT form (less than 49,5 or more than 84 points) indicates
increased disease activity.
According to experts, the use of the proposed information technology to de-
termine the CTD activity according the ECG parameters will allow the physician
to identify the initial stages of the process in an outpatient setting. The advantage
of this approach is the possibility of simultaneous assessment of CVS functional
Information Technology for Classification of Donosological and Pathological States Using
ISSN 2663-2586 (Online), ISSN 2663-2578 (Print). Cyb. and Comp. Eng. 2021. № 1 (203) 91
changes and the disease activity level before the clinical manifestations of the
inflammatory process. The STT shape indicator gives an opportunity to select a
group of children with apropriate changes in the STT segment. Such changes
reflect dysmetabolic, hypoxic changes of the myocardium that accompany the
manifestations of inflammatory processes [44]. At the same time, it is also nec-
essary to take into account changes in the T wave amplitude, changes in the an-
gle alpha angle QRS, the T wave symmetry index.
Prospects for solving the problem of physician` information support. Further
development of a clinical desition support system in disease severity determining will
be aimed at analyzing large arrays of clinical, laboratory and instrumental data in order
to improve the classification accuracy for an extended range of tasks.
CONCLUSIONS
The created information technology, which combines the generalized stages:
data pre-processing to reduce the studied data set, clustering (data segmentation,
likelihood function biulding), predictors` identification by analysis of Data Min-
ing models and classification of human condition with formation of final charac-
teristics allows to determine pecularities of human functional state change under
external factors influence and severity patients by analysis of heart rate variabil-
ity and expert characteristics.
The combination of Data Mining methods used at different stages of IT
allows solving consistently the necessary tasks: by filtering indicators, the rele-
vant features are determined; the use of clustering provides the homogeneous
groups detection; the decision tree method (CART algorithm) makes it possible
to build a classification rules and high classification accuracy.
Using the developed IT, specific changes in HRV indicators in operators,
which occur under the influence of various types of information loads, are de-
termined taking into account the type of vegetative regulation. Loads of dynamic
memorization and rapid response cause greater changes in HRV than activation
of attention concentration and short-term memory. Thus, the following shifts in
HRV regulation occur during the performance of these tasks: acceleration of
heart rate (decrease in MoRR), increase of low-frequency (VLF) and high-
frequency (HF) heart rate fluctuations, increase of centralization index (influ-
ence of central regulation loop), which indicates stress loads and significant
psycho-emotional strain in operators. In the recovery state after all test loads,
only in sympathonics, the spectrum total power (due to an increase in the LF
component) does not return to the initial values.
The use of developed models and technologies to classify the patients` severity al-
lowed to assess the CVS state of children with dysplasia, identify markers of stages of
different disease activity and build diagnostic rules, the use of which make it possible
to predict the disease severity and to adjust treatment tactics.
Kryvova O.A., Kozak L.M.
ISSN 2663-2586 (Online), ISSN 2663-2578 (Print). Cyb. and Comp. Eng. 2021. № 1 (203) 92
REFERENCES
1. Ian H. Data Mining Practical Machine Learning Tools and Techniques Witten, Eibe
Frank and Mark A. Hall Data Mining: Practical Machine Learning Tools and Tech-
niques. 3rd Edition. Morgan Kaufmann, 2011, 665 p.
2. Yoo I., Alafaireet P., Marinov M., Pena-Hernandez K., Gopidi R., Chang J. F. Data Min-
ing in Healthcare and Biomedicine: A Survey of the Literature. Journal of medical sys-
tems. 2012, no 36(4), pp. 2431–2448.
3. Chen M., Hao Y. , Hwang K., Wang L., Wang L. Disease Prediction by Machine Learn-
ing Over Big Data From Healthcare Communities. IEEE Access 2017;5:8869-8879.
4. Safdar S., Zafar S., Zafar N., Khan N.F. Machine learning based decision support
systems (DSS) for heart disease diagnosis: a review. Artificial Intelligence Review. 2018,
50 (4), pp. 597–623.
5. Roopa C. K., Harish B. S. Survey on various Machine Learning Approaches for ECG Analysis.
International Journal of Computer Applications. 2017, no 9, vol. 163, pp.25–33.
