Content uploaded by Saurabh Pal
Author content
All content in this area was uploaded by Saurabh Pal on Jun 10, 2020
Content may be subject to copyright.
56| International Journal of Pharmaceutical Research | Oct - Dec 2020 | Vol 12 | Issue 4
Research Article
Prediction of Heart Disease Using Feature Selection and
Random Forest Ensemble Method
DHYAN CHANDRA YADAV, SAURABH PAL
VBS Purvanchal University, Jaunpur, India
Email ID: dc9532105114@gmail.com, drsaurabhpal@yahoo.co.in
Received: 12.03.19, Revised: 28.05.20, Accepted: 02.06.20
ABSTRACT
The heart is very soft and sensitive part of body by which brain handles blood related system in body. The
heart disease that greatly affects in body as like: pulmonary artery, atalata, enzaina and birth defects included.
Heart disease is mainly related to contraction or blocked blood vessels in the heart. The symptoms of heart
disease depend on the type of disease. Heart disease occurs not only in adults but also in children. The
infection affecting the tissues is known as percarditis. In this, the tissues closest to the heart are affected.
Infections affecting the lining of the heart muscle are known as myocardium .The study of medical datasets is
made very intuitive by machine learning algorithms. The machine learning algorithms provide techniques to
identify dataset attributes and the relationship between them.
In this research work, we used heart disease related information from UCI repository. The dataset contained
1025 Instances with 14 attributes, sick and nonstick patients in target variable. In this paper, we proposed and
analyzed classification accuracy, precision and sensitivity by four tree based classification algorithms: M5P,
random Tree and Reduced Error Pruning with Random forest ensemble method. All the prediction based
algorithms have applied after the features selection of heart patient’s dataset. In this paper, we used three
features based algorithms: Pearson Correlation, Recursive Features Elimination and Lasso Regularization. The
data table analyzed by different feature selection methods for better prediction. All the analysis is done by
three experimental setup; First experiment applied Pearson Correlation on M5P, random Tree, Reduced Error
Pruning and Random forest ensemble method. In the second experiment we used Recursive Features
Elimination and applied on above four tree based algorithms. In the third experiment we used Lasso
Regularization and applied on as above tree based algorithms. After all the performance we analyzed and
calculated classification accuracy, precision and sensitivity.
With the results, we finally concluded that feature selection methods Pearson correlation and Lasso
Regularization with random forest ensemble method provide better results 99% accuracy. We analyzed and
find the random forest ensemble method predicted better result compare to other algorithms in the previous
year’s works.
Keywords: Data mining Tree based Algorithms, Random Forest Ensemble Method, Features Relevant
Method, Features Elimination Method Lasso Regularization Method and Heart Disease.
INTRODUCTION
Research is going on, in large research
institutions to ensure factors related to heart
disease. In some institutions, smoking, age,
high/low blood pressure, obesity, diabetes and
lack of exercise have been included as main
factors for diseases. According to the instructions
of the researchers, it is considered helpful to
identify the disease related to heart disease. Heart
disease is also revealed due to blockage in the
blood vessels, which later expresses the possibility
of heart attack, chest pain or stroke. Valve and
heart muscles are mainly affected in heart
disease. The level of mortality among the world
population by heart disease is quite large.
Cardiovascular data are available in very large
quantities in healthcare. Due to the large amount
of data, it becomes very difficult to study it in
general. But with the help of data mining, large
collections are easily converted into information.
Which shows how the condition of heart disease
has been in children and adults in the past years
and its study also helps in estimating how to
reduce the mortality caused by cardiovascular
diseases in the future. Machine learning
algorithms can improve the treatment of a person
suffering from the disease by comparing its
factors.
