Classification of Cardiotocogram Data using Neural Network based Machine Learning Technique

Abstract

Cardiotocography (CTG) is a simultaneous recording of fetal heart rate (FHR) and uterine contractions (UC). It is one of the most common diagnostic techniques to evaluate maternal and fetal well-being during pregnancy and before delivery. By observing the Cardiotocography trace patterns doctors can understand the state of the fetus. There are several signal processing and computer programming based techniques for interpreting a typical Cardiotocography data. Even few decades after the introduction of cardiotocography into clinical practice, the predictive capacity of the these methods remains controversial and still inaccurate. In this paper, we implement a model based CTG data classification system using a supervised artificial neural network(ANN) which can classify the CTG data based on its training data. According to the arrived results, the performance of the supervised machine learning based classification approach provided significant performance. We used Precision, Recall, F-Score and Rand Index as the metric to evaluate the performance. It was found that, the ANN based classifier was capable of identifying Normal, Suspicious and Pathologic condition, from the nature of CTG data with very good accuracy.

Full text

International Journal of Computer Applications (0975 – 888)
Volume 47– No.14, June 2012
19
Classification of Cardiotocogram Data using Neural
Network based Machine Learning Technique
Sundar.C
Christian College of Engineering
Technology,
Oddanchatram – 624619
M.Chitradevi
PRIST University
Trichy Campus – Tamilnadu
Tiruchirappalli – 620 009
Dr.G.Geetharamani
Anna University of Technology
Tiruchirappalli - 620024
ABSTRACT
Cardiotocography (CTG) is a simultaneous recording of fetal
heart rate (FHR) and uterine contractions (UC). It is one of the
most common diagnostic techniques to evaluate maternal and
fetal well-being during pregnancy and before delivery. By
observing the Cardiotocography trace patterns doctors can
understand the state of the fetus. There are several signal
processing and computer programming based techniques for
interpreting a typical Cardiotocography data. Even few
decades after the introduction of cardiotocography into
clinical practice, the predictive capacity of the these methods
remains controversial and still inaccurate. In this paper, we
implement a model based CTG data classification system
using a supervised artificial neural network(ANN) which can
classify the CTG data based on its training data. According to
the arrived results, the performance of the supervised machine
learning based classification approach provided significant
performance. We used Precision, Recall, F-Score and Rand
Index as the metric to evaluate the performance. It was found
that, the ANN based classifier was capable of identifying
Normal, Suspicious and Pathologic condition, from the nature
of CTG data with very good accuracy.
Keywords
Multidimensional Data Classification, Medical Data
Classification, Cardiotocography, CTG, fetal heart rate, FHR.
uterine contractions, UC, ANN.
1. INTRODUCTION
Data Mining (DM) and the technology of Knowledge
Discovery from Data (KDD) has brought many new
developments, methods, and technologies in the recent
decade. Also the improvement of integration of techniques
and the application of data mining techniques had contributed
in handling of new kinds of data types and applications.
However, the field of data mining and its application in
medical domain is still young enough so that the possibilities
of the application are still limitless [20].
One of the major challenges in medical domain is the
extraction of comprehensible knowledge from medical
diagnosis data such as CTG data. In this information era, the
use of machine learning tools in medical diagnosis is
increasing gradually. This is mainly because the effectiveness
of classification and recognition systems has improved in a
great deal to help medical experts in diagnosing diseases[21].
1.1 Cardiotocography (CTG)
Cardiotocography (CTG) is a technical means of recording
the fetal heart rate (FHR) and the uterine contractions (UC)
during pregnancy, typically in the third trimester to evaluate
maternal and fetal well-being. FHR patterns are observed
manually by obstetricians during the process of CTG analysis.
In the recent past fetal heart rate baseline and its frequency
analysis has been taken in to research on many aspects [2],[6].
