50720140101001 2

Indian Journal of Information Technology (INDJIT), ISSN 2347-3916 (Print), ISSN
xxxx - xxxx (Online), Volume 1, Issue 1, January - June (2014), Pp-1-13 ©IAEME
1
FEATURE SELECTION IN FREQUENCY DOMAIN FOR
STATIONARY ECG SIGNAL FOR ECG BEAT CLASSIFICATION
R GANESH KUMAR
Research Scholar, Department of CSE, Sathyabama University,
Chennai, Tamil Nadu 600119, INDIA
Dr. Y S KUMARASWAMY
Sr. Professor & HOD, Department of MCA, Dayananda Sagar College of Engineering,
Bangalore 560078, INDIA
ABSTRACT
Arrhythmia is a common term used for cardiac rhythm deviating from normal sinus
rhythm. Many heart diseases are detected through electrocardiograms (ECG) analysis.
Manual analysis of ECG is time consuming and error prone. Thus, an automated system for
detecting arrhythmia in ECG signals is essential. In this paper, investigations are carried out
to evaluate the performance of soft computing techniques for classifying cardiac arrhythmia.
In this study, features are extracted from time series ECG data with Discrete Cosine
Transform (DCT) computing the distance between RR waves. The feature is the beat’s
extracted RR interval. Frequency domain extracted features are classified using Classification
and Regression Tree (CART), Radial Basis Function (RBF), Support Vector Machine (SVM)
and Multilayer Perceptron Neural Network (MLP-NN). Experiments were conducted on the
MIT-BIH arrhythmia database.
Keywords: Cardiac Arrhythmia, Discrete Cosine Transform (DCT), Electrocardiograms
(ECG), RR Waves, Soft Computing.
INDIAN JOURNAL OF INFORMATION TECHNOLOGY
(INDJIT)
ISSN 2347-3916 (Print)
ISSN xxxx - xxxx (Online)
Volume 1, Issue 1, January - June (2014), pp. 01-13
© IAEME: http://www.iaeme.com/INDJIT.asp
INDJIT
© I A E M E

2
1. INTRODUCTION
Cardiovascular diseases kill more people than any other diseases. Arrhythmia is a
collection term for any cardiac rhythm, which deviates from the normal sinus rhythm.
Arrhythmia may be due to a disturbance in impulse formation or conduction, or both, but it is
not always an irregular heart activity [1]. Extracting such signal features helps to explain and
identify different cardiac arrhythmias. The ECG is an important tool to assess a patient’s
cardiac status, as this signal represented the difference between two points on body surface,
versus time [2]. This becomes difficult as ECG signal size is large and the presence of many
noise types contaminates ECG signals [3].Also manual analysis is time consuming and error
prone; thus, extraction of automatic features in ECG signals gains importance.
Many tools / algorithms were proposed for feature extraction from ECG signals like
total least squares based. Most techniques involve high consumption and processing time for
features extraction / classification. Another disadvantage is limited to arrhythmias which are
classified. Hence, a new technique is required to classify many arrhythmias. Also, the
suggested technique should be amenable for real time implementation to ensure its use in
intensive care units / ECG signal collected. Feature selection is the selection of a features
subset occurring in a training set and used as features in classification. Feature selection
serves two things; it ensures efficiency in training and applying a classifier through decrease
of effective feature size. And secondly, feature selection improves classification accuracy
through noisy feature elimination. A noisy feature being one that when represented increases
classification error in new data.
The position and magnitudes of the peaks in ECG such as PR interval, ST interval,
QRS interval and QT interval are commonly used to classify arrhythmia [4, 5]. The features
are either selected from the time domain or frequency domain [6, 7]. Several methods were
suggested for cardiac arrhythmias automatic detection/classification in the literature,
including: artificial immune recognition system with fuzzy weighted [8], neural networks [9],
fuzzy neural networks [10], wavelet transforms [11, 12], combined wavelet transformation
[13], Bayesian classifiers [14] support vector machines [15], and Markov models [16].
This paper investigates performance analysis of soft computing techniques for
classifying cardiac arrhythmia using RBF, SVM and MLP-NN. Experiments were conducted
on the MIT-BIH arrhythmia database. Features are extracted from time series ECG data with
DCT computing the distance between RR waves. The feature is the beat’s extracted
RRinterval. The rest of the paper is organized as follows: Section 2 reviews related works
available in the literature. Section 3 details the methods and materials used. Section 4
reports the results and Section 5 concludes the paper.
2. RELATED WORKS
Karaolis, et al., [17] investigated data-mining systems to assess heart related risk
factors and to CHD events. A total of 528 cases from Paphos district in Cyprus was collected
for examination, a majority of them being affected with an event. Data-mining analysis was
done through the use of c4.5 decision tree algorithm, for the three events applying five
different splitting criteria mentioned above. Risk factors considered for classification rules
analysis include, 1) MI, age, history of hypertension and smoking; 2) PCI family,
hypertension and diabetes histories; 3) for CABG, age, smoking and hypertension history.
These risk factors were also extracted by other examiners. For the MI, PCI, and CABG