ISSN 0975-2366
DOI:https://doi.org/10.31838/ijpr/2020.12.04.013
Dhyan Chandra Yadav et al / Prediction of Heart Disease Using Feature Selection and Random Forest
Ensemble Method
57| International Journal of Pharmaceutical Research | Oct - Dec 2020 | Vol 12 | Issue 4
Fig.1: Representation of blockage in heart .https://images.app.goo.gl/sSdy8qxDpni7fFTj6
Some of the symptoms of heart disease are as
follows:
Heart tightness, pressure and pain.
Chest arms or neck jaw and back pain.
Heart attacks are as follows:
Having a dizzy head.
Face turning brown.
Restlessness.
Trouble breathing, etc.
Heart diseases that are not easily understood
like:
Arrhythmias: Heart beat to be irregular.
Cardiogenic: Shock in person to properly do
not get the blood that person's blood
pressure suddenly collapsed.
Hypoxemia: There is much difficulty in
breathing due to lack of oxygen in the blood.
Pulmonary Edema: Pulmonary edema
involves the accumulation of fluid in or
around the lungs of a heart patient.
DVT or deep win thrombosis: Due to an
excess of blood clots in the veins obstructing
the blood flow.
Mycordial rupture: In this, damage the wall
of heart, of heart patients, which indicates a
major danger.
Ventricular aneurysm: A bulge in the heart
chamber of the afflicted person, causing
difficulty in breathing with blood flow [1].
In this paper, we predict various heart diseases by
variety of feature selection algorithms, applied on
tree based machine learning algorithms. Machine
learning algorithms provide correlations between
various related attributes.
RELATED WORKS
Cai et al., [2020], discussed about heart
arrhythmia and 12 lead electro cardiogram. They
used one dimensional deep densely connected
neural network to detect artial fibrillation. Authors
found accuracy, sensitivity and specificity (99.35) ,
(99.19) and (99.44) respectively the results on test
dataset [2].
Buettner et al., [2020], considered
electroencephalography recording of heart
patients. Authors explained five granular divisions
of EEG spectra by machine learning classifiers.
They used Random Forest algorithm to make a
balance between paranoid schizophrenic and
non- schizophrenic persons with (96.77) percent
classification accuracy [3].
Magesh and Swarnalatha [2020] analyzed about
cardiovascular ailment centers in ruler side. They
found some risk factors or illness in coronary
disease by smoking. Authors examined target
level distribution from samples and identify
features through entropy. They used Random
Forest in the prediction of heart disease and
found accuracy (89.30) percent with cluster based
Dhyan Chandra Yadav et al / Prediction of Heart Disease Using Feature Selection and Random Forest
Ensemble Method
58| International Journal of Pharmaceutical Research | Oct - Dec 2020 | Vol 12 | Issue 4
DT learning and (76.70) percent without cluster
based DT learning [4].
Shen et al., [2020], discussed about a trial
fibrillation arrhythmia. They used neural network
and manual extraction features on the prediction
of a trial fibrillation. Authors used decision tree,
Random Forest, GBDT, XG Boost, LightGBM and
find (99.91) percent accuracy by stacking model
[5] .
Kar et al., [2020], observed the condition of heart
of a patient by electrocardiogram signal. They
analyzed ECG, signal by continues and discrete
wavelet transforms. Authors used time interval,
statistical features and classify irregular
heartbeats. They calculated K-NN, DT-CWT
features and find (98.92) percent classification
accuracy [6].
Harimoorthy and Thangavelu [2020], discussed
about hidden pattern in chronic kidney disease.
They reduced some features from chronic kidney
disease and improved in SVM Redial biaskernal.
Authors compared SVMRBK with (SVM-Linear,
SVM-Polynomial, Random forest and Decision
Tree) and find improvement in accuracy of SVM-
RBK (98.3) percent, (98.7) percent AND (89.9)
percent [7].
Miled et al., [2020], analyzed electronic medical
record of diagnosis, prescriptions and medical
notes. They used machine learning algorithms to
identify dementia and non dementia cases and
predict the fact. They developed Random forest
algorithms in three EMR dataset and find (77.43)
percent accuracy [8].