Fetal heart rate (FHR) monitoring is mainly used to find out
the amount of oxygen a fetus is acquiring during the time of
labor [7]. Even then death and long term disablement occurs
due to hypoxia during delivery. More than 50% of these
deaths were caused by not recognizing the abnormal FHR
pattern, even after recognizing not communicating the same
without knowing the seriousness and the delay in taking
appropriate action [7]. The currently proposed computation
and datamining techniques for FHR can be used for analyzing
and classifying the CTG data to avoid human mistakes and
helps the doctors to take a decision.
2. PROBLEM DEFINITION
Cardiotocography (CTG), consisting of fetal heart rate (FHR)
and tocographic (TOCO) measurements, is used to evaluate
fetal well-being during the delivery. Since 1970 many
researchers have employed different methods to help the
doctors to interpret the CTG trace pattern from the field of
signal processing and computer programming [2]. They have
supported doctors with interpretations in order to reach a
satisfactory level of reliability so as to act as a decision
support system in obstetrics. Up to now, none of them has
been adopted worldwide for everyday practice (Van
Geijnt, 1996). There is currently no consensus on the best
methodology for baseline estimation in computer analysis of
cardiotocographs [2]. More than 30 years after the
introduction of antepartum cardiotocography into clinical
practice, the predictive capacity of the method remains
controversial. In a review of lot of articles published on this
subject, it was found that its reported sensitivity varies
between 2 and 100%, and its specificity between 37 and 100%
[5]. So, in this work, we are going to evaluate some of the
statistical, machine learning and datamining techniques for the
classification of CTG data.
Classification can be viewed as a supervised learning
scenario. Here a training data set of records is accompanied
by class labels. New data can be classified based on the
training set by generating descriptions of the classes. In
addition to the training set, there is also a test data set which is
used to determine the effectiveness of a classification. In
principle, the popular neural network can be trained to
recognize the data directly. However, a simple network can be
very complex and difficult to train. Further, if the dimension
of the input data is high, then the training process will
consume very lot of time and the accuracy of classification
also vary with the increase of dimension in the training data.
Generally, the techniques used in the neural network systems
will depend on the application of the system.

International Journal of Computer Applications (0975 – 888)
Volume 47– No.14, June 2012
20
As means of data collection have become more capable, the
need for non-linear modeling techniques has become more
and more apparent. Traditional statistical methods rely on an
assumption of linearity. However, since most of the data
collected concerns, or is the result of, human behavior, and
humans rarely behave linearly, methods that assume linear
separability are ultimately doomed to failure. Furthermore,
data collection streams are broadening. The number of
variables of concern to modelers has increased by at least an
order of magnitude. Traditional methods simply were not
designed to work with one hundred or more variables.
In answer to this, the last decade has seen the emergence of
neural networks as a means of non-linear modeling. These
devices resulted from the efforts of a number of cognitive
scientists to mimic learning and memory in the human brain.
The back-propagation neural network in particular has proven
successful in creating useful models from large masses of
complex data. The algorithm has been successfully applied in
a variety of settings including direct marketing, intelligence
and process control. Because of its pattern recognition nature
it has proven robust with respect to missing data and other
data irregularities.
2.1 The Medical Background Of
Cardiotocography (CTG)
Cardiotocography is a medical test conducted during
pregnancy that records fetal heart rate(FHR) and uterine
contractions. The tests may be conducted by either internal or
external methods. In internal testing, a catheter is placed in the
uterus after a specific amount of dilation has taken place.
With external tests, a pair of sensory nodes is affixed to the
mother's stomach. The CTG trace generally shows two lines.
The upper line is a record of the fetal heart rate in beats per
minute. The lower line is a recording of uterine contractions
from the TOCO [4].
2.1 Baseline Heart Rate
The baseline heart rate helps to evaluate the healthy
functioning of the cardiovascular system. The baseline fetal
heart rate is determined by approximating the mean FHR
rounded to increments of 5 beats per minute (bpm) during a
10-minute window, excluding accelerations and decelerations
and periods of marked FHR variability (greater than 25 bpm.