3
models, classification efficiency was 66%, 75%, and 75% respectively. Hence, data-mining
system recognizes subjects high and low risk subgroups, and also decides on therapy
selection; (i.e.,) whether medical or surgical.
Kampouraki, et al., [18] examined SVMs to benefit heartbeat time series (i.e., heart
rate signals) classification. Statistical methods and signal analysis techniques extracted signal
features. By comparing other neural network-based classification methods, SVM classifier
performed eave-one-out type cross validation. Regarding other state-of-the-art classifiers,
SVM classifies heart rate signal with reduced low signal-to-noise ratio. Examination was
undertaken for two data sets with many features. The first dataset included long-term ECG
recordings of juvenile and matured healthy subjects while the second set included long-term
ECG recordings of normal subjects classification when compared to the common Heart Rate
Variability (HRV) analysis which was unable to classify similar signs correctly. The issue of
feature selection is considered in feature space by the classifier.
Pasolli, et al., [19] suggested 3 learning strategies to classify ECG Signals by first
selecting from a small, suboptimal training with learning strategies then selecting beat
samples additionally from unlabeled large set data. Samples are labeled manually and added
to the needed training set. The entire process is iterated till the final training set is
constructed. Based on support vector machine classification, the proposed strategies are
developed on margin sampling, posterior probability and query by committee principles. All
proposed strategies show high performance regarding stability and accuracy in relation to a
completely random selection strategy (R).
Faezipour, et al., [20] presented a repetition-detection concept in a patient-adaptive
cardiac profiling system, where a wavelet-based beat-detection mechanism first extracts
fiducial ECG points following which a new local ECG beat classifier profiles patients normal
cardiac behavior. Beat detection is a hybrid of Pan and Tompkins algorithm and wavelet
analysis approach. Experiments were undertaken on a MIT-BIH arrhythmia database proving
that this system was able to detect beats with 99.5% accuracy and identify abnormalities with
97.42% accuracy.
Kim, et al., [21] proposed an ECG holter system, a signal processing technique
comprising of a three-step compression and classification flows with Quad Level Vector
(QLV). QLV improved performance with low-computation in ECG processing for both
compression and classification flows. Classification was applied to both heartbeat
segmentation and R-peak detection. The algorithm’s reliability is upped through a noise
robust test. Average compression ratio was 16.9:1 with 0.641% percentage root mean square
difference value and a 6.4 kbps encoding rate. Without noise, R-peak detection accuracy
performance is 100% and with it, -10dB SNR noise 95.63%. Overall processing cost
reduction was by 45.3% through this method.
Sufi, et al., [22] presented a new procedure executing real-time CVD (Cardiovascular
Disease) classification. This real-time CVD classification automatically informs emergency
personnel/hospital through SMS/MMS/e-mail, when a CVD patient has a life-threatening
cardiac abnormality. The method uses data mining techniques, like attribute selection (only
selects features from compressed ECG) and expectation maximization (EM)-based clustering.
Data mining techniques in a hospital server bear constraints in representing each abnormality.
A patient’s mobile gets these constraints and a rule-based system that recognizes abnormal
beats in real time is used. Experiments undertaken on 50 MIT-BIH ECG entries for cardiac
abnormality detection (ventricular flutter/fibrillation, premature ventricular contraction, atrial
fibrillation etc.) successfully ended up with 97% average accuracy.