6. Mohan S., Thirumalai C., Srivastava G. Effective heart disease prediction using hybrid
machine learning techniques. IEEE Access, 2019. 7:81542–81554.
7. Goldstein B.A., Navar A.M., Pencina M.J., Ioannidis J.P. Opportunities and challenges
in developing risk prediction models with electronic health records data: a systematic
review. J Am Med Inform Assoc. 2017, Jan; 24(1):198-208.
8. Antomonov M.Yu. Algorithmization of the choice of adequate mathematical methods in
the analysis of medical and biological data. Kibernetika i vyčislitel`naâ tehnika. 2007,
Iss. 153, pp. 12–23. (In Russian)
9. Georga E.I., Tachos N.S., Sakellarios A.I., Kigka V.I., Exarchos T.P., Pelosi G. Artificial
intelligence and data mining methods for cardiovascular risk prediction Cardiovascular
Computing. Methodologies and Clinical Applications. 2019, pp. 279–301
10. Amin M., Chiam Y. Identification of significant features and data mining techniques in
predicting heart disease. Telematics and Informatics. 2019, Vol. 36, pp. 82–93.
11. Kaieski N., da Costa C.A., da Rosa Righi R., Lora P.S. Application of artificial intelli-
gence methods in vital signs analysis of hospitalized patients: A systematic literature re-
view. Applied Soft Computing. 2020, Vol. 96,
12. Owens W.D., Felts J.A., et al. A physical status classification: A study of consistency of
ratings. Anesthesiology. 1978, Vol. 49, pp. 239–243.
13. Lemeshow S., Le Gall J.R: Modeling the severity of illness of ICU patients. JAMA.
1994, Vol 272, pp.1049–1055.
14. Le Gall J.R., Lemeshow S., Saulnier F: A new simplified acute physiology score
(SAPS II) based on a European/North American multicenter study. JAMA. 1993,
270 (24), pp. 2957–2963.
15. Knaus W.A., Draper E.A., Wagner D.P., Zimmerman J.E: APACHE II: A severity of
disease classification system. Cri.t Care Med .1985, 13:818-829.
16. Lemeshow S., Teres D., Klar J., Avrunin J.S., Gehlbach S.H., Rapoport J. Mortality
probability models (MPM II) based on an international cohort of intensive care unit pa-
tients. JAMA 1993, 270, pp. 2478-86
17. Trujillano J., Badia M, Serviá L. Stratification of the severity of critically ill patients with
classification trees. BMC medical research methodology. 2009, V 9, no 7, pp. 83–95.
18. Kim S., Kim W., Park R.W. A Comparison of intensive care unit mortality prediction models
through the use of Data Mining Techniques. Health Inform Res 2011,17, pp. 232–43.
19. Allyn J. et all. A comparison of a machine learning model with EuroSCORE II in pre-
dicting mortality after elective cardiac surgery: a decision curve analysis. PLoS one
2017, 12(1), pp. 1–12.
20. Amosov N.M. Thinking about health. Мoskow: 1978, 178 p. (In Russian)
21. Baevsky R.M., Berseneva A.P. Introduction to prenosological diagnostics. Мoskow:
Slovo, 2008, 174 p. (In Russian)
22. HRV analysis software URL: http://www.nevrokard.eu/maini/hrv.html (last access 20.10.2020)
23. Fainzilberg L.S. Computer diagnostics based on the phase portrait of an electrocardio-
gram. Kyiv: Osvita Ukrainy. 2013, 191 p. (In Russian)
Information Technology for Classification of Donosological and Pathological States Using
ISSN 2663-2586 (Online), ISSN 2663-2578 (Print). Cyb. and Comp. Eng. 2021. № 1 (203) 93
24. Gritsenko V.I., Fainzilberg L.S. Intelligent information technologies in digital medicine
on the example of phasagraphy. Kyiv: Naukova Dumka. 2019, 423 p. (In Russian)
25. Fainzilberg L.S., Dykach Ju.R. Linguistic approach for estimation of electrocardio-
grams’s subtle changes based on the Levenstein distance. Cybernetics and Computer
Engineering. 2019, no. 2 (196), pp. 3–26.
