KNOWLEDGE BASED ANALYSIS OF VARIOUS STATISTICAL TOOLS IN DETECTING BREAST CANCER

D.C. Wyld, et al. (Eds): CCSEA 2011, CS & IT 02, pp. 37–45, 2011.
© CS & IT-CSCP 2011 DOI: 10.5121/csit.2011.1205
KNOWLEDGE BASED ANALYSIS OF VARIOUS
STATISTICAL TOOLS IN DETECTING BREAST
CANCER
S. Aruna1
, Dr S.P. Rajagopalan2
and L.V. Nandakishore3
1,2
Department of Computer Applications, Dr M.G.R University, Chennai-95, India
arunalellapalli@yahoo.com1
, sasirekharaj@yahoo.co.in2
3
Department of Mathematics, Dr M.G.R University, Chennai-95, India
lellapalliarunakishore@gmail.com
ABSTRACT
In this paper, we study the performance criterion of machine learning tools in classifying breast
cancer. We compare the data mining tools such as Naïve Bayes, Support vector machines,
Radial basis neural networks, Decision trees J48 and simple CART. We used both binary and
multi class data sets namely WBC, WDBC and Breast tissue from UCI machine learning
depositary. The experiments are conducted in WEKA. The aim of this research is to find out the
best classifier with respect to accuracy, precision, sensitivity and specificity in detecting breast
cancer.
KEYWORDS
J48, Naïve Bayes, RBF neural networks, Simple Cart, Support vector machines.
1. INTRODUCTION
Data mining is a collection of techniques for efficient automated discovery of previously
unknown, valid, novel, useful and understandable patterns in large databases [1]. The term Data
Mining or Knowledge Discovery in databases has been adopted for a field of research dealing
with the automatic discovery of implicit information or knowledge within databases [2]. Machine
learning refers to a system that has the capability to automatically learn knowledge from
experience and other ways [3]. Classification and prediction are two forms of data analysis that
can be used to extract models describing important data classes or to predict future data trends [4]
In this paper we analyze the performance of supervised learning algorithms such as Naïve Bayes,
SVM Gaussian RBF kernel, RBF neural networks, Decision trees.J48and simple CART. These
algorithms are used for classifying the breast cancer datasets WBC, WDBC, Breast tissue from
UCI Machine learning depository (http://archive.ics.uci.edu/ml).We conducted our experiments
using WEKA tool. These algorithms have been used by many researchers and found efficient in
some aspects. The goal of this research is to find the best classifier which outperforms other
classifiers in all the aspects.
This paper is organized as follows. Section 2 gives a brief description about the data mining
algorithms and section 3 gives the description about the datasets used for this experiment. Section
4 gives the results obtained and the concluding remarks are given in section 5 to address further
research issues.

38 Computer Science & Information Technology (CS & IT)
2. DATA MINING ALGORITHMS
2.1. Naive Bayes
Naive Bayes classifier is a probabilistic classifier based on the Bayes theorem, considering a
strong (Naive) independence assumption. Thus, a Naive Bayes classifier considers that all
attributes (features) independently contribute to the probability of a certain decision. Taking into
account the nature of the underlying probability model, the Naive Bayes classifier can be trained
very efficiently in a supervised learning setting, working much better in many complex real-world
situations, especially in the computer-aided diagnosis than one might expect [5], [6]. Because
independent variables are assumed, only the variances of the variables for each class need to be
determined and not the entire covariance matrix.
(1)
where P is the probability, C is the class variable and F1.......Fn are Feature variables F1 through Fn
The denominator is independent of C.
2.2. Decision trees CART and J48
Decision trees are supervised algorithms which recursively partition the data based on its
attributes, until some stopping condition is reached [4] Decision Tree Classifier (DTC) is one of
the possible approaches to multistage decision-making. The most important feature of DTCs is
their capability to break down a complex decision making process into a collection of simpler
decisions, thus providing a solution, which is often easier to interpret [7].
The classification and regression trees (CART) methodology proposed by [8] is perhaps best
known and most widely used. CART uses cross-validation or a large independent test sample of
data to select the best tree from the sequence of trees considered in the pruning process. The basic
CART building algorithm is a greedy algorithm in that it chooses the locally best discriminatory
feature at each stage in the process. This is suboptimal but a full search for a fully optimized set
of question would be computationally very expensive. The CART approach is an alternative to
the traditional methods for prediction [8] [9] [10]. In the implementation of CART, the dataset is
split into the two subgroups that are the most different with respect to the outcome. This
procedure is continued on each subgroup until some minimum subgroup size is reached.
Decision tree J48 [11] implements Quinlan’s C4.5 algorithm [12] for generating a pruned or
unpruned C4.5 tree. C4.5 is an extension of Quinlan's earlier ID3 algorithm. The decision trees
generated by J48 can be used for classification. J48 builds decision trees from a set of labelled
training data using the concept of information entropy. It uses the fact that each attribute of the
data can be used to make a decision by splitting the data into smaller subsets.
J48 examines the normalized information gain (difference in entropy) that results from choosing
an attribute for splitting the data. To make the decision, the attribute with the highest normalized
information gain is used. Then the algorithm recurs on the smaller subsets. The splitting
procedure stops if all instances in a subset belong to the same class. Then a leaf node is created in
the decision tree telling to choose that class. But it can also happen that none of the features give
any information gain. In this case J48 creates a decision node higher up in the tree using the
expected value of the class.
J48 can handle both continuous and discrete attributes, training data with missing attribute values
and attributes with differing costs. Further it provides an option for pruning trees after creation