4
Neto, et al [23] presented the use of Particle Swarm Optimization (PSO), Wavelets
and neural networks for automatic detection of cardiac arrhythmias based on analysis of the
ECG. The ECG signal has been evaluated in time-frequency domain using wavelets. Wavelet
coefficients were presented as the input of a Multilayer Perceptron (MLP)
Artificial Neural Network (ANN) with three layers, which was trained by the PSO algorithm.
Then the trained network classified the ECG signal in normal signal, atrial fibrillation or
ventricular tachycardia. The database utilized was the MIT-BIH Arrhythmia Database. The
accuracy rate was 97.03%.
Ramirez, et al [24] described a hybrid architecture for classification
of cardiac arrhythmias taking as a source the ECG records MIT-BIH Arrhythmia database.
The Samples were taken from the LBBB, RBBB, PVC and Fusion Paced and
Normal arrhythmias, as well as the normal beats. These were segmented and normalized and
three methods of classification were used: Fuzzy KNN, MLP with Gradient Descent with
Momentum Back propagation, and MLP with Scaled Conjugate Gradient Back propagation.
Mamdani fuzzy inference system was used to combine the outputs of each classifier, and
achieved a very high classification rate of 98%.
Abibullaev, et al [25] presented a technique for the detection and classification
of cardiac arrhythmias using biorthogonal wavelet functions and SVM. The wavelet
transforms was applied to decompose the ECG signal into wavelet scales. Then soft
thresholding technique was applied to denoise and detect important cardiac events in the
signal. Also SVM classifier was used to discriminate the detected events into normal or
pathological ones in the signal. Numeric computations demonstrate that the efficient wavelet
pre-processing provides an accurate estimation of important physiological features of ECG
and moreover it improves the SVM classification performance.
Rua, et al [26] presented the development of two machine learning algorithms on a
32-bit ARM®
Cortex®
M4 microcontroller core from Freescale Semiconductors. Artificial
neural network (ANN) and a SVM were implemented for real time detection of Ventricular
Tachycardia (VT) and Ventricular Fibrillation (VF), and they were compared in terms of
accuracy. In the feature extraction step a Fast Wavelet Transform (FWT) was used; which
was analyzed using the time-frequency characteristics of energy in each sub-band frequency.
For the training and validation algorithms, signals from MIT-BIH database with normal sinus
rhythm, VF and VT in a time window of 2 seconds were used. Validation results achieve test
accuracy of 99.46% by ANN and SVM in VT/VF detection.
Mehdi, et al [27] proposed a system to detect cardiac arrhythmia. The techniques were
used in pattern recognition which includes: signal pre-processing, filter design analysis and
neural network toolboxes in Mat lab environment. Normal and abnormal ECG signals were
tested by using recordings obtained from the Massachusetts Institute of Technology Beth
Israel Hospital (MIT-BIH) arrhythmia database. Supervised Radial Basis Function (RBF)
network architecture was used for filtration, extraction of morphological features
of ECG signal and for classification of abnormal cardiac conditions. Different threshold
techniques were implemented to detect best fudicial points on the derivative of
filtered ECG signal through signal processing. All fudicial points and intervals extracted were
learned to the RBF network for extraction of features through the overall record which was
further processed to ANN for classification.