Abnormal baseline is termed bradycardia and tachycardia.
The fluctuations are visually quantitated as the amplitude of
the peak- to-trough in bpm. Using this definition, the baseline
FHR variability is categorized by the quantitated amplitude
as:
Absent- undetectable
Minimal- greater than undetectable, but less than or equal
to 5 bpm
Moderate- 6 bpm - 25 bpm
Marked- greater than 25 bpm
Bradycardia: It is the resting heart rate of under 60 beats
per minute, though it is seldom symptomatic until the rate
drops below 50 beats/min. It may cause cardiac arrest in some
patients
Tachycardia: It typically refers to a heart rate that
exceeds the normal range for a resting heart rate (heart rate in
an inactive or sleeping individual). It can be dangerous
depending on the speed and type of rhythm.
Type 1 (early)
This occurs during the peak of the uterine contraction. It will
be uniform, repetitive, periodic slowing of FHR with onset
early in the contraction and return to baseline at the end of the
contraction. The reasons behind this may be fetal head
compression, cord compression or early hypoxia. This occurs
in first and second stage labor with decent of the head [4].
This is synchronous with uterine contraction
Type 2 (late)
This occurs after the peak of the uterine contraction. It will
also be uniform, repetitive, slowing of FHR with onset mid to
end of the contraction and nadir more than 20 seconds after
the peak of the contraction and ending after the contraction. If
the lag time is high seriousness is also high. This is also
synchronous with uterine contraction. Mx: a fetal pH
measurement is mandatory [4].
Type 3 (variable)
This is variable, repetitive, periodic slowing of FHR with
rapid onset and recovery. Variable and isolated time
relationships with contraction cycles may occur. In some
cases, they resemble other types of deceleration patterns in
timing and shape. If they occur consistently, there is a chance
of fetal hypoxia. This is unrelated to uterine contractions.
Mx: check fetal pH if the pattern persists after turning the
patient on her side (or if other adverse features are present)
[4].
3. CLASSIFICATION USING
ARTIFICIAL NEURAL NETWORK
3.1 ANN Based Classification
Here in this classification, we use supervised learning by
using a set of training data which is accompanied by class
labels. When a new data arrive, then classification of that data
will be done based on the training set by generating
descriptions of the classes. In addition to training set we also
have a test data set that is used to determine the effectiveness
of a classification. In general, commonly used and popular
neural networks can be trained to recognize the data directly,
whereas in simple networks there is a chance of the system
being complex and training may be difficult. The time taken
and the accuracy of classification depend on the dimension of
the input given and also on the dimension in the training data.
For input data with high dimension, the process will take a
longer time.
3.2 Structuring the Network
The number of layers and the number of processing elements
per layer are important decisions. These parameters to a feed
forward, back-propagation topology are also the most ethereal
- they are the "art" of the network designer. There is no
quantifiable, best answer to the layout of the network for any
particular application. There are only general rules picked up
over time and followed by most researchers and engineers
applying this architecture to their problems.
Rule One: As the complexity in the relationship between
the input data and the desired output increases, the number of
the processing elements in the hidden layer should also
increase.
Rule Two: If the process being modeled is separable into
multiple stages, then additional hidden layer(s) may be

International Journal of Computer Applications (0975 – 888)
Volume 47– No.14, June 2012
21
required. If the process is not separable into stages, then
additional layers may simply enable memorization of the
training set, and not a true general solution effective with
other data.
Fig 1: Feed forward Network
Rule Three: The amount of training data available sets an
upper bound for the number of processing elements in the
hidden layer(s). To calculate this upper bound, use the number
of cases in the training data set and divide that number by the
sum of the number of nodes in the input and output layers in
the network. Then divide that result again by a scaling factor
between five and ten. Larger scaling factors are used for
relatively less noisy data. If you use too many artificial
neurons the training set will be memorized. If that happens,
generalization of the data will not occur, making the network
useless on new data sets.