Computer Science & Information Technology (CS & IT) 39
2.3. Radial Basis Neural Networks
Radial Basis Function (RBF networks) is the artificial neural network type for application of
supervised learning problem [13]. By using RBF networks, the training of networks is relatively
fast due to the simple structure of RBF networks. Other than that, RBF networks are also capable
of universal approximation with non-restrictive assumptions [14]. The RBF networks can be
implemented in any types of model whether linear on non-linear and in any kind of network
whether single or multilayer [13].
The design of a RBFN in its most basic form consists of three separate layers. The input layer is
the set of source nodes (sensory units). The second layer is a hidden layer of high dimension. The
output layer gives the response of the network to the activation patterns applied to the input layer.
The transformation from the input space to the hidden-unit space is nonlinear. On the other hand,
the transformation from the hidden space to the output space is linear [15]. A mathematical
justification of this can be found in the paper by Cover [16].
2.4. Support Vector Machines
Support vector machines (SVM) are a class of learning algorithms which are based on the
principle of structural risk minimization (SRM) [17] [18]. SVMs have been successfully applied
to a number of real world problems, such as handwritten character and digit recognition, face
recognition, text categorization and object detection in machine vision [19],[20],[21]. SVMs find
applications in data mining, bioinformatics, computer vision, and pattern recognition. SVM has a
number of advanced properties, including the ability to handle large feature space, effective
avoidance of over fitting, and information condensing for the given data set.etc.[22]
Each kind of classifier needs a metric to measure the similarity or distance between patterns.
SVM classifier uses inner product as metric. If there are dependent relationships among pattern’s
attributes, such information will be accommodated through additional dimensions, and this can be
realized by a mapping [23]. In SVM literature, the above course is realized through kernel
function
)(),(),( yxyxk φφ= (2)
Kernels can be regarded as generalized dot products [23]. For our experiments we used Gaussian
RBF kernel. A Gaussian RBF kernel is formulated as





 −−
= 2
2
2
||||
exp),(
σ
yx
yxk (3)
3. DATASETS DESCRIPTION
3.1 Wisconsin Diagnostic Breast Cancer Dataset
Features are computed from a digitized image of a fine needle aspirate (FNA) of a breast mass.
They describe characteristics of the cell nuclei present in the image. Number of instances: 569,
Number of attributes: 32 (ID, diagnosis, 30 real-valued input features)

Attribute information
1) ID number
2) Diagnosis (M = malignant, B = benign)
3-32) Ten real-valued features are computed for each cell nucleus:
a) radius (mean of distances from center to points on the perimeter)
b) texture (standard deviation of gray-scale values)
c) perimeter
d) area
e) smoothness (local variation in radius lengths)
f) compactness (perimeter^2 / area - 1.0)
g) concavity (severity of concave portions of the contour)
h) concave points (number of concave portions of the contour)
i) symmetry
j) fractal dimension ("coastline approximation" -1)
The mean, standard error, and "worst" or largest (mean of the three largest values) of these
features were computed for each image, resulting in 30 features. For instance, field 3 is Mean
Radius, field 13 is Radius SE, and field 23 is Worst Radius. All feature values are recoded with
four significant digits. Class distribution: 357 benign, 212 malignant
3.2 Wisconsin Breast Cancer Dataset
This has 699 instances (Benign: 458 Malignant: 241) of which 16 instances has missing attribute
values removing that we have 683 instances of which 444 benign and 239 are malignant. Features
are computed from a digitized image of a Fine Needle Aspiration (FNA) of a breast mass. Table 1
presents the description about the attributes of the WBC dataset
Table 1. Description about the attributes of the WBC dataset
No Attribute Domain
1. Sample code number Id-number
2. Clump thickness 1-10
3. Uniformity of cell size 1-10
4. Uniformity of cell shape 1-10
5. Marginal Adhesion 1-10
6. Single Epithelial cell size 1-10
7. Bare Nuclei 1-10
8. Bland Chromatin 1-10
9. Normal Nucleoli 1-10
10. Mitoses 1-10
11. Class (2 for benign, 4 for malignant)