5
3. MATERIALS AND METHODS
3.1.MIT-BIH Arrhythmia Database
For training and testing of ECG datasets signals for evaluating the performance of the
classifiers, Massachusetts Institute of Technology/Beth Israel Hospital (MIT-BIH)
arrhythmia database is used [28]. Cardiac Arrhythmias for classification include LBBB (Left
Bundle Branch Block), RBBB (Right Bundle Branch Block) and normal beats. Each ECG
beat used is a matrix (275x1) for one ECG lead with every ECG lead with every ECG signal
having five distinct points (P, Q, R, S and T) used for ECG interpretation.
3.2. R-R Interval
It is known that the morphology of the QRS complex along with the instantaneous RR
interval (interval between two successive R peaks) play an important role in the heart
diseases diagnosis [29]. For the classification of an ECG beat the ratio of the RR interval
before it to the one after it is a useful feature. The peaks are detected as follows:
• The moving average of the signal is calculated using a number of records.
• New signal is derived by subtracting the moving average from the original signal.
• Peak of the signal ‘R’ is found.
• Peaks of P, Q, R, S and T are found by the relative position.
The peak amplitude is measured from k line. The k line is given by:
k = max(θi, i = 1, 2, …, 11) + c (1)
Where θ is the greatest amplitude, type of heartbeat and c is a constant. Figure 1 shows a
typical ECG with various peaks and the RR interval.
Fig. 1. ECG with R-R interval pattern

6
3.3.Discrete Cosine Transform (DCT)
Discrete Cosine Transform (DCT) translates time series signal into basic frequency
components. The data is pre-processed to find RR intervals using Fast DCT. The FDCT
[30], of a lsit of n real number, s(x), x = 0,…., n-1, is the list of length n given by the
following equation:
( ) ( ) ( )
( )1
0
2 1
2 c o s
2
n
x
x u
S u n C u s x
n
π−
=
+
= ∑ (2)
where ( )
1
2
2C u
−
= for u=0 or otherwise ( ) 1C u =
The constant factors are chosen so that the basis vectors are orthogonal and normalized. The
inverse cosine transform (IDCT) is given as:
( ) ( ) ( )
( )1
0
2 1
2 c o s
2
n
x
x u
S x n C u s u
n
π−
=
+
= ∑ (3)
3.4.Classifiers
3.4.1. Classification and Regression Trees (CART)
Classification and Regression Tree (CART) based on class label produce output that
is either classification or regression. If class label is categorical, CART provides a
classification tree and regression tree, if it is numeric. A decision tree is created by rules
formed on given data’s attribute values. Each rule is based on dependent attribute values.
This rule is applied to all nodes till the last node is reached, which is the class node [31].
A Classification/regression tree is a binary tree, that given an input X produces output
Ÿ approximating random variable of interest Y, stochastically related to X. this deterministic
mapping is as follows. Associated with the tree’s every internal node is a binary function of
input X and linked to each external node is specific output label Y. Starting at root node,
binary function tests input X. When the result is “0”, left branch is followed; if it results in
“1”, then right branch is followed. This is repeated till an external node/leaf is reached when
associated label Y is the output. The tree minimizes (approximately) anticipated loss
between Y and Ÿ.
3.4.2. Multi-Layer Perceptron Neural Networks (MLP-NN)
Neural networks are analytical techniques modelled analogous to learning process
in the cognitive system and the brain’s neurological functions. They can predict new
observations from earlier ones after executing the learning process by using past data. Neural
networks or widely used Multilayer Perceptron Neural Networks (MLPNNs) are
computational systems comprising of highly interconnected processing elements sets, called
neurons, that process information in response to external stimuli which in turn are
transmitted from one processing element to another through synapses/interconnection, which
are either excitatory or inhibitory. MLPNNs are used in applications like pattern recognition,
and classification [32].