26. Gritsenko V.I., Fainzilberg L.S. Current state and prospects for the development of digi-
tal medicine. Cybernetics and Computer Engineering. 2020, no. 1 (199), pp. 59–84.
27. Richman J.S. Randall M.J. Physiological time–series analysis using approximate entropy and
sample entropy. Am J. Physiol. Heart Circ. Physiol. 2000, Vol. 278, № 6, pp. H22039–H2049.
28. İşler Y., Kuntalp M. Combining classical HRV indices with wavelet entropy measures
improves to performance in diagnosing congestive heart failure. Computers in Biology
and Medicine. 2007, Vol. 37, no. 10, pp. 1502–1510.
29. Valupadasu R., Chunduri B. R., Chanagoni V. Identification of Cardiac Ischemia using
bispectral analysis of ECG. Biomedical Engineering and Sciences (IECBES). 2012:
IEEE EMBS Conference on, Langkawi. 2012, pp. 999–1003.
30. Romanyuk O.A., Kozak L.M., Kovalenko A.S., Kryvova O.A. Digital transformation in
medicine: from formalized medical documents to information technologies of digital
medicine. Cybernetics and Computer Engineering. 2018, no. 4(194), pp. 61–78.
31. Krivova O.A., Kozak L.M. Comprehensive assessment of regional demographic devel-
opment. Kibernetika i vyčislitel`naâ tehnika. 2015, Iss 182, pp. 70–84 (In Russian)
32. Wolf L., Shashua A. Features Selection for Unsupervised and Supervised Inference:
The Emergence of Sparsity in a Weight-Based Approach. J. Machine Learning Res.
2005, V. 6, pp. 1855–1887.
33. Guyon I., Elisseeff A. An Introduction to Variable and Feature Selection. Journal of
Machine Learning Research. 2003, V 3, pp. 1157–1182.
34. Mandel I.D. Cluster analysis. Moscow: Finance and Statistics. 1988. 128 p. (In Russian)
35. Tzortzis G., Likas A. The MinMax k-Means clustering algorithm. Pattern Recognition.
2014, no 47 (7), pp. 2505–2516.
36. McLachlan G. Krishnan T. The EM algorithm and extensions. New York, United
States: Wiley. 1997, 274 p.
37. Wang K., Wang B., Peng L. CVAP: Validation for cluster analyses. Data Science Jour-
nal. 2009, no 8, pp. 88–93.
38. Fayn J. A classification tree approach for cardiac ischemia detection using spatiotempo-
ral information from three standard ECG leads. IEEE Trans. Biomed. Eng. 2011, V. 58,
no 1, pp. 95–102.
39. Pecchia L., Melillo P. Bracale M. Remote health monitoring of heart failure with data
mining via CART method on HRV features. IEEE Transactions Biomedical Engineer-
ing. 2011, V. 58(3), pp. 800–804.
40. Sokolova M., Lapalme G. A systematic analysis of performance measures for classifica-
tion tasks. Information processing & management. 2009, V. 45, N 4, pp. 427–437.
41. Kalnish V.V., Shvets A.V. Information technology for psychophysiological support of
high reliability of operator activities. Kibernetika i vyčislitel`naâ tehnika. 2014, Iss. 177,
pp. 54–67. (In Russian)
42. Shvets A.V., Kalnysh V.V. Features of influence of various psychophysiological states on reli-
ability of operator` activity. Military medicine of Ukraine. 2009, no 1, pp. 84–91. (In Ukrainian)
43. Consolaro A., Ruperto N, Bazso A. Development and validation of a composite disease activity
score for juvenile idiopathic arthritis. Arthritis & Rheumatism, 2009, vol. 61, pp. 658–666.
44. Ansari S., Farzaneh N, Duda M, Horan K. A review of automated methods for detection
of myocardial ischemia and infarction using electrocardiogram and electronic health re-
cords. IEEE reviews in biomedical engineering. 2017, Vol. 10, pp. 264–298.
Received 31.11.2020
Kryvova O.A., Kozak L.M.