3.3 Breast Tissue Dataset
This is a dataset with electrical impedance measurements in samples of freshly excised tissue
from the Breast. It consists of 106 instances. 10 attributes: 9 features+1class attribute. Six classes
of freshly excised tissue were studied using electrical impedance measurements. Table 2 presents
the details about the 6 classes and number of cases that belong to those classes.
Table 2. Description about the 6 classes of breast tissue dataset
Class # of cases
Car Carcinoma 21
Fad Fibro-adenoma 15
Mas Mastopathy 18
Gla Glandular 16
Con Connective 14
Adi Adipose 22
Impedance measurements were made at the frequencies: 15.625, 31.25, 62.5, 125, 250, 500, 1000
KHz. These measurements plotted in the (real, -imaginary) plane constitute the impedance
spectrum from where the features are computed. Table 3 presents the description about the
attributes of the breast tissue dataset
Table 3. Description about the attributes of the breast tissue dataset
Id Attribute Description
1 I0 Impedivity (ohm) at zero frequency
2 PA500 phase angle at 500 KHz
3 HFS high-frequency slop e of phase angle
4 DA impedance distance between spectral ends
5 AREA area under spectrum
6 A/DA area normalized by DA
7 MAX IP maximum of the spectrum
8 DR distance between I0 and real part of the
maximum frequency point
9 P length of the spectral curve
4 RESULTS
From the confusion matrix to analyze the performance criterion for the classifiers in detecting
breast cancer, accuracy, precision (for multiclass dataset), sensitivity and specificity have been
computed to give a deeper insight of the automatic diagnosis [24]. Accuracy is the percentage of
predictions that are correct. The precision is the measure of accuracy provided that a specific
class has been predicted. The sensitivity is the measure of the ability of a prediction model to
select instances of a certain class from a data set. The specificity corresponds to the true negative
rate which is commonly used in two class problems. Accuracy, precision, sensitivity and
specificity are calculated using the equations 4, 5, 6 and 7 respectively, where TP is the number
of true positives, TN is the number of true negatives, FP is the number of false positives and FN
is the number of false negatives.
Accuracy =
FNFPTNTP
TNTP
+++
+ (4)
Precision =
FPTP
TP
+
(5)

Sensitivity =
FNTP
TP
+
(6)
Specificity =
FPTN
TN
+
(7)
Table 4, 5, 6 shows the accuracy percentage for WBC, Breast tissue and WDBC datasets
respectively. From the results we can see that all the classifiers except SVM-RBF kernel have
varying accuracies but SVM-RBF kernel always has higher accuracy than the other classifiers for
both binary and multiclass datasets.
Table 4. Accuracy percentage for WBC dataset
Table 5. Accuracy percentage for Breast tissue dataset
Table 6. Accuracy percentage for WDBC dataset
Algorithm Accuracy (%)
Naïve Bayes 92.61
RBF networks 93.67
Trees-J48 92.97
Trees-CART 92.97
SVM-RBF kernel 98.06
WBC and WDBC datasets are binary class datasets whereas breast tissue is a 6 class dataset.
Hence for binary class datasets for measuring the performance criterion sensitivity and specificity
are calculated, for multiclass dataset sensitivity and precision are calculated. Table 7 and 8 shows
the performance criterion such as for WBC and WDBC datasets respectively. Table 9 and 10
shows the percentage of sensitivity and precision for Breast tissue dataset.
Naïve Bayes 96.50
RBF networks 96.66
Trees-J48 94.59
Trees-CART 94.27
Naïve Bayes 94.33
RBF networks 92.45
Trees-J48 95.28
Trees-CART 96.22

