International Journal of Scientific Engineering and Applied Science (IJSEAS) - Volume-1, Issue-5, August 2015

ISSN: 2395-3470
www.ijseas.com

Mining Big Data: Breast Cancer Prediction

using DT - SVM Hybrid Model
K.Sivakami, Assistant Professor, Department of Computer Application

Nadar Saraswathi College of Arts & Science, Theni.

Abstract - Breast Cancer is becoming a leading data. It encompasses a set of processes

cause of death among women in the whole performed automatically, whose task is to
world; meanwhile, it is confirmed that the early discover and extract hidden features (such
detection and accurate diagnosis of this disease as: various patterns, regularities and
can ensure a long survival of the patients. This
anomalies) from large datasets.
paper work presents a disease status prediction
employing a hybrid methodology to forecast the Big data is a broad term for datasets so large
changes and its consequence that is crucial for or complex that traditional data processing
lethal infections. To alarm the severity of the
applications are inadequate. Analysis of data
diseases, our strategy consists of two main parts:
sets can find new correlations, to "spot
1. Information Treatment and Option Extraction,
business trends, prevent diseases, and
and 2. Decision Tree-Support Vector Machine
(DT-SVM) Hybrid Model for predictions. We combat crime and so on. Scientists,
analyse the breast Cancer data available from the practitioners of media and advertising
Wisconsin dataset from UCI machine learning and governments alike regularly meet
with the aim of developing accurate prediction difficulties with large data sets in areas
models for breast cancer using data mining including Internet search, finance and
techniques. In this experiment, we compare business informatics [1].
three classifications techniques in Weka
software and comparison results show that DT- Building accurate and efficient classifiers
SVM has higher prediction accuracy than for large databases is one of the essential
Instance-based learning (IBL), Sequential tasks of data mining and machine learning
Minimal Optimization (SMO) and Nave based research [10]. Usually, classification is a
classifiers. preliminary data analysis step for examining
Index Terms - breast cancer; classification; a set of cases to see if they can be grouped
Decision Tree- Support Vector Machine, based on similarity" to each other. The
Nave Bayes, Instance-based learning, ultimate reason for doing classification is to
Sequential Minimal Optimization, and increase understanding of the domain or to
weka; improve predictions compared to
unclassified data. Many different types of
1 INTRODUCTION classification techniques have been proposed
in literature that includes DT-SVM, SMO,
Data mining, also known as knowledge IBL, etc.
discovery in databases is defined as the
extraction of implicit, previously unknown, Clustering is a Data mining technique which
and potentially useful information from segments a heterogeneous data into a

418
 International Journal of Scientific Engineering and Applied Science (IJSEAS) - Volume-1, Issue-5, August 2015
ISSN: 2395-3470
www.ijseas.com