7
3.4.3. Support Vector Machines (SVM)
Support Vector Machine (SVM) maps input vectors to higher dimensional spaces
where maximal separating hyper planes are constructed. Two parallel hyper planes are
constructed on both sides of a data searching hyper plane. The separating hyper plane is that
which maximizes distance between two parallel hyper planes as shown in Figure 2. It is
assumed that larger the margin between parallel hyper planes, better a classifier’s impact.
Though SVMs were originally designed for binary classification, it can also be used in multi-
class classification issues [33].
Fig. 2. Representation of Hyper planes
For a training set consisting of instances (xi; yi), i = 1, …., i, where n
ix R∈ and { }1, 1
l
y ∈ − ,
the support vector machines (SVM) has to solve the following optimization problem [19]:
Polynomial and Radial Basis Function (RBF) is the different kernel functions used in this
study.
1] Polynomial:
k (x, x’
) = (x, x’
)d
(4)
k (x, x’
) = ((x, x’
) + 1)d
(5)
wx+b=1
wx+b=0
wx’+b=-1

8
2] Gaussian Radial Basis Function: A RBF kernel models non-linear relation of class
labels and features Radial Basis functions most commonly with a Gaussian form
(6)
4. RESULTS AND DISCUSSION
Two sets of experiments were conducted, in the first experiment, 153 images were
used and in the second with a dataset of 13300 beats from 20 recordings.
4.1 With dataset 153 instances
A total of 153 images are used for evaluating the performance of classifiers. The
dataset consists of 68 instances of left bunch bundle block, 30 instances of right bunch bundle
block and 56 normal instances. In this paper, a method of energy extraction using Discrete
Cosine Transform was used and RR interval was extracted and used as feature. Using 10 fold
cross validation Classification and Regression tree (CART), Radial Basis Function (RBF),
MLP and SVM was tested. Table 1 tabulates the summary of results of classification
experiments conducted and the classification accuracy is given in Table 2.
Table 1.Summary of Results for various Classifiers
Properties CART
SVM-RBF
kernel
SVM Polynomial
Kernel
MLP NN
Correctly Classified Instances 135 141 95 135
Incorrectly Classified Instances 18 12 58 18
Kappa statistic 0.814 0.876 0.3688 0.8141
Mean absolute error 0.1079 0.0523 0.2527 0.0883
Root mean squared error 0.2688 0.2287 0.5027 0.2679
Relative absolute error 25.54 % 12.38 % 59.84 % 20.91 %
Root relative squared error 58.52 % 49.78 % 109.44 % 58.33 %
Total Number of Instances 153 153 153 153
Table 2. Classification Accuracy
Classification Method
Classification Accuracy in
%
CART 88.2353
SVM – RBF kernel 92.16
SVM - Polynomial kernel 62.09
MLP 88.2353
The Figure 3 shows the root mean squared error (RMSE) obtained by various
classifiers and the Figure 4 charts the classification accuracy achieved.

9
Fig. 3. Root mean squared error (RMSE) achieved Fig. 4. Classification Accuracy
for the classifiers used
It is seen from Figure 4, that the Gaussian RBF kernel performs better than the
Polynomial kernel and MLP. The classification accuracy obtained by RBF kernel is 92.16%.
Table 3 tabulates the precision, recall and F measure achieved by the classifiers. The
precision, recall and F measure for the techniques used is shown in Figure 5 and 6
respectively.
Table 3. Precision, Recall and F-Measure
Classification Method
Precision Recall F measure
CART 0.884 0.882 0.883
SVM – RBF kernel 0.922 0.922 0.922
SVM - Polynomial kernel 0.429 0.621 0.494
MLP 0.882 0.882 0.882
Fig. 5. Precision and Recall Fig. 6. F Measure