A single-layer network of S logsig neurons having R inputs is
shown below in full detail on the left and with a layer diagram
on the right [16].
Feed forward networks often have one or more hidden layers
of sigmoid neurons followed by an output layer of linear
neurons [11], Multiple layers of neurons with nonlinear
transfer functions allow the network to learn nonlinear and
linear relationships between input and output vectors. The
linear output layer lets the network produce values outside the
range -1 to +1. On the other hand, if you want to constrain the
outputs of a network (such as between 0 and 1), then the
output layer should use a sigmoid transfer function.
3.3 The ANN based CTG Data
Classification System
The Fig. 2 shows the ANN based CTG data Classification
system.
The Metrics Used for the Evaluation
Precision, recall and F-Score are computed for every (class,
cluster) pair. But Rand index is a metric which will consider
all the classes and the clusters as the whole.
Rand Index
The Rand index or Rand measure is a commonly used
technique for measure of such similarity between two data
clusters.
Given a set of n objects S = {O1, ..., On} and two data
clusters of S which we want to compare: X = {x1, ..., xR} and
Y = {y1, ..., yS} where the different partitions of X and Y are
disjoint and their union is equal to S; we can compute the
following values:
a is the number of elements in S that are in the same
partition in X and in the same partition in Y,
b is the number of elements in S that are not in the same
partition in X and not in the same partition in Y,
c is the number of elements in S that are in the same
partition in X and not in the same partition in Y,
d is the number of elements in S that are not in the same
partition in X but are in the same partition in Y.
Intuitively, one can think of a + b as the number of
agreements between X and Y and c + d the number of
disagreements between X and Y. The Rand index, R, then
becomes,

International Journal of Computer Applications (0975 – 888)
Volume 47– No.14, June 2012
22
Fig 2: The ANN based CTG Data Classifier
dcba
da
RI 


The Rand index has a value between 0 and 1 with 0 indicating
that the two set of data clusters do not agree on any pair of
points and 1 indicating that the two data clusters are exactly
similar.
Precision
Precision is calculated as the fraction of correct objects among
those that the algorithm believes belonging to the relevant
class. It can be loosely equated to accuracy and it will roughly
answers the question: “How many of the points in this cluster
belong there/ correctly classified?”
The Precision is calculated as :
P(Lr, Si) = nri/ni
for
class Lr of size nr
cluster Si if size ni
nri data points in Si from class Lr
Recall
Recall roughly answers the question: "Did all of the
documents that belong in this cluster make it in?". In other
words, recall is the fraction of actual objects that were
identified.
The recall is calculated as :
R(Lr, Si) = nri/nr
F-Score
F-Score is the harmonic mean of Precision and Recall and will
tries to give a good combination of the two. It is calculated
with the equation:
),(),(
),(),(2
),(
irir
irir
ir SLPSLR
SLPSLR
SLF 


4. RESULTS AND DISCUSSION
4.1 Data Set Information
For evaluating the algorithms under consideration, we used
cardiotocograms data from UCI Machine Learning
Repository.
This data set contains 2126 fetal cardiotocograms belonging
to different classes. The data contains 21 attributes and two
class labels. The CTGs were classified by three expert
obstetricians and a consensus classification label assigned to
each of them. Classification was both with respect to a
morphologic pattern (A, B, C. ...) and to a fetal state (N, S, P).
Therefore the dataset can be used either for 10-class or 3-class
experiments. Here we use this data set for these evaluations.