number of homogeneous subgroups or medical problem. Thus, finding an accurate

clusters. In clustering, the records are and effective diagnosis method is very
grouped together based on self-similarity important. In recent years machine learning
without any predefined classes or examples. methods have been widely used in
The quality of a cluster is measured by prediction, especially in medical diagnosis
cluster diameter which is the distance [2]. Medical diagnosis is one of major
between any two objects in a cluster. In this problem in medical application. The
study k-means clustering is chosen because classification of Breast Cancer data can be
of its popularity and its proved useful to predict the outcome of some
effectiveness. diseases or discover the genetic behavior of
tumors [3]. A major class of problems in
In data mining breast cancer research has medical science involves the diagnosis of
been one of the important research topics in disease, based upon various tests performed
medical science during the recent years The upon the patient. For this reason the use of
classification of Breast Cancer data can be classifier systems in medical diagnosis is
useful to predict the result of some diseases gradually increasing.
or discover the genetic behavior of tumors.
There are many techniques to predict and The aim of this dissertation is to develop the
classification breast cancer pattern. This various phase. They are:
work empirically compares performance of
different classification rules that are suitable Predictive Framework for Breast
for direct interpretability of their results. Cancer Disease
Hybrid model
Breast cancer is one of the most common Generation of rule to predict Breast
cancers among women. Breast cancer is one Cancer
of the major causes of death in women when Qualitative measures used for
compared to all other cancers. Cancer is a Comparison (accuracy, error rate,
type of diseases that causes the cells of the sensitivity, and specificity).
body to change its characteristics and cause GUI Development
abnormal growth of cells. Most types of
cancer cells eventually become a mass 1.2 OVERVIEW
called tumor. The occurrence of breast
cancer is increasing globally. It is a major Breast cancer is becoming a leading cause of
health problem and represents a significant death among women in the whole world,
worry for many women [1]. meanwhile, it is confirmed that the early
detection and accurate diagnosis of this
Early detection of breast cancer is essential disease can ensure a long survival of the
in reducing life losses. However earlier patients. In this research work, a decision
treatment requires the ability to detect breast intelligence technique based support vector
cancer in early stages. Early diagnosis machine classifier (DT-SVM) is proposed
requires an accurate and reliable diagnosis for breast cancer diagnosis. In the proposed
procedure that allows physicians to DT-SVM, the issue of model selection and
distinguish benign breast tumors from feature selection in SVM is simultaneously
malignant ones. The automatic diagnosis of solved using Net beans and WEKA
breast cancer is an important, real-world

419
 International Journal of Scientific Engineering and Applied Science (IJSEAS) - Volume-1, Issue-5, August 2015
ISSN: 2395-3470
www.ijseas.com

analytical tool. A weighted function is Djebbari et al. consider the effect of

adopted to design the objective function of ensemble of machine learning techniques to
DT-SVM, which takes into account the predict the survival time in breast cancer.
average accuracy rates of SVM, the number Their technique shows better accuracy on
of support vectors and the selected features their breast cancer data set comparing to
simultaneously. previous results. Liu Ya-Qins [5]
experimented on breast cancer data using C5
2 RELATED WORKS algorithm with bagging to predict breast
Delen et al in their work preprocessed the cancer survivability.
SEER data for to remove redundancies and Bellaachi et al. used naive bayes, decision
missing information. They have compared tree and back-propagation neural network to
the predictive accuracy of the SEER data on predict the survivability in breast cancer
three prediction models indicated that the patients. Although they reached good results
decision tree (C5) is the best predictor with (about 90% accuracy), their results were not
93.6% accuracy on the holdout sample. significant due to the fact that they divided
Endo et al. implemented common machine the data set to two groups; one for the
learning algorithms to predict survival rate patients who survived more than 5 years and
of breast cancer patient. This study is based the other for those patients who died before
upon data of the SEER program with high 5 years.
rate of positive examples (18.5 %). Logistic Tsirogianniss et al. applied bagging
regression had the highest accuracy; algorithm on medical databases using the
artificial neural network showed the highest classifiers neural networks, SVMS and
specificity and J48 decision trees model had decision trees. Results exhibits improved
the best sensitivity accuracy of bagging than without bagging.
Kotsiantis et al. did a work on Bagging, 3 RELATED METHODS
Boosting and Combination of Bagging and
Boosting as a single ensemble using The overall design procedure is shown is shown
different base learners such as C4.5, Nave in Figure 3.1.
Bayes, OneR and Decision Stump. These
were experimented on several benchmark
datasets of UCI Machine Learning
Repository.

Bittern et al. used artificial neural network to

predict the survivability for breast cancer
patients. They tested their approach on a
limited data set, but their results show a Figure.3.1: Ensemble Model for Breast cancer
good agreement with actual survival. identification

Vikas Chaurasia et al. used RepTree, RBF The design is the structure of any scientific
Network and Simple Logistic to predict the work. It gives direction and systematizes
survivability for breast cancer patients. the research. Most scientists are interested in
getting reliable observations that can help the

420
 International Journal of Scientific Engineering and Applied Science (IJSEAS) - Volume-1, Issue-5, August 2015
ISSN: 2395-3470
www.ijseas.com

understanding of a phenomenon. DT SVM

method has been adopted in this research START
was so carefully designed.