ISSN 2663-2586 (Online), ISSN 2663-2578 (Print). Cyb. and Comp. Eng. 2021. № 1 (203) 94
ЛІТЕРАТУРА
1. Ian H. Data Mining Practical Machine Learning Tools and Techniques Witten, Eibe
Frank and Mark A. Hall Data Mining: Practical Machine Learning Tools and Tech-
niques. 3rd Edition. Morgan Kaufmann, 2011. 665 p.
2. Yoo I., Alafaireet P., Marinov M., Pena-Hernandez K., Gopidi R., Chang J. F. Data Min-
ing in Healthcare and Biomedicine: A Survey of the Literature. Journal of medical sys-
tems. 2012. No 36(4). P. 2431–2448.
3. Chen M., Hao Y. , Hwang K., Wang L., Wang L. Disease Prediction by Machine Learn-
ing Over Big Data From Healthcare Communities. IEEE Access. 2017;5:8869-8879.
4. Safdar S., Zafar S., Zafar N., Khan N.F. Machine learning based decision support sys-
tems (DSS) for heart disease diagnosis: a review. Artificial Intelligence Review. 2018, 50
(4), 597-623.
5. Roopa C. K., Harish B. S. Survey on various Machine Learning Approaches for ECG Analysis.
International Journal of Computer Applications. 2017. no 9. Vol. 163. pp.25–33.
6. Mohan S., Thirumalai C., Srivastava G. Effective heart disease prediction using hybrid
machine learning techniques. IEEE Access, 2019. 7:81542–81554.
7. Goldstein B.A., Navar A.M., Pencina M.J., Ioannidis J.P. Opportunities and challenges
in developing risk prediction models with electronic health records data: a systematic
review. J Am Med Inform Assoc. 2017, Jan; 24(1):198-208.
8. Антомонов М.Ю. Алгоритмизация выбора адекватных математических методов
при анализе медико-биологических данных. Кибернетика и вычислительная
техника. 2007. Вып. 153. С. 12–23.
9. Georga E.I., Tachos N.S., Sakellarios A.I., Kigka V.I., Exarchos T.P., Pelosi G. Artificial
intelligence and data mining methods for cardiovascular risk prediction Cardiovascular
Computing. Methodologies and Clinical Applications. 2019. P. 279–301
10. Amin M., Chiam Y. Identification of significant features and data mining techniques in
predicting heart disease. Telematics and Informatics. 2019. Vol. 36. P. 82–93.
11. Kaieski N., da Costa C.A., da Rosa Righi R., Lora P.S. Application of artificial intelli-
gence methods in vital signs analysis of hospitalized patients: A systematic literature re-
view. Applied Soft Computing. 2020. Vol. 96.
12. Owens W.D., Felts J.A., et al. A physical status classification: A study of consistency of
ratings. Anesthesiology. 1978. Vol. 49. P. 239–243.
13. Lemeshow S., Le Gall J.R: Modeling the severity of illness of ICU patients. JAMA.
1994, Vol 272. P.1049–1055.
14. Le Gall J.R., Lemeshow S., Saulnier F: A new simplified acute physiology score (SAPS II)
based on a European/North American multicenter study. JAMA. 1993, 270 (24). P.: 2957–2963.
15. Knaus W.A., Draper E.A., Wagner D.P., Zimmerman J.E: APACHE II: A severity of
disease classification system. Cri.t Care Med .1985. 13:818-829.
16. Lemeshow S., Teres D., Klar J., Avrunin J.S., Gehlbach S.H., Rapoport J. Mortality
probability models (MPM II) based on an international cohort of intensive care unit pa-
tients. JAMA.1993. 270:2478-86
17. Trujillano J., Badia M, Serviá L. Stratification of the severity of critically ill patients with classi-
fication trees. BMC medical research methodology. 2009. V 9. no 7. P. 83–95.
18. Kim S., Kim W., Park R.W. A Comparison of intensive care unit mortality prediction
models through the use of Data Mining Techniques. Health Inform Res 2011;17:232-43.
19. Allyn J. et all. A comparison of a machine learning model with EuroSCORE II in pre-
dicting mortality after elective cardiac surgery: a decision curve analysis. PloS one 2017.
12(1). P. 1–12.
20. Амосов Н.М. Раздумья о здоровье — М: 1978. 178 с.
21. Баевский Р.М., Берсенева А.П. Введение в донозологическую диагностику. М.:
Слово, 2008. 174 с.