Attribute Information
1) LB - FHR baseline (beats per minute)
2) AC - # of accelerations per second
3) FM - # of fetal movements per second
4) UC - # of uterine contractions per second
5) DL - # of light decelerations per second
6) DS - # of severe decelerations per second
7) DP - # of prolongued decelerations per second
8) ASTV - percentage of time with abnormal short
term variability
9) MSTV - mean value of short term variability
10) ALTV - percentage of time with abnormal long
term variability
11) MLTV - mean value of long term variability
12) Width - width of FHR histogram
13) Min - minimum of FHR histogram
14) Max - Maximum of FHR histogram
15) Nmax - # of histogram peaks
16) Nzeros - # of histogram zeros
17) Mode - histogram mode
18) Mean - histogram mean
19) Median - histogram median
20) Variance - histogram variance
21) Tendency - histogram tendency
The CTG Data
The Training
Data with Class
Labels
The Testing
Data
Normalize the
Training Data
Normalize the
Testing Data
Train ANN using Training
Data and Class Labels
Classify the Test Data
using the Trained
Network
New Class
Labels
Class Labels of
the Test Data
Measure Performance Using
Rand Index, Precision, Recall and
F-Score

International Journal of Computer Applications (0975 – 888)
Volume 47– No.14, June 2012
23
22) CLASS - FHR pattern class code (1 to 10)
23) NSP - fetal state class code (Normal=1; Suspect=2;
Pathologic=3)
Class Information
We used the data for a three class classification problem. The
descriptions for the three classes are
Normal
A CTG where all four features fall into the reassuring
category
Suspicious
A CTG whose features fall into one of the non-
reassuring categories and the reassuring category and the
remainder of features are reassuring
Pathological
A CTG whose features fall into two or more of the Non-
reassuring the reassuring category or two or more abnormal
categories.
4.2 The Visualization of Data Space
The following image shows the projection of this 21 attribute
(dimension) data in to a virtual three dimensional data space.
We used three principal components of the data for this
projection. In this plot, the normal CTG data points are shown
in black dots, the suspicious data points are shown as blue
dots, and the Pathologic data points are shown as red ‘x’
mark. This figure roughly shows the distribution of the data in
the virtual space.
Fig 1 : The 3D projection of CTG data
The Numerical Results
The following tables show the average performance of the
three different methods. Here we tabulate the average results
of ten trials. (The detailed results of all the trials can be found
in the tables presented in annexure section)
Table 1.The Performance in terms of
Rand Index and CPU time
Table 2. The Average Performance of ANN Based
Classifier
Metric
Normal
Suspicious
Pathological
Precision
0.9663
0.5897
0.9706
Recall
0.991
0.3688
0.9745
F-Score
0.9784
0.4514
0.9724
The Analysis of Results
The performance of the algorithms in terms of Rand Index
was good and always greater than 0.9. The proposed model
consumed around 2.5 seconds for training and testing. 2.5
seconds is not a big figure to consider and will not be a
obstacle in practical use of the method in real world
application.
0
0.2
0.4
0.6
0.8
1
1.2
Precision
Recall
F-Score
Performance Index
Metric
Analysis of Performance of ANN
Normal
Suspicious
Pathological
Fig 4. Performance of ANN
SlNo
RI
Time
01
0.9146
3.6719
02
0.9428
2.4844
03
0.9317
2.3750
04
0.9266
2.5469
05
0.9396
2.3750
06
0.9481
2.4688
07
0.9395
2.3750
08
0.9325
2.4688
09
0.9348
2.4531
10
0.9178
2.6719
Avg
0.9328
2.5891

International Journal of Computer Applications (0975 – 888)
Volume 47– No.14, June 2012
24
The above chart (Fig. 4) obviously shows the good
performance of ANN based classifier. It gives good precision,
recall and f-score for normal as well as pathological
records. But giving poor performance in the case of suspicious
records.
Arrived results obviously show that supervised machine
learning based methods can be used for the classification of
CTG data. We realize that there are some training glitches in
the case of suspicious records which caused some unexpected
poor results while classifying the CTG data class
“suspicious”.