3.1 EXPERIMENTAL DESIGN

Classification predicts categorical class TRAINING TEST

labels (discrete or nominal) and also A AS

classifies data (constructs a model) based on
DATASET
the training set and the values (class labels)
USER ANALYSIS
in a classifying attribute and uses it in
classifying new data. The figure 3.2 shoes
classification of new data. PARAMETER PARTITION
ANALYSIS

DECISION
SUPPORT &
CONFIDENCE

SUPPORT &
CONFIDENCE

RESULT

TRANSFORMATION

Figure 3.2: Data classification

3.1.1 SUPPORT VECTOR MACHINE Figure 3.3: SVM Data Flow

Classification in SVM is an example of 3.1.2 Decision Tree

Supervised Learning. Known labels help
indicate whether the system is performing in DT is a most popular and powerful
a right way or not. This information points classification technique where each internal
to a desired response, validating the node (non-leaf node) denotes a test on an
accuracy of the system, or be used to help attribute, each branch represents an outcome
the system learn to act correctly. A step in of the test, and each leaf node (or terminal
SVM classification involves identification as node) holds a class label. DT with four
which are intimately connected to the known decision nodes and five leaf nodes are
classes is called feature selection. Feature shown in figure 3.5. The top most nodes in a
selection and SVM classification together tree are the root nodes. DT is so popular
have been used even, when prediction of because construction of DT classifiers does
unknown samples is not necessary. They can not require any domain knowledge or
be used to identify key sets which are parameter setting and, therefore, is
involved in whatever processes distinguish appropriate for exploratory knowledge
the classes. The flow diagram for the SVM discovery. Various decision tree based
method is shown is given in figure 3.3. techniques are widely accepted and applied
on health care diagnosis process. On the
other hand some statistical technique like

421
 International Journal of Scientific Engineering and Applied Science (IJSEAS) - Volume-1, Issue-5, August 2015
ISSN: 2395-3470
www.ijseas.com

SVM is also used as classifier for health care Information Gain: The information gain is
diagnosis. Interactive Dichotomiser 3 (ID3) based on the decrease in entropy after a
and C4.5 are the two very popular DT dataset is split on an attribute. Constructing
algorithms proposed by Quinlan [20]. ID3 a decision tree is all about finding attribute
uses Entropy and Information Gain to that returns the highest information gain
construct a decision tree. (i.e., the most homogeneous branches).

3.1.3 INSTANCE BASED LEARNING

IBL is a basic instance-based learner which

finds the training instance closest in
Euclidean distance to the given test instance
and predicts the same class as this training
distance. If several instances qualify as the
closest, the first one found is used. IBL
algorithms do not construct extensional
CDs. These functions are two of the three
Figure 3.4: Decision Tree with 4 decision nodes
components in the following framework that
and 5 leaf nodes
describes all IBL algorithms:
Entropy: A decision tree is built top-down
Similarity Function: This calculates
from a root node and involves partitioning
the similarity between training
the data into subsets that contain instances
instances i and the instances in the
with similar values (homogenous). ID3
concept depiction. Similarities are
algorithm uses entropy to calculate the
numeric-valued.
homogeneity of a sample. If the sample is
completely homogeneous the entropy is zero Classification Function: This obtains
and if the sample is an equally divided it has the similarity functions results and
entropy of one. the classification performance
records of the instances in the
To build a decision tree, we need to concept description. It returns a
calculate two types of entropy using classification for i.
frequency tables as follows: CD Updater: This retains records on
classification performance and
a) Entropy using the frequency table of
decides which instances to include in
one attribute:
the concept description. Inputs
include i, the classification results,
the similarity results, and a current
concept description. It returns the
modified concept description.
b) Entropy using the frequency table of
two attributes:
3.1.4SEQUENTIAL MINIMAL
, OPTIMIZATION