METHODOLOGY
In this phase, we have described the heart
patient’s attributes and applied algorithms. We
visualized all the attributes measured their
distribution and considered applied algorithms
with experimental setup.
Data Description:
In this paper, we organized dataset from
recorded UCI website. The dataset is related with
heart patients and measure the distribution of
heart disease patient attributes. The class
distribution, box whisker plotting and visualizing
of dataset have discussed by Python language. In
this dataset, we used 1025 instances and 14
attributes.
Class Distribution
target
0 499
1 526
dtype: int64
The class level distribution of dataset represents
how much TRUE /FALSE positive in the target
variable.
Box and Whisker Plots
Each attributes and their numeric values provide
help in disease prediction. By the help of box and
whisker, we have implemented the heart disease
attributes in brief and measured each attributes
distribution [9].
Fig.2: Representation of Box and Whisker plotting of heart disease attributes
Dhyan Chandra Yadav et al / Prediction of Heart Disease Using Feature Selection and Random Forest
Ensemble Method
59| International Journal of Pharmaceutical Research | Oct - Dec 2020 | Vol 12 | Issue 4
Histograms Representation
The histograms are graphical representation of
each attributes separately in the graph and
measure their visualization [10]. In this paper, we
used heart disease 14 attributes. Each attributes
represents their valuable representation in whole
dataset.
Fig.3: Representation of Histogram plotting of heart disease attributes
Algorithms:
M5P algorithm
Fig.4: Representation of M5P algorithms for 40 instances and 14 attributes of heart disease
In this paper, we used this model for numeric
prediction with the results; at the leaf find the
class values of instances. The work of this
algorithm as an expert to search on node, each
node due to prediction [11]. For example we
considered some instances for heart disease and
their performance.
Dhyan Chandra Yadav et al / Prediction of Heart Disease Using Feature Selection and Random Forest
Ensemble Method
60| International Journal of Pharmaceutical Research | Oct - Dec 2020 | Vol 12 | Issue 4
Random Tree Algorithm
Fig.5: Random Tree algorithms for 40 instances and 14 attributes of heart disease
Random tree algorithm used for randomly
selection of attributes in decision node [12]. The
main work of this algorithm to measure the
performance of class predictions with their
probability and try to improve prediction
performance at each node.
Reduced Error Pruning
Fig.6: Representation of REP algorithms for 40 instances and 14 attributes of heart disease
The performance of reduced error pruning is
based on C4.5 algorithms [13]. In this experiment
we used, batch size=100, max Depth=-1,
minimum variance probability=0.001 num
folds=3 and seed=1 for fast learner on each
node. The main objective of algorithm to reduce
error pruning on each node of the tree.
Formula Representation:
Table 1: Computational Formula for Prediction [14]
S.No.
Measure
Formula
1.
Accuracy
(TP+TN)/(TP+TN+FP+FN)
2.
Sensitivity
(TP)/(TP+TN)
3.
Specificity
(TN)/(TN+FP)
Dhyan Chandra Yadav et al / Prediction of Heart Disease Using Feature Selection and Random Forest
Ensemble Method
61| International Journal of Pharmaceutical Research | Oct - Dec 2020 | Vol 12 | Issue 4
Proposed Ensemble Method:
In this research paper, we used random forest as
an ensemble method. The Random Forest is a
powerful decision making tree ensemble method
[15]. The main property of this algorithm is to
select decision randomly from other tree. In this
paper, we used M5P tree, Random tree and Error
Reduced Pruning tree with Random Forest
Ensemble method. After the features selection
trained on (75%) dataset and the test on (25%)
with tree algorithms with ensemble method. The
final prediction has measured by average voting
algorithms. In this experiment, we used bag size=
100%, batch size==100 and seed = 1 for better
prediction.