10
From table 3 and figure 5 it is evident that the precision and recall of RBF kernel is
much higher than the other classifiers. The SVM-RBF kernel achieves better precision of
4.3%, 4.54% than CART, and MLP respectively. The SVM-RBF kernel achieves higher
recall of 4.54% than CART, 48.47% than SVM-Polykernel and 4.54% than MLP.
4.2 With dataset 13300 beats were taken from 20 recordings
Here 13300 beats were taken from 20 recordings. Four additional Markers including
sex, Age, Family History and smoker / nonsmoker were considered for the classification
problem. The results obtained are tabulated in table 4 and 5.
Table 4 Classification accuracy
Classification accuracy
CART 0.8447
SVM-RBF kernel 0.868
SVM - polykernel 0.8597
MLP 0.8594
Fig. 7 Classification Accuracy Fig. 8 Precision and Recall
From table 4 and figure 7 it is observed that the SVM-RBF kernel achieves better accuracy
by 2.76% than CART, by 0.97% than SVM-Polykernel and by 1% than MLP.
Table 5 Precision and Recall
Precision Recall
CART 0.7958 0.767067
SVM-RBF kernel 0.838333 0.807333
SVM - polykernel 0.823533 0.793733
MLP 0.826633 0.803

11
From table 5 and figure 8 it is shows that the precision and recall is obtained for
various classifiers. The SVM-RBF kernel achieves better precision than other classifiers; by
5.34% than CART, by 1.8% than SVM-Polykernel and by 1.42% than MLP. The SVM-RBF
kernel achieves better recall of 5.25% than CART, 1.71% than SVM-Polykernel and 0.54%
than MLP.
5. CONCLUSION
Cardiac Arrhythmia is an irregular heartbeat due to irregular rhythm caused by the
heartbeat slowing down or increasing. ECG is used to assess heart arrhythmia. Different
arrhythmia types relevant to heart disease are accurately diagnosed through ECG. ECG’s
automatic arrhythmia assessment is widely researched. In this paper, performance analysis of
soft computing techniques for classifying cardiac arrhythmia using RBF, SVM and MLP-NN
is investigated. Experiments were conducted on the MIT-BIH arrhythmia database.
Features like RR interval extracted from ECG are used for arrhythmia classification. For
feature extraction, MIT-BIH time series data is converted to frequency domain using Discrete
Cosine Transform (DCT). Two arrhythmia types, LBB and RBB were classified with normal
beats. Experiments demonstrate Support Vector Machine (SVM) with Gaussian Radial Basis
Function (RBF) kernel achieving classification accuracy of 92.16%. In second part of the
investigation, 13300 beats were taken from 20 recordings. Four additional Markers including
sex, Age, Family History and smoker / nonsmoker were considered for the classification
problem. The SVM-RBF kernel achieves best accuracy of 86.8%.
REFERENCES
1. Sandoe, B. Sigurd (1991): Arrhythmia – A Guide to Clinical Electrocardiology,
Bingen Publishing Partners Verlags GmBH.
2. McSharry, P. E., et al. (2003): A dynamical model for generating synthetic
electrocardiogram signals. IEEE Trans. Biomed. Eng., 50(3), pp. 289–294.
3. Singh, B. N.; Tiwari, A. K. (2006): Optimal selection of wavelet basis function
applied to ECG signal denoising. Dig. Signal Process., 16(3), pp. 275–287.
4. Christov, G., et al. (2006): Comparative study of morphological and time-frequency
ECG descriptors for heartbeat classification. Med. Eng. Phys., 28, pp. 876–887.
5. Chazal, P.; O’Dwyer, M.; Reilly, R. B. (2004): Automatic classification of heart beats
using ECG morphology and heartbeat interval features. IEEE Trans. Biomed. Eng.,
51, pp. 1196–1206.
6. Martinez, J. P.; Olmos, S.; Laguna, P. (2000): Evaluation of a wavelet-based ECG
waveform detector on the QT database. Comput. Cardiol., 27, pp. 81–84.
7. Agante, P. M,; Marques de Sa, J. P. (1999): ECG noise filtering using wavelets with
soft-thresholding methods. Comput. Cardiol., 26, pp. 535–538.
8. Polat, K.; Şahan, S.; Güneş, S. (2006): A new method to medical diagnosis: Artificial
immune recognition system (AIRS) with fuzzy weighted pre-processing and
application to ECG arrhythmia. Expert Systems with Applications, 31(2), pp. 264-269.
9. Wang, J. S., et al. (2012): ECG Arrhythmia Classification Using a Probabilistic
Neural Network with a Feature Reduction Method. Neurocomputing.