422
 International Journal of Scientific Engineering and Applied Science (IJSEAS) - Volume-1, Issue-5, August 2015
ISSN: 2395-3470
www.ijseas.com

SMO is an algorithm for solving the 9. Normal Nucleoli 1-10

10. Mitoses 1-10
quadratic programming (QP) problem that
2-
arises during the training of SVM. SMO is Benign,
11. Class
widely used for training support vector 4 for
malignant
machines and is implemented by the popular
LIBSVM tool.
Table 3.1 Attribute Information: (class
SMO is an iterative algorithm for solving
attribute has been moved to last column)
the optimization problem described above.
SMO breaks this problem into a series of 3. Missing attribute values: 16
smallest possible sub-problems, which are
then solved analytically. There are 16 instances in Groups 1
to 6 that contain a single missing
3.2 DATA SOURCES (i.e., unavailable) attribute value,
now denoted by "?".
In this study, we have performed our
conduction on the Wisconsin Breast Cancer 4. Class distribution:
Dataset (WBCD) taken from UCI machine
learning repository (UCI Repository of Benign: 458 (65.5%)
Machine Learning Databases). The dataset Malignant: 241 (34.5%)
contains 699 instances taken from needle
aspirates from patients breasts, of which
458 cases belong to benign class and the
remaining 241 cases belong to malignant
class. It should be noted that there are16
instances which have missing values, in this
study all the missing values are replaced by
the mean of the attributes. Each record in the
database has nine attributes. These nine
attributes were found to differ significantly
between benign and malignant samples. Figure 3.5: Diagrammatic Representation of
Breast Cancer Attributes
3.3 DATA ANALYSIS
4 RESULTS & DISCUSSION
1. Number of Instances: 699
The proposed DT-SVM diagnostic system is
2. Number of Attributes: 10 + class attribute
implemented using Weka tool with Java
platform. For SVM, LIBSVM
S.NO ATTRIBUTE DOMAIN
id
implementation was utilized. The empirical
1. Sample code number experiment was conducted on Intel Quad-
number
2. Clump Thickness 1-10 Core Xeon 5130 CPU (2.0 GHz) with 4GB
3. Uniformity of Cell Size 1-10
4. Uniformity of Cell Shape 1-10
of RAM.
5. Marginal Adhesion 1-10
6. Single Epithelial Cell Size 1-10 In order to guarantee the valid results, the k-
7. Bare Nuclei 1-10 fold Crossover Validation (CV) was used to
8. Bland Chromatin 1-10 evaluate the classification accuracy [45].

423
 International Journal of Scientific Engineering and Applied Science (IJSEAS) - Volume-1, Issue-5, August 2015
ISSN: 2395-3470
www.ijseas.com

Data was divided into ten subsets. Each

time, one of the ten subsets is used as the
test set and the other nine subsets are put
together to form a training set. Then the
average error across all ten trials is
computed. The advantage of DT-SVM
method is that all of the test sets are
independent and the reliability of the results
could be improved. We attempted to design
our experiment using two loops. The inner
loop is used to determine the optimal
parameters and best feature subset. The
outer loop is used for estimating the
performance of the SVM classifier.
Figure 4.1 Interface to feed individual patient
We selected Breast cancer data set and
data
selected different class of patient groups
from the collection for our experiments. The
To forecast the severity of breast cancer
group data is pre-processed and made as
with help of slider by feeding the field
simple text files thereby removing all the
values are given in figure 4.2. The
headers, replies and other non- related
attribute values are also displayed in
information. The preprocessed data from
chart.
these selected classes are divided as 60%
training data and 40% testing data. But we
selected only 10 parameters for training and
around 400 patients documents for testing.
The feature selection was simple and did not
involve any special techniques, to test the
classifier with such features. In our
implementation of the classifier we used
below interface to train the SVM. The
classes which we selected for our
experiments are different age groups, breast
size and menopause. The classifier was
trained with these classes and then tested
with the test data.
Figure 4.2: Predictive framework to forecast
The figure 4.1 is act as an interface to feed the severity of cancer.
individual patients data to predict the
response which is diagnosed against the The figure 4.3 is used to evaluate the dataset
classified ones using DT-SVM. Show based on the class distribution benign and
method used to view all available data set. malignant with predefined decision rule by
calculating the support & confidence values.