Table 7. Performance Criteria for WBC
Algorithm Sensitivity(%) Specificity(%)
Naïve Bayes 95.7 97.8
RBF Networks 95.9 97.8
Trees-J48 94.4 92.6
Trees-CART 94.4 93.9
SVM-RBF kernel 97.2 97.8
Table 8. Performance Criteria for WDBC
Algorithm Sensitivity(%) Specificity(%)
Naïve Bayes 89.6 94.3
RBF Networks 90.0 95.7
Trees-J48 91.5 93.8
Trees-CART 89.1 95.2
SVM-RBF kernel 95.7 99.4
Table 9. Performance Criteria (sensitivity) for Breast tissue
Algorithm Sensitivity(%)
Car Fad Mas Gla Con Adi
Naïve Bayes 100 100 88.8 93.7 92.8 90.9
RBF Networks 100 80.0 88.8 93.7 92.8 95.4
Trees-J48 95.2 93.3 94.4 100 92.8 95.4
Trees-CART 100 93.3 94.4 93.7 92.8 100
SVM-RBF kernel 100 93.3 100 100 100 100
Table 10 Performance Criteria (precision) for Breast tissue
Algorithm Precision(%)
Car Fad Mas Gla Con Adi
Naïve Bayes 95.4 93.7 94.1 100 86.6 95.2
RBF Networks 91.3 100 84.2 93.7 92.8 95.4
Trees-J48 100 93.3 94.4 94.1 92.8 95.4
Trees-CART 100 100 94.4 93.7 92.8 95.6
SVM-RBF kernel 95.4 100 100 100 100 100
From the results we can see that the percentage of sensitivity, specificity and precision of SVM-
RBF kernel is higher than that of other classifiers. SVM-RBF kernel always outperforms than the
other classifiers in performance for both binary and multiclass datasets.
5 CONCLUSION
In this paper we compared the performance criterion of supervised learning classifiers such as
Naïve Bayes, SVM RBF kernel, RBF neural networks, Decision trees J48 and Simple CART.
The aim of this research is to find a best classifier. All the classifiers are used to classify the
breast cancer datasets namely WBC, WDBC and Breast tissue. The experiments were conducted
in WEKA. The results are compared and found that SVM RBF Kernel outperforms other
classifiers with respect to accuracy, sensitivity, specificity and precision for both binary and
multiclass datasets. In future work we propose to analyze the linear and non linear SVM with and
without dimensionality reduction techniques.

REFERENCES
[1] Julie M. David and Kannan Balakrishnan, (2010) “Significance of Classification Techniques In
Prediction Of Learning Disabilities”, International Journal of Artificial Intelligence & Applications
(IJAIA), Vol.1, No.4.
[2] S.J. Cunningham, G Holmes, (1999) “Developing innovative applications in agricultural using data
mining”. In: The Proceedings of the Southeast Asia Regional Computer Confederation Conference.
[3] D.K. Roy, L.K. Sharma, (2010) “Genetic k-Means clustering algorithm for mixed numeric and
categorical data sets”, International Journal of Artificial Intelligence & Applications, Vol 1, No. 2,
pp 23-28.
[4] H. Jiawei and K. Micheline, (2008) Data Mining-Concepts and Techniques, Second Edition, Morgan
Kaufmann - Elsevier Publishers, ISBN: 978-1-55860-901-3.
[5] S. Belciug, (2008) “Bayesian classification vs. k-nearest neighbour classification for the non-invasive
hepatic cancer detection”, Proc. 8th International conference on Artificial Intelligence and Digital
Communications.
[6] F. Gorunescu, (2006) Data Mining: Concepts, models and techniques, Blue Publishing House, Cluj
Napoca.
[7] T. Pang-Ning, S. Michael, K. Vipin, (2008), Introduction to Data Mining, Low Price edn. Pearson,
Education, Inc., London, ISBN 978-81-317-1472-0.
[8] L. Breiman, J. Friedman., R. Olshen, C. Stone, (1984), Classification and Regression Trees.
Wadsworth, Belmont, CA.
[9] D.Steinberg., and P.L. Colla, (1995) ”CART: Tree-Structured Nonparametric Data Analysis”,
Salford Systems: SanDiego, CA.
[10] D.Steinberg., and P.L. Colla, (1997) “CART-Classification and Regression Trees”, Salford Systems:
San Diego, CA.
[11] Ross J. Quinlan, (1992) ” Learning with Continuous Classes”. 5th Australian Joint Conference on
Artificial Intelligence, Singapore, 343-348.
[12] Ross Quinlan, (1993) C4.5: Programs for Machine Learning, Morgan Kaufmann Publishers, San
Mateo, CA.
[13] M. J. L. Orr, (1996) Radial Basis Function Networks. Edinburgh, Scotland.
[14] J. Park, and I.W. Sandberg, (1993) “Approximation and Radial-Basis-Function Networks”, Neural
Computation. 5, 305-316.
[15] S. Haykin, (1994) Neural Networks a Comprehensive Foundation, New Jersey, PrenticeHall.
[16] T. M. Cover, (1965) “Geometrical and Statistical Properties of Systems of Linear with Applications
in Pattern Recognition,” IEEE Transactions on Electronic Computers EC-14