12
10. Ceylan, R.; Özbay, Y.; Karlik, B. (2009): A novel approach for classification of ECG
arrhythmias: Type-2 fuzzy clustering neural network. Expert Systems with
Applications, 36(3) pp. 6721- 6726.
11. Özbay, Y.; Ceylan, R.; Karlik, B. (2011): Integration of type-2 fuzzy clustering and
wavelet transform in a neural network based ECG classifier. Expert Systems with
Applications, 38(1), pp. 1004-1010.
12. Khorrami, H.; Moavenian, M. (2010) A comparative study of DWT, CWT and DCT
transformations in ECG arrhythmias classification. Expert systems with Applications,
37(8), pp. 5751-5757.
13. Kim, M. S., et al. (2011): A new method of ECG feature detection based on combined
wavelet transform for u-health service. Biomedical Engineering Letters, 1(2), pp. 108-
115.
14. Jadhav, S. M.; Nalbalwar, S. L.; Ghatol. (2010): ECG arrhythmia classification using
modular neural network model. IEEE EMBS Conference on Biomedical Engineering
and Sciences, pp. 62-66.
15. Homaeinezhad, M. R.; Ghaffari, A.; Rahmani, R. (2012): Multi-lead Discrete
Wavelet-based ECG Arrhythmia Recognition via Sequential Particle Support Vector
Machine Classifiers. Journal of Medical and Biological Engineering, 32(6), pp. 381-
396.
16. Chang, P.C., et al. (2012): Myocardial infarction classification with multi-lead ECG
using hidden Markov models and Gaussian mixture models. J. Applied Soft
Computing, Journal, 12(10), pp. 3165-3175.
17. Karaolis, M. A.; Joseph A. M; Hadjipanayi, D.; Constantinos, S. P. (2010):
Assessment of the Risk Factors of Coronary Heart Events Based on Data Mining
With Decision Trees. IEEE Transactions On Information Technology In Biomedicine,
14(3), pp. 559-566.
18. Kampouraki, A.; George Manis; Nikou, C. (2009): Heartbeat Time Series
Classification With Support Vector Machines. IEEE Transactions On Information
Technology In Biomedicine, 13(4), pp. 512-518.
19. Edoardo Pasolli; Farid Melgani. (2010): Active Learning Methods for
Electrocardiographic Signal Classification. IEEE Transactions On Information
Technology In Biomedicine, 14(6), pp. 1405-1416.
20. Faezipour, M., et al. (2010): A Patient-Adaptive Profiling Scheme for ECG Beat
Classification. IEEE Transactions On Information Technology In Biomedicine, 14(5),
pp. 1153-1165.
21. Hyejung Kim, et al. (2010): ECG Signal Compression and Classification Algorithm
With Quad Level Vector for ECG Holter System. IEEE Transactions On Information
Technology In Biomedicine, 14(1), 93-100.
22. Sufi, F.; Mahmood, A.; Khalil, I. (2011): Diagnosis of Cardiovascular abnormalities
from Compressed ECG: A Data mining based approach. IEEE Trans. Inf. Technol.
Biomed., 15(1), pp. 33–39.
23. Neto, A. B., Nievola, J. C., Figueredo, M. V. M., & Rogal, S. R. (2010, March).
Automatic detection of cardiac arrhythmias using wavelets, neural networks and
particle swarm optimization. In Industrial Technology (ICIT), 2010 IEEE
International Conference on (pp. 194-198). IEEE.