424
 International Journal of Scientific Engineering and Applied Science (IJSEAS) - Volume-1, Issue-5, August 2015
ISSN: 2395-3470
www.ijseas.com

Figure 4.5: Calculation of Minimum Support

Threshold values
Figure 4.3: Rule to evaluate dataset for class
malignant The truncated database will be imported to a
DT (ID3) algorithm to find the predictive
To find the rule to predict breast cancer
rule for breast cancer. In figure 4.6 present
using DT, the transaction database will be
the complete DT with 25 decision nodes
imported are given in figure 4.4. The
with three leaf nodes with class of
number of transactions will be calculated
malignant.
with respect to the attribute values. It also
shows the Class distribution: Benign,
Malignant.

In order to find the rule based on

transactional database. We must truncate the
discarded data by calculating the minimum
support threshold through interesting
measures such as support and confidence
values are shown in figure 4.5.

Figure 4.6: Generation of Final Decision

Tree

We have designed a Prototype of

Breast cancer disease prediction model. This
model will predict the breast cancers
disease class based on the rules created by
ID3 algorithms.Figure.4.1 shows the
interface for input, which takes Medical
hFigure 4.4: Imported transaction database profiles of a patient such as age, sex, blood
to find the rule pressure and blood sugar etc as input and it

425
 International Journal of Scientific Engineering and Applied Science (IJSEAS) - Volume-1, Issue-5, August 2015
ISSN: 2395-3470
www.ijseas.com

100%
can predict about presence or absence of
90%
Breast cancer disease. The path from root to 80%
leaf node shows if the patient has these 70%

Accuracy in %
combinations of cell values is high, and then 60%

the patient is affected by cancer. This is the 50%

40%
rule we generate through DT.
30%

It has been observed that the classes 20%

10%
which are predicted at the top level of the
0%
tree have good accuracy and the ones DT+SVM IBK SMO NAVE
ACCURACY 91% 85.23% 72.56% 89.48%
predicted at the lower levels have poor
accuracy. In our experiment which was
classified at the root node has 95% accuracy. Figure 4.11: Accuracy of Classification
Methods
INCORRECTLY
DESCRIPTION

ERROR RATE

CORRECTLY
CLASSIFIED

CLASSIFIED
ACCURACY

INSTANCE

INSTANCE
14
12

Error Rate In %
10
DT+SVM 91% 2.58 459 240 8
6
IBL 85.23% 12.63 184 515
4
SMO 72.56% 5.96 325 374 2
0
NAVE 89.48% 9.89 291 408 DT+SVM IBK SMO NAVE
ERRORRATE 2.58 12.63 5.96 9.89
The performance of a chosen (IBL, SMO
and Nave based) classifiers are performed
through weka tool and its validated based
on error rate and accuracy. The Figure 4.12: Error Rate of Classification
classification accuracy is predicted in terms Methods
of Sensitivity and Specificity. The 500
Number of Instance

evaluation parameters are the specificity,

sensitivity, and overall accuracy. Hence 400
DT+SVM perform well in classifying the 300
breast cancer data, compared to all other
algorithms. 200

100
From the above figures and table we find
that highest accuracy of Classification 0 DT+SVM IBK SMO NAVE
model is DT - SVM (91%), low error rate Correctly
Classified 459 184 325 291
(2.58%), correctly classified instance (459) Instance
and incorrectly classified instance (240) in
breast cancer data as shown in Figure 4.11, Figure 4.13: Correctly Classified Instance
4.12, 4.13 and 4.14. of Classification Methods