22. HRV analysis software. URL: http://www.nevrokard.eu/maini/hrv.html (дата звернен-
ня: 20.10.2020)
23. Файнзильберг Л.С. Компьютерная диагностика по фазовому портрету электрокар-
диограммы. К.: Освита Украины, 2013. 191 c.
Information Technology for Classification of Donosological and Pathological States Using
ISSN 2663-2586 (Online), ISSN 2663-2578 (Print). Cyb. and Comp. Eng. 2021. № 1 (203) 95
24. Гриценко В.И., Файнзильберг Л.С. Интеллектуальные информационные технологии в
цифровой медицине на примере фазаграфии. Киев: Наукова Думка, 2019. 423 с.
25. Fainzilberg L.S., Dykach Ju.R. Linguistic approach for estimation of
electrocardiograms’s subtle changes based on the Levenstein distance. Cybernetics and
Computer Engineering. 2019. No. 2 (196). P. 3–26.
26. Gritsenko V.I., Fainzilberg L.S. Current state and prospects for the development of digi-
tal medicine. Cybernetics and Computer Engineering. 2020. No. 1 (199). P. 59–84.
27. Richman J.S. , Randall M.J.Physiological time–series analysis using approximate entropy and
sample entropy. Am J. Physiol. Heart Circ. Physiol. 2000. Vol. 278. № 6. P. H22039–H2049.
28. İşler Y., Kuntalp M. Combining classical HRV indices with wavelet entropy measures
improves to performance in diagnosing congestive heart failure. Computers in Biology
and Medicine. 2007. Vol. 37. No. 10. P. 1502–1510.
29. Valupadasu R., Chunduri B. R., Chanagoni V. Identification of Cardiac Ischemia using
bispectral analysis of ECG. Biomedical Engineering and Sciences (IECBES). 2012:
IEEE EMBS Conference on, Langkawi. 2012. P. 999–1003.
30. Romanyuk O.A., Kozak L.M., Kovalenko A.S., Kryvova O.A. Digital transformation in
medicine: from formalized medical documents to information technologies of digital
medicine. Cybernetics and Computer Engineering. 2018. No. 4(194). P. 61–78.
31. Кривова О.А., Козак Л.М. Комплексная оценка регионального демографического
развития. Киб.и выч.техн., 2015. Вып 182. С. 70–84.
32. Wolf L., Shashua A. Features Selection for Unsupervised and Supervised Inference:
The Emergence of Sparsity in a Weight-Based Approach. J. Machine Learning Res.
2005. V. 6. P. 1855–1887.
33. Guyon I., Elisseeff A. An Introduction to Variable and Feature Selection. Journal of
Machine Learning Research. 2003. V 3. P. 1157–1182.
34. Мандель И.Д. Кластерный анализ. М.: Финансы и статистика. 1988. 128 с.
35. Tzortzis G., Likas A. The MinMax k-Means clustering algorithm. Pattern Recognition.
2014. No 47 (7). Pp 2505–2516.
36. McLachlan G. Krishnan T. The EM algorithm and extensions. New York, United
States: Wiley. (1997) 274 p.
37. Wang K., Wang B., Peng L. CVAP: Validation for cluster analyses. Data Science Jour-
nal. 2009. No 8. Pp. 88–93.
38. Fayn J. A classification tree approach for cardiac ischemia detection using spatiotempo-
ral information from three standard ECG leads. IEEE Trans. Biomed. Eng. 2011. V. 58.
no 1. P. 95–102.
39. Pecchia L., Melillo P. Bracale M. Remote health monitoring of heart failure with data
mining via CART method on HRV features. IEEE Transactions Biomedical Engineer-
ing, 2011 V. 58(3). P. 800–804.
40. Sokolova M., Lapalme G. A systematic analysis of performance measures for classifica-
tion tasks. Information processing & management. 2009 (45). N 4. V. 45. P. 427–437.
41. Кальниш В.В., Швец А.В. Информационная технология психофизиологического
обеспечения высокой надежности операторской деятельности. Киберн. И выч.
техн. 2014. Вып. 177. С. 54–67.