5. CONCLUSION
The performance neural network based classification model
has been analyzed with CTG dataset.. According to the
arrived results, the performance of the supervised machine
learning based classification approach provided significant
performance. It was found that, the ANN based classifier was
capable of identifying Normal, Suspicious and Pathologic
condition, from the nature of CTG data with very good
accuracy. ANN based classifier provided excellent
performance in terms of Rand Index, Precision, Recall and F-
Score. It was capable of identifying Normal and Pathologic
condition with almost equal accuracy. But if we carefully see
the comparative chart of ANN (the last figure), we can tell
that, it’s performance to identify the Suspicious CTG pattern
is little bit poor than the other two classes. So future works
my address the way to improve the system to recognize the
Suspicious CTG patterns with the same accuracy.
Even though the ANN based classifier provided excellent
average performance, if we carefully watch the results of ten
trials with ANN (the last table in annexure), we may find
another weakness of this system. If we see some cells of the
columns P2, R2 and F2 there are some bad results
(highlighted in gray colour) during some trials. It means, in
that trial, the system was absolutely incapable of identifying a
single suspicious record. It means, even though we train the
system with all the classes of samples, there is a chance by
which the trained system may be incapable of identifying
suspicious record. That is why we are getting comparatively
poor average performance while classifying suspicious
records. It is a major weakness of the system which should be
overcome in future design. One may address the way to
improve the system for getting proper training with different
classes of CTG patterns. Future works may address hybrid
models using statistical and machine learning techniques for
improved classification accuracy.
ANNEXURE - 1
In the following tables, P1 is the precision for normal record, P2 is the precision for suspicious record, P3 is the precision for
pathological records. R1 is the recall for normal record, R2 is the recall for suspicious record, R3 is the recall for pathological records.
F1 is the f-score for normal record, F2 is the f-score for suspicious records, F3 is the f-score for pathological records.
Table 3. Results with ANN (10 Trials)
SlNo
Precision
Recall
F-score
P1
P2
P3
R1
R2
R3
F1
F2
F3
01
0.9485
0.0000
0.9787
0.9978
0.0000
0.9787
0.9725
0.0000
0.9787
02
0.9744
0.7714
0.9785
0.9892
0.5625
0.9681
0.9817
0.6506
0.9733
03
0.9652
0.7037
1.0000
0.9913
0.3958
0.9574
0.9781
0.5067
0.9783
04
0.9650
0.6522
0.9485
0.9881
0.3125
0.9787
0.9764
0.4225
0.9634
05
0.9733
0.7429
0.9785
0.9881
0.5417
0.9681
0.9806
0.6265
0.9733
06
0.9785
0.7500
0.9691
0.9881
0.5625
1.0000
0.9833
0.6429
0.9843
07
0.9713
0.7857
0.9583
0.9902
0.4583
0.9787
0.9807
0.5789
0.9684
08
0.9682
0.7407
0.9474
0.9892
0.4167
0.9574
0.9785
0.5333
0.9524
09
0.9682
0.7500
0.9785
0.9902
0.4375
0.9681
0.9791
0.5526
0.9733
10
0.9504
0.0000
0.9688
0.9978
0.0000
0.9894
0.9735
0.0000
0.9789
Avg
0.9663
0.5897
0.9706
0.9910
0.3688
0.9745
0.9784
0.4514
0.9724

International Journal of Computer Applications (0975 – 888)
Volume 47– No.14, June 2012
25
6. REFERENCES
[1] Xiaojun Chen, Yunming Ye, Xiaofei Xu, Joshua Zhexue
Huang , “A feature group weighting method for
subspace clustering of high-dimensional data”, Pattern
Recognition 45 (2012) 434-446, Elsevier
[2] Shahad Nidhal, M. A. Mohd. Ali1 and Hind Najah, “A
novel cardiotocography fetal heart rate baseline
estimation algorithm”, Scientific Research and Essays
Vol. 5(24), pp. 4002-4010, 18 December, 2010
[3] ANA. KLIMEŠOVÁ, EVA OCELÍKOVÁ,
Multidimensional Data Classification, Proceedings of the
10th WSEAS International Conference on
AUTOMATION & INFORMATION, ISSN: 1790-5117,
ISBN: 978-960-474-064-2
[4] Stirrat, Mills and Draycott, "Notes on Obstetrics and
Gynaecology for the MRCOG, 5th Edition", 04 Aug
2003, ISBN: 9780443072239
[5] Diogo Ayres-de-Camposa, Cristina Costa-Santosb, Joa˜o
Bernardesa, "Prediction of neonatal state by computer
analysis of fetal heart rate tracings: the antepartum arm
of the SisPorto1 multicentre validation study”, European
Journal of Obstetrics & Gynecology and Reproductive
Biology 118 (2005) 52-60.