426
 International Journal of Scientific Engineering and Applied Science (IJSEAS) - Volume-1, Issue-5, August 2015
ISSN: 2395-3470
www.ijseas.com

600
The experimental results given in the Figure

Number of Instance
500 4.10 and Table 4.1 showed the effectiveness
400 of the proposed algorithm. Overall, DT-
SVM classification accuracy is better than
300
other classifier algorithm. However, from a
200
relatively low error rate, the results show
100 that the DT-SVM will be the best prognosis
0
in clinical practice.
DT+SVM IBK SMO NAVE
InCorrectly
Classified 240 515 374 408 The optimum breast cancer disease
Instance
predictive model obtained in this study
adopts DT-SVM classification algorithm,
Figure 4.13: InCorrectly Classified Instance
this research may provide references for
of Classification Methods future research on selecting the optimal
predictive models to lower the incidence of
5 CONCLUSION breast cancer.

This work proposed a hybrid classification REFERENCES

algorithm for breast cancer patients which
integrates DT and SVM algorithms. The [1] American Cancer Society. Breast Cancer
proposed algorithm was composed of two Facts & Figures 2005-2006. Atlanta: American
Cancer Society, Inc. (http://www.cancer.org/).
main phases. The first phase is Information
treatment and option extraction followed by [2] G. Ravi Kumar, Dr. G. A. Ramachandra,
DT-SVM hybrid model predictions. K.Nagamani, An Efficient Prediction of Breast
Classification had two main phases, Cancer Data using Data Mining Techniques,
Training and Testing phases. The input International Journal of Innovations in
parameters for SVM were optimized using Engineering and Technology (IJIET), Vol. 2
DT algorithm. The SVM algorithm was used Issue 4 August 2013.
to classify breast cancer patients into one of
[3] Miss Jahanvi Joshi, Mr. RinalDoshi , Dr.
two classes (Benign/Malignant).
Jigar Patel, Diagnosis and Prognosis Breast
In comparison of data mining techniques, Cancer Using Classification Rules,
International Journal of Engineering Research
this research used accuracy indicator to
and General Science Volume 2, Issue 6,
evaluate classification efficiency of different
October-November, 2014,
algorithms. The proposed algorithm was
compared with different classifier [4] A.Bellachia and E.Guvan,Predicting breast
algorithms using Weka tool. Specifically, cancer survivability using data mining
we used popular data mining methods: DT- techniques, Scientific Data Mining Workshop,
SVM, SMO, IBL, and Nave. An important in conjunction with the 2006 SIAM Conference
challenge in data mining and machine on Data Mining, 2006.
learning areas is to build precise and
[5] A. Endo, T. Shibata and H. Tanaka (2008),
computationally efficient classifiers for
Comparison of seven algorithms to predict
Medical applications. breast cancer survival, Biomedical Soft
Computing and Human Sciences, vol.13, pp.11-
16.

427
 International Journal of Scientific Engineering and Applied Science (IJSEAS) - Volume-1, Issue-5, August 2015
ISSN: 2395-3470
www.ijseas.com