42. Швець А.В., Кальниш В.В. Особливості впливу різних психофізіологічних станів на
надійність операторської діяльності. Військова медицина України. 2009. № 1. С. 84–91.
43. Consolaro A., Ruperto N, Bazso A. Development and validation of a composite disease activity
score for juvenile idiopathic arthritis. Arthritis & Rheumatism. 2009. Vol. 61. P. 658–666.
44. Ansari S., Farzaneh N, Duda M, Horan K. A review of automated methods for detection
of myocardial ischemia and infarction using electrocardiogram and electronic health re-
cords. IEEE reviews in biomedical engineering. 2017. Vol. 10. P. 264–298.
Отримано 31.11.2020
Kryvova O.A., Kozak L.M.
ISSN 2663-2586 (Online), ISSN 2663-2578 (Print). Cyb. and Comp. Eng. 2021. № 1 (203) 96
Кривова О.А., наук. співроб.,
відд. медичних інформаційних систем
ORCID: 0000-0002-4407-5990
e-mail: ol.kryvova@gmail.com
Козак Л.M., д-р біол. наук, старш. наук. співроб.,
провід. наук. співроб. відд. медичних інформаційних систем
ORCID: 0000-0002-7412-3041
e-mail: lmkozak52@gmail.com
Міжнародний науково-навчальний центр інформаційних
технологій та систем НАН України та МОН України,
пр. Акад. Глушкова, 40, м. Київ, 03187, Україна
ІНФОРМАЦІЙНА ТЕХНОЛОГІЯ КЛАСИФІКАЦІЇ
ДОНОЗОЛОГІЧНИХ ТА ПАТОЛОГІЧНИХ СТАНІВ ЗДОРОВ’Я
З ВИКОРИСТАННЯМ АНСАМБЛЮ МЕТОДІВ DATA MINING
Вступ. Впровадження цифрових технологій забезпечує реєстрацію великих обсягів біо-
медичних даних (ЕКГ, ЕЕГ, електронних медичних записів) як основи для оцінювання
і прогнозування стану пацієнтів. Методи Data Mining дають змогу виявити найбільш
інформативні показники, типологічні групи, класифікувати функційний стан людини і
стадії захворювання для прогнозування їхніх змін.
Метою роботи є розроблення інформаційної технології класифікації стану здоро-
в'я людини за допомогою комплексу методів Data Mining за об'єктивними та експерт-
ними характеристиками.
Результати. Розроблена інформаційна технологія об'єднує кілька етапів: I — по-
переднє оброблення даних; II — кластеризація, вибір однорідних груп (сегментація
даних); III — ідентифікація предикторів; IV — класифікація досліджуваних станів,
розроблення прогнозних моделей за допомогою алгоритмів машинного навчання (де-
рев рішень (Decision trees, опорних векторних машин Support vector machine, нейрон-
них мереж) та методу перевірки навчальної вибірки (cross-validation). Запропоновану
ІТ використано для дослідження функційного стану операторів та класифікації тяжкос-
ті стану пацієнтів у разі прогресування захворювання.
Висновки. Використання інформаційної технології для оцінювання успішності
діяльності операторів дало можливість виділити найінформативніші показники ВРС, за
змінами яких можна прогнозувати надійність діяльності операторів з урахуванням
типу вегетативної регуляції. Оцінювання активності захворювання дітей з дисплазією з
використанням ІТ дало змогу ідентифікувати діагностичні маркери ССС та розробити
діагностичні правила для визначення стадій захворювання за параметрами ЕКГ (симе-
трія зубця Т, інтегральний показник форми сегмента STT).
Ключові слова: інформаційна технологія, Data Mining, моделі машинного навчан-
ня, тяжкість стану пацієнта.