Fig.7: Proposed Model of Random Forest algorithm as a ensemble model
RESULTS
In this paper, we used various features selection
method and applied on various machine learning
algorithms for better prediction.
Pearson correlation with output variables, find
score of some features: cp, exang, oldpeak
and target (.43), (.43), (.43) and 1.00
respectively, these are highly correlated
features.
The Pearson correlation features selection
method with Random Forest Ensemble
method calculated (99.9%) accuracy.
The Recursive Features selection method
provide optimal number of features:12 and
the score with 12 features: 0.54
The Recursive Features selection method
applied with Random Forest and calculated
(94.12%) accuracy.
Lasso Regularization by lassoCV() calculated:
Best alpha= 0.0048, Best score =.51
In the performance Lasso Model avoid some
features: fbs, chol and age
Lasso Regularization with Random Forest
ensemble method finds (99.9%) accuracy.
DISCUSSION
In this section, we discussed about all feature
selection performance with machine learning
algorithms:
Dhyan Chandra Yadav et al / Prediction of Heart Disease Using Feature Selection and Random Forest
Ensemble Method
62| International Journal of Pharmaceutical Research | Oct - Dec 2020 | Vol 12 | Issue 4
Fig.8: Representation of Pearson Correlation for heart disease attributes
The matrix has an absolute value (0.3) with the
output variable and gives the results for highly
correlate attributes [16]. We used Pearson
Correlation matrix with output variable and select
the highly correlated features as:
Table 2: Valuable Score with Features of Pearson Correlation
Features
Correlation
cp
0.434854
thalach
0.422895
exang
0.438029
oldpeak
0.438441
slope
0.345512
ca
0.382085
thal
0.337838
target
1.000000
Name: target, dtype: float64
Table 3: Measure Prediction Performance for With/ Without PC by Tree Classifiers
Algorithms
FSM (Without PC)
FSM (With PC)
Specificity
Sensitivity
Accuracy
Specificity
Sensitivity
Accuracy
M5PT
41.2
94.5
89.3
83.1
91.2
93.4
RT
42.7
95.7
91.2
92.6
95.3
95.2
REPT
50.6
95.8
91.5
89.7
96.6
96.6
RFT
62.3
95.1
93.8
90.3
99.6
99.9
For the table.3, it is clear that PC= Pearson
Correlation feature selection on RFT = Random
Forest calculated highest accuracy and sensitivity.
We Initialized Recursive Features Elimination
model for fitting the data to model and find the
result as:
[False True True False False False False False
True True True True True]
[3 1 1 6 7 4 2 5 1 1 1 1 1]
Recursively remaining heart attributes and
building a model on those heart attributes remain
in table. All the True are most relevant features in
dataset and False are irrelevant features [17] .
Dhyan Chandra Yadav et al / Prediction of Heart Disease Using Feature Selection and Random Forest
Ensemble Method
63| International Journal of Pharmaceutical Research | Oct - Dec 2020 | Vol 12 | Issue 4
So calculate no of features with variable to store
the optimum features as:
Optimum number of features: 12; Score with 12
features: 0.541462
By the experiment find the transforming data
using RFE and fitting the data to model as :
Index(['age', 'sex', 'cp', 'trestbps', 'fbs', 'restecg',
'thalach', 'exang', 'oldpeak', 'slope', 'ca', 'thal'],
dtype='object')
Table 4: Measure Prediction Performance for With/ Without RFE by Tree Classifiers
Algorithms
FSM (Without RFE)
FSM (With RFE)
Specificity
Sensitivity
Accuracy
Specificity
Sensitivity
Accuracy
M5PT
52.4
83.7
74.8
92.3
82.5
80.6
RT
53.3
84.2
80.3
81.7
84.7
86.3
REPT
41.2
84.3
85.8
78.6
84.8
85.6
RFT
51.1
84.8
87.9
91.6
98.8
98.2
For the table.4, it is clear that RFE= Recursive
Features Elimination on RFT = Random Forest
calculated highest accuracy and sensitivity.