[6] Breast Cancer Wisconsin Data [online]. [16] Vapnik, V. N., The nature of statistical
Available: http://archive.ics.uci.edu/ml/machine- learning theory. Springer, Berlin, 1995.
learning-databases/breast-
cancerwisconsin/breast-cancer-wisconsin.data. [17] Weka: Data Mining Software in Java,
http://www.cs.waikato.ac.nz/ml/weka/
[7] Brenner, H., Long-term survival rates of
cancer patients achieved by the end of the 20th [18] Witten H.I., Frank E., Data Mining:
century: a period analysis. Lancet. 360:1131 Practical Machine Learning Tools and
1135, 2002. Techniques, Second edition, Morgan Kaufmann
Publishers, 2005.
[8] D. Delen, G. Walker and A. Kadam (2005),
Predicting breast cancer survivability: a [19] Y Rejani- Early detection of breast cancer
comparison of three data mining methods, using SVM. 2009 arxiv
Artificial Intelligence in Medicine, vol.34,
pp.113-127. [20] Ilias Maglogiannis, E Zafiropoulos An
intelligent system for automated breast cancer
[9] Ian H. Witten and Eibe Frank. Data Mining: diagnosis and prognosis using SVM based
Practical machine learning tools and techniques, classifiers Applied Intelligence, 2009
2nd Edition. San Fransisco:Morgan Kaufmann; Springer.
2005.
[21] Zhang Qinli; Wang Shitong; Guo Qi; A
[10] J. Han and M. Kamber, Data Mining Novel SVM and Its Application to Breast
Concepts and Technique (The Morgan Cancer Diagnosis
Kaufmann Series in Data Management http://ieeexplore.ieee.org/xpl/freeabs_all.jsp?arn
Systems), 2nd ed. San Mateo, CA: Morgan umber=4272649.
Kaufmann, 2006.
[22] Quinlan.J.R. (1993).C4.5:Programs for
[11] J. R. Quinlan, C4.5: Programs for Machine machine learning (1st edition), San Francisco,
Learning. San Mateo, CA:Morgan Kaufmann; Morgan Kaufmann Publishers,1993.
1993.
[23] Pendharkar PC, Rodger JA, Yaverbaum GJ,
[12] Mitchell, T. M., Machine Learning, Herman N, Benner M (1999) Association,
McGraw-Hill Science/Engineering/Math, 1997 statistical, mathematical and neural approaches
for mining breast cancer patterns. Expert
[13] P.-N. Tan, M. Steinbach, and V. Kumar, Systems with Applications 17: 223-232.
Introduction to Data Mining. Reading, MA:
Addison-Wesley, 2005. [24] Zhou ZH, Jiang Y (2003) Medical
diagnosis with C4.5 Rule preceded by artificial
[14] Razavi, A. R., Gill, H., Ahlfeldt, H., and neural network ensemble. IEEE Trans Inf
Shahsavar, N., Predicting metastasis in breast Technol Biomed 7: 37-42.
cancer: comparing a decision tree with domain
experts. J. Med. Syst. 31:263273, 2007. [25] Lundin M, Lundin J, Burke HB, Toikkanen
S, Pylkkanen L, et al. (1999) Artificial neural
[15] S.B.Kotsiantis and networks applied to survival prediction in breast
P.E.Pintelas,Combining Bagging and cancer. Oncology 57: 281-286.
Boosting, International Journal of Information
[26] Shweta Kharya, Using data mining
and Mathematical Sciences, 1:4 2005.
techniques for diagnosis and prognosis of
Cancer Disease, International Journal of
Computer Science, Engineering and Information

428
 International Journal of Scientific Engineering and Applied Science (IJSEAS) - Volume-1, Issue-5, August 2015
ISSN: 2395-3470
www.ijseas.com

Technology (IJCSEIT), Vol.2, No.2, April 2012,

pp. 55-66.

[27] P.Ramachandran, N.Girija and

T.Bhuvaneswari, Health care Service Sector:
Classifying and finding Cancer spread pattern in
Southern india using data mining techniques,
International Journal on Computer Science and
Engineering (IJCSE), Vol. 4 No. 05 May 2012,
pp. 682-687.

[28] S.B.Kotsiantis and

P.E.Pintelas,Combining Bagging and
Boosting, International Journal of Information
and Mathematical Sciences, 1:4 2005.

[29] K. P. Bennett and O. L. Mangasarian:

"Robust Linear Programming Discrimination of
Two Linearly Inseparable Sets", Optimization
Methods and Software 1, 1992, 23-34.

429