<<
/ASCII85EncodePages false
/AllowTransparency false
/AutoPositionEPSFiles true
/AutoRotatePages /All
/Binding /Left
/CalGrayProfile (Dot Gain 20%)
/CalRGBProfile (sRGB IEC61966-2.1)
/CalCMYKProfile (U.S. Web Coated \050SWOP\051 v2)
/sRGBProfile (sRGB IEC61966-2.1)
/CannotEmbedFontPolicy /Warning
/CompatibilityLevel 1.4
/CompressObjects /Tags
/CompressPages true
/ConvertImagesToIndexed true
/PassThroughJPEGImages true
/CreateJDFFile false
/CreateJobTicket false
/DefaultRenderingIntent /Default
/DetectBlends true
/DetectCurves 0.0000
/ColorConversionStrategy /LeaveColorUnchanged
/DoThumbnails false
/EmbedAllFonts true
/EmbedOpenType false
/ParseICCProfilesInComments true
/EmbedJobOptions true
/DSCReportingLevel 0
/EmitDSCWarnings false
/EndPage -1
/ImageMemory 1048576
/LockDistillerParams false
/MaxSubsetPct 100
/Optimize true
/OPM 1
/ParseDSCComments true
/ParseDSCCommentsForDocInfo true
/PreserveCopyPage true
/PreserveDICMYKValues true
/PreserveEPSInfo true
/PreserveFlatness true
/PreserveHalftoneInfo false
/PreserveOPIComments false
/PreserveOverprintSettings true
/StartPage 1
/SubsetFonts true
/TransferFunctionInfo /Apply
/UCRandBGInfo /Preserve
/UsePrologue false
/ColorSettingsFile ()
/AlwaysEmbed [ true
]
/NeverEmbed [ true
]
/AntiAliasColorImages false
/CropColorImages true
/ColorImageMinResolution 300
/ColorImageMinResolutionPolicy /OK
/DownsampleColorImages true
/ColorImageDownsampleType /Bicubic
/ColorImageResolution 300
/ColorImageDepth -1
/ColorImageMinDownsampleDepth 1
/ColorImageDownsampleThreshold 1.50000
/EncodeColorImages true
/ColorImageFilter /DCTEncode
/AutoFilterColorImages true
/ColorImageAutoFilterStrategy /JPEG
/ColorACSImageDict <<
/QFactor 0.15
/HSamples [1 1 1 1] /VSamples [1 1 1 1]
>>
/ColorImageDict <<
/QFactor 0.15
/HSamples [1 1 1 1] /VSamples [1 1 1 1]
>>
/JPEG2000ColorACSImageDict <<
/TileWidth 256
/TileHeight 256
/Quality 30
>>
/JPEG2000ColorImageDict <<
/TileWidth 256
/TileHeight 256
/Quality 30
>>
/AntiAliasGrayImages false
/CropGrayImages true
/GrayImageMinResolution 300
/GrayImageMinResolutionPolicy /OK
/DownsampleGrayImages true
/GrayImageDownsampleType /Bicubic
/GrayImageResolution 300
/GrayImageDepth -1
/GrayImageMinDownsampleDepth 2
/GrayImageDownsampleThreshold 1.50000
/EncodeGrayImages true
/GrayImageFilter /DCTEncode
/AutoFilterGrayImages true
/GrayImageAutoFilterStrategy /JPEG
/GrayACSImageDict <<
/QFactor 0.15
/HSamples [1 1 1 1] /VSamples [1 1 1 1]
>>
/GrayImageDict <<
/QFactor 0.15
/HSamples [1 1 1 1] /VSamples [1 1 1 1]
>>
/JPEG2000GrayACSImageDict <<
/TileWidth 256
/TileHeight 256
/Quality 30
>>
/JPEG2000GrayImageDict <<
/TileWidth 256
/TileHeight 256
/Quality 30
>>
/AntiAliasMonoImages false
/CropMonoImages true
/MonoImageMinResolution 1200
/MonoImageMinResolutionPolicy /OK
/DownsampleMonoImages true
/MonoImageDownsampleType /Bicubic
/MonoImageResolution 1200
/MonoImageDepth -1
/MonoImageDownsampleThreshold 1.50000
/EncodeMonoImages true
/MonoImageFilter /CCITTFaxEncode
/MonoImageDict <<
/K -1
>>
/AllowPSXObjects false
/CheckCompliance [
/None
]
/PDFX1aCheck false
/PDFX3Check false
/PDFXCompliantPDFOnly false
/PDFXNoTrimBoxError true
/PDFXTrimBoxToMediaBoxOffset [
0.00000
0.00000
0.00000
0.00000
]
/PDFXSetBleedBoxToMediaBox true
/PDFXBleedBoxToTrimBoxOffset [
0.00000
0.00000
0.00000
0.00000
]
/PDFXOutputIntentProfile ()
/PDFXOutputConditionIdentifier ()
/PDFXOutputCondition ()
/PDFXRegistryName ()
/PDFXTrapped /False
/Description <<