In the Lasso regularization model, we used CV
based function for better feature importance[18] .
reg = LassoCV()
Best alpha using built-in LassoCV: 0.004860
Best score using built-in LassoCV: 0.513496
Text(0.5, 1.0, 'Feature importance using Lasso
Model') and reduced some less important as: fbs,
chol and age
Fig.9: Lasso regularization model for features selection
If the features are irrelevant then Lasso penalizes,
with the results, we find features: fbs, chol and
age are penalized. The last top and bottom
features are highly related with each other.
Table 5: Measure Prediction Performance for with / Without Lasso regularization by Tree
Classifiers
Algorithms
Without LRM
With LRM
Specificity
Sensitivity
Accuracy
Specificity
Sensitivity
Accuracy
M5PT
63.5
72.2
83.8
75.2
73.2
89.8
RT
31.7
78.3
83.7
79.5
79.8
91.9
REPT
61.5
79.4
78.7
91.5
63.2
79.4
RFT
73.5
76.8
88.9
91.3
97.1
99.9
Dhyan Chandra Yadav et al / Prediction of Heart Disease Using Feature Selection and Random Forest
Ensemble Method
64| International Journal of Pharmaceutical Research | Oct - Dec 2020 | Vol 12 | Issue 4
For the table.5, it is clear that LRM= Lasso regularization model on RFT = Random Forest calculated
highest accuracy and sensitivity.
Fig.10: Representation of Accuracy by Tree based Algorithms in heat disease
With the results, table.3, 4 &5 represents the
comparison of other M5P, RT and REPT
algorithms. Fig.10., represents the above all the
obtained experiments and we find that the highest
accuracy and Sensitivity of Random Forest
ensemble method.
Table 6: Representation of Previous Year Paper Accuracy Score
Authors
Instances
Algorithms
Accuracy
Hui et al.,[2012] [19]
9800
SRBC, ICA & RR
98.35
Martis et ai., [2013][20]
110,094
ICA, DWT & PNN
99.28
Ince et al.,[2015][21]
100,389
CNN & BP
98.90
Naomin et al.,[2016][22]
110,094
NN, SVM & PCA
98.90
Hua et al., [2017][23]
90808
SVM & Weighted RR
98.46
Oh et al.,[2017][24]
109949
CNN
94.47
Yildirim et al.,[2018][25]
7376
DBLSTM
99.39
Yildirim et al., [2019][26]
100,022
CAE-LSTM
99.23
Haotien et al.,[2020][27]
100630
CNN
99.06
We have studies near 2012 -2020 and find the
highest accuracy near about (99%). In the work,
we have compared different algorithms
individually but did not cover (100%) accuracy. In
this research work, we have tried to test with
different features selection method applied on
different machine learning tree classifiers
algorithms and finally find Random Forest
ensemble method provide better result (99.9%)
accuracy.
CONCLUSION
In this research paper, we used Pearson
Correlation, Recursive Features Elimination and
Lasso Regularization, features selection methods
and applied on Machine learning tree based
classifiers algorithms: M5P, Random Tree and
Reduced Error Pruning with Random Forest
ensemble method. In this analysis, we evaluated
the value of classification accuracy, precision,
sensitivity and ROC. We have used UCI
Repository dataset for 1025 instances and 14
attributes. In this research, we Identify whether a
person is suffering from heart problem are not in
heart disease machine learning algorithms
provide various way to implement the medical
data set. In this research work, the important
features were identified by Pearson correlation,
Recursive Features Elimination and Lasso
Regularization with the selected important
features we examine with improved, Random
Forest, Random Tree, Reduced Error Pruning and
M5P classifiers algorithms in heart disease. With
the results, we find that improved Random Forest
ensemble method with batch size (100) and seed
(1) provide batter accuracy compare to other.