13
24. Ramirez, E., Castillo, O., & Soria, J. (2010, July). Hybrid system for cardiac
arrhythmia classification with fuzzy k-nearest neighbors and Multi Layer Perceptrons
combined by a fuzzy inference system. In Neural Networks (IJCNN), The 2010
International Joint Conference on (pp. 1-6). IEEE.
25. Abibullaev, B., Kang, W. S., Lee, S. H., & An, J. (2010, May). Classification of
cardiac arrhythmias using biorthogonal wavelet preprocessing and SVM.
InNetworked Computing (INC), 2010 6th International Conference on (pp. 1-5).
IEEE.
26. Rua, S., Zuluaga, S. A., Redondo, A., Orozco-Duque, A., Restrepo, J. V., &
Bustamante, J. (2012, September). Machine learning algorithms for real time
arrhythmias detection in portable cardiac devices: microcontroller implementation and
comparative analysis. In Image, Signal Processing, and Artificial Vision (STSIVA),
2012 XVII Symposium of (pp. 50-55). IEEE.
27. MIT-BIH Arrhythmia Database. (2010). [Online]. Available:
www.physionet.org/physiobank/database/mitdb
28. Khoureich, et al. (2011): ECG beats classification using waveform similarity and RR
interval. J. of Medical and Bio. Engineering, 32(6), pp. 417-422.
29. Ahmed, N.; Natarajan, T. (1983): Discrete-Time Signals and Systems, Reston
Publishing Company.
30. Berk, R. A. (2008): Classification and regression trees (CART). Statistical Learning
from a Regression Perspective, New York: Springer, pp. 1- 65.
31. Gardner, M. W.; Dorling, S. R. (1998): Artificial neural networks (the multilayer
perceptron)-A review of applications in the atmospheric sciences. Atmospheric
Environment, 32(14 – 15), pp. 2627- 2636.
32. Gunn, S. R. (1998): Support Vector Machines for Classification and Regression. ISIS
Technical Report, 14.
33. Ch. Ramesh, Dr.N.B. Venkateswarlu, Dr. J.V.R. Murthy, “Fast DCT Algorithm Using
Winograd’s Method” International journal of Computer Engineering & Technology
(IJCET), Volume 3, Issue 1, 2012, pp. 98 - 110, ISSN Print: 0976 – 6367, ISSN
Online: 0976 – 6375
34. M. M. Kodabagi and Hemavati C. Purad, “Script Identification Using Dct
Coefficients” International Journal of Graphics And Multimedia (IJGM), Volume 4,
Issue 1, 2013, pp. 31 - 40, ISSN Print: 0976 – 6448, ISSN Online: 0976 –6456,
35. Kritika Parganiha and Prasanna Kumar Singh, “ECG Interpretation Using Backward
Propagation Neural Networks” International journal of Electronics and
Communication Engineering &Technology (IJECET), Volume 5, Issue 4, 2014, pp.
19 - 24, ISSN Print: 0976- 6464, ISSN Online: 0976 –6472.
36. Shivajirao M. Jadhav, Sanjay L. Nalbalwar and Ashok A. Ghatol, “PERFORMANCE
EVALUATION OF MULTILAYER PERCEPTRON NEURAL NETWORK BASED
CARDIAC ARRHYTHMIA CLASSIFIER” International journal of Computer
Engineering & Technology (IJCET), Volume 3, Issue 2, 2012, pp. 1 - 11, ISSN Print:
0976 – 6367, ISSN Online: 0976 – 6375

50720140101001 2

Recommended

Recommended

More Related Content

What's hot

What's hot (18)

Viewers also liked

Viewers also liked (8)

Similar to 50720140101001 2

Similar to 50720140101001 2 (20)

More from IAEME Publication

More from IAEME Publication (20)

Recently uploaded

Recently uploaded (20)

50720140101001 2