/CHS <FEFF4f7f75288fd94e9b8bbe5b9a521b5efa7684002000500044004600206587686353ef901a8fc7684c976262535370673a548c002000700072006f006f00660065007200208fdb884c9ad88d2891cf62535370300260a853ef4ee54f7f75280020004100630072006f0062006100740020548c002000410064006f00620065002000520065006100640065007200200035002e003000204ee553ca66f49ad87248672c676562535f00521b5efa768400200050004400460020658768633002>
/CHT <FEFF4f7f752890194e9b8a2d7f6e5efa7acb7684002000410064006f006200650020005000440046002065874ef653ef5728684c9762537088686a5f548c002000700072006f006f00660065007200204e0a73725f979ad854c18cea7684521753706548679c300260a853ef4ee54f7f75280020004100630072006f0062006100740020548c002000410064006f00620065002000520065006100640065007200200035002e003000204ee553ca66f49ad87248672c4f86958b555f5df25efa7acb76840020005000440046002065874ef63002>
/DAN <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>
/DEU <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>
/ESP <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>
/FRA <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>
/ITA <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>
/JPN <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>
/KOR <FEFFc7740020c124c815c7440020c0acc6a9d558c5ec0020b370c2a4d06cd0d10020d504b9b0d1300020bc0f0020ad50c815ae30c5d0c11c0020ace0d488c9c8b85c0020c778c1c4d560002000410064006f0062006500200050004400460020bb38c11cb97c0020c791c131d569b2c8b2e4002e0020c774b807ac8c0020c791c131b41c00200050004400460020bb38c11cb2940020004100630072006f0062006100740020bc0f002000410064006f00620065002000520065006100640065007200200035002e00300020c774c0c1c5d0c11c0020c5f40020c2180020c788c2b5b2c8b2e4002e>
/NLD (Gebruik deze instellingen om Adobe PDF-documenten te maken voor kwaliteitsafdrukken op desktopprinters en proofers. De gemaakte PDF-documenten kunnen worden geopend met Acrobat en Adobe Reader 5.0 en hoger.)
/NOR <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>
/PTB <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>
/SUO <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>
/SVE <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>
/ENU (Use these settings to create Adobe PDF documents for quality printing on desktop printers and proofers. Created PDF documents can be opened with Acrobat and Adobe Reader 5.0 and later.)
>>
/Namespace [
(Adobe)
(Common)
(1.0)
]
/OtherNamespaces [
<<
/AsReaderSpreads false
/CropImagesToFrames true
/ErrorControl /WarnAndContinue
/FlattenerIgnoreSpreadOverrides false
/IncludeGuidesGrids false
/IncludeNonPrinting false
/IncludeSlug false
/Namespace [
(Adobe)
(InDesign)
(4.0)
]
/OmitPlacedBitmaps false
/OmitPlacedEPS false
/OmitPlacedPDF false
/SimulateOverprint /Legacy
>>
<<
/AddBleedMarks false
/AddColorBars false
/AddCropMarks false
/AddPageInfo false
/AddRegMarks false
/ConvertColors /NoConversion
/DestinationProfileName ()
/DestinationProfileSelector /NA
/Downsample16BitImages true
/FlattenerPreset <<
/PresetSelector /MediumResolution
>>
/FormElements false
/GenerateStructure true
/IncludeBookmarks false
/IncludeHyperlinks false
/IncludeInteractive false
/IncludeLayers false
/IncludeProfiles true
/MultimediaHandling /UseObjectSettings
/Namespace [
(Adobe)
(CreativeSuite)
(2.0)
]
/PDFXOutputIntentProfileSelector /NA
/PreserveEditing true
/UntaggedCMYKHandling /LeaveUntagged
/UntaggedRGBHandling /LeaveUntagged
/UseDocumentBleed false
>>
]
>> setdistillerparams
<<
/HWResolution [2400 2400]
/PageSize [612.000 792.000]
>> setpagedevice
|