[6] http://www.academicjournals.org/SRE, ISSN 1992 –
2248 © 2010 Academic Journals.
[7] Antonia Costa, MD; Diogo Ayres-de-Campos, PhD;
Fernada Costa, MD; Cristina Santos, MS; Joao
Bernardes, PhD, “Prediction of neonatal academia by
Computer analysis of fetal heart rate and ST event
sibnals” 2009 AJOG – American Journal of Obstetrics
and Gynecology.
[8] Ben Kao, Sau Dan Lee, Foris K.F.Lee, David W.
Cheung, Wai-Shing Ho,” Clustering Uncertain Data
using Voronoi Diagrams and R-Tree Index” IEEE
Transactions on Knowledge and Data Engineering, Vol.
22(9), pp. 1219 – 1233, sep 2010
[9] E. Ocelikova, D. Klimesova, “ Bays Classifier in
multidimensional data classification “ 15th Int.
Conference Process Control 2005, pp. 188-1 – 188-5.
Strbske Pleso, Slovakia.
[10] E. Ocelikovć, J Krištof, “Classification of multispectral
data” Zbornik radova, Volume 25, Number 1(2001).
[11] http://www-h.eng.cam.ac.uk/help/tpl/programs/
matlab.html.
[12] S.Anto, Dr. S.Chandramathi, “Supervised Machine
Learning Approaches for Medical Data Set
Classification – A Review” IJCST Nol. No.2, Issue 4, pp.
234 – 240, Oct – Dec 2011, ISSN : 2229-4333.
[13] Frank, A. Asuncion, UCI Machine Learning Repository
{http://archive.ics.uci.edu/ml}, 2010.
[14] Zhaohong Deng , Kup-Sze Choi , Fu-Lai Chung ,
Shitong Wang, Enhanced soft subspace clustering
integrating within-cluster and between-cluster
information, Pattern Recognition, v.43 n.3, p.767-781,
March, 2010 [doi>10.1016/j.patcog.2009.09.010]
[15] Hans-Peter Kriegel , Peer Kröger , Arthur Zimek,
Clustering high-dimensional data: A survey on subspace
clustering, pattern-based clustering, and correlation
clustering, ACM Transactions on Knowledge Discovery
from Data (TKDD), v.3 n.1, p.1-58, March 2009
[doi>10.1145/1497577.1497578]
[16] http://www.mathworks.in/help/toolbox/nnet/ug/bss33y1-
1.html.
[17] S.Angle Latha Mary, K.R.Shankar Kumar,” Evaluation
of Clustering Algorithm with Cluster Validation Metrics”
European Journal of Scientific Research ISSN 1450-
216X Vol.69 No.1 (2012), pp.61-72
[18] https://sites.google.com/site/dataclusteringalgorithms/fuz
zy-c-means-clustering-algorithm.
[19] http://home.dei.polimi.it/matteucc/Clustering/tutorial_ht
ml/cmeans.html.
[20] YI PENG, GANG KOU“A descriptive framework for
the field of data mining and knowledge discovery”
International Journal of Information Technology &
Decision Making Vol. 7, No. 4 (2008) pp. 639–682.
[21] Michael Lloyd-Williams, “Discovering the hidden
secrets in your data - the data mining approach to
information”, Information Research,
{http://informationr.net/ir/3-2/paper36.html},Vol. 3 No.
2, September 1997.