Since this work is based on recorded data from
UCI repository, for future work planning , we will
Dhyan Chandra Yadav et al / Prediction of Heart Disease Using Feature Selection and Random Forest
Ensemble Method
65| International Journal of Pharmaceutical Research | Oct - Dec 2020 | Vol 12 | Issue 4
try train and test on huge medical data set with
more than one ensemble method and try to
improvement in their performance.
Conflict of Interest
Authors have no conflict of Interest.
Funding
This study was not funded.
Acknowledgements
The author is grateful to Veer Bahadur Singh
Purvanchal University Jaunpur, Uttar Pradesh, for
Providing financial support to work as Post
Doctoral Research Fellowship.
REFERENCES
1. Lui, C. K., Kerr, W. C., Li, L., Mulia, N., Ye, Y.,
Williams, E., ... & Lown, E. A. (2020). Lifecourse
Drinking Patterns, Hypertension, and Heart
Problems Among US Adults. American Journal of
Preventive Medicine.
2. Cai, W., Chen, Y., Guo, J., Han, B., Shi, Y., Ji, L.,
... & Luo, J. (2020). Accurate detection of atrial
fibrillation from 12-lead ECG using deep neural
network. Computers in biology and medicine, 116,
103378.
3. Buettner, R., Beil, D., Scholtz, S., & Djemai, A.
(2020, January). Development of a machine
learning based algorithm to accurately detect
schizophrenia based on one-minute EEG
recordings. In Proceedings of the 53rd Hawaii
International Conference on System Sciences.
4. Magesh, G., & Swarnalatha, P. (2020). Optimal
feature selection through a cluster-based DT
learning (CDTL) in heart disease
prediction. Evolutionary Intelligence, 1-11.
5. Shen, M., Zhang, L., Luo, X., & Xu, J. (2020,
January). Atrial Fibrillation Detection Algorithm
Based on Manual Extraction Features and
Automatic Extraction Features. In IOP Conference
Series: Earth and Environmental Science (Vol. 428,
No. 1, p. 012050). IOP Publishing.
6. Kar, N., Sahu, B., Sabut, S., & Sahoo, S. (2020).
Effective ECG Beat Classification and Decision
Support System Using Dual-Tree Complex
Wavelet Transform. In Advances in Intelligent
Computing and Communication (pp. 366-374).
Springer, Singapore.
7. Harimoorthy, K., & Thangavelu, M. (2020). Multi-
disease prediction model using improved SVM-
radial bias technique in healthcare monitoring
system. Journal of Ambient Intelligence and
Humanized Computing, 1-9.
8. Miled, Z. B., Haas, K., Black, C. M., Khandker, R.
K., Chandrasekaran, V., Lipton, R., & Boustani, M.
A. (2020). Predicting dementia with routine care
EMR data. Artificial Intelligence in Medicine, 102,
101771.
9. Norris, D. J. (2020). Introduction to machine
learning (ML) with the Raspberry Pi (RasPi).
In Machine Learning with the Raspberry Pi (pp. 1-
47). Apress, Berkeley, CA.
10. del Rio, A. A. H., Cuevas, E., & Zaldivar, D.
(2020). Multi-level Image Thresholding
Segmentation Using 2D Histogram Non-local
Means and Metaheuristics Algorithms.
In Applications of Hybrid Metaheuristic Algorithms
for Image Processing (pp. 121-149). Springer,
Cham.
11. Mudali, P., Roopa, J., Raju, M. G., & Yadav, A.
(2020). Analysis of Parallel M5P and Random
Forest Regression for Visualization of Traffic
Behavior. In Computational Intelligence in Pattern
Recognition (pp. 231-241). Springer, Singapore.
12. Nachmias, A. (2020). Uniform Spanning Trees of
Planar Graphs. In Planar Maps, Random Walks and
Circle Packing (pp. 89-103). Springer, Cham.
13. Thomas, T., Vijayaraghavan, A. P., & Emmanuel,
S. (2020). Applications of Decision Trees.
In Machine Learning Approaches in Cyber Security
Analytics (pp. 157-184). Springer, Singapore.
14. Shehab, M. (2019). Artificial Intelligence in Diffusion
MRI: Enhanced Cuckoo Search Algorithm with
Metaheuristic Components for Extracting the
Maxima of the Orientation Distribution
Function (Vol. 877). Springer Nature.
15. Sniatala, P., Amini, M. H., & Boroojeni, K. G.
(2020). A Novel Fault Tolerant Random Forest
Model Using Brooks–Iyengar Fusion.
In Fundamentals of Brooks–Iyengar Distributed
Sensing Algorithm (pp. 159-165). Springer, Cham.
16. Jain, G., Mahara, T., & Tripathi, K. N. (2020). A
Survey of Similarity Measures for Collaborative
Filtering-Based Recommender System. In Soft
Computing: Theories and Applications (pp. 343-
352). Springer, Singapore.
17. Kumari, P., & Haider, M. T. U. (2020). Sentiment
Analysis on Aadhaar for Twitter Data—A Hybrid
Classification Approach. In Proceeding of
International Conference on Computational Science
and Applications (pp. 309-318). Springer,
Singapore.
18. Chen, Q., & Huang, L. (2020). Research on
Prediction Model of Gas Emission Based on
Lasso Penalty Regression Algorithm. In Artificial
Intelligence in China (pp. 165-172). Springer,
Singapore.
19. Huang, H. F., Hu, G. S., & Zhu, L. (2012). Sparse
representation-based heartbeat classification
using independent component analysis. Journal of
medical systems, 36(3), 1235-1247.
20. Martis, R. J., Acharya, U. R., & Min, L. C. (2013).
ECG beat classification using PCA, LDA, ICA and
discrete wavelet transform. Biomedical Signal
Processing and Control, 8(5), 437-448.
21. Kiranyaz, S., Ince, T., & Gabbouj, M. (2015). Real-
time patient-specific ECG classification by 1-D
convolutional neural networks. IEEE Transactions
on Biomedical Engineering, 63(3), 664-675.
Dhyan Chandra Yadav et al / Prediction of Heart Disease Using Feature Selection and Random Forest
Ensemble Method
66| International Journal of Pharmaceutical Research | Oct - Dec 2020 | Vol 12 | Issue 4
22. Elhaj, F. A., Salim, N., Harris, A. R., Swee, T. T., &
Ahmed, T. (2016). Arrhythmia recognition and
classification using combined linear and nonlinear
features of ECG signals. Computer methods and
programs in biomedicine, 127, 52-63.
23. Chen, S., Hua, W., Li, Z., Li, J., & Gao, X. (2017).
Heartbeat classification using projected and
dynamic features of ECG signal. Biomedical Signal
Processing and Control, 31, 165-173.
24. Acharya, U. R., Oh, S. L., Hagiwara, Y., Tan, J. H.,
Adam, M., Gertych, A., & San Tan, R. (2017). A
deep convolutional neural network model to
classify heartbeats. Computers in biology and
medicine, 89, 389-396.
25. Yildirim, Ö. (2018). A novel wavelet sequence
based on deep bidirectional LSTM network
model for ECG signal classification. Computers in
biology and medicine, 96, 189-202.
26. Yildirim, O., Baloglu, U. B., Tan, R. S., Ciaccio, E.
J., & Acharya, U. R. (2019). A new approach for
arrhythmia classification using deep coded
features and LSTM networks. Computer methods
and programs in biomedicine, 176, 121-133.
27. Wang, H., Shi, H., Chen, X., Zhao, L., Huang, Y.,
& Liu, C. (2020). An Improved Convolutional
Neural Network Based Approach for Automated
Heartbeat Classification. Journal of Medical
Systems, 44(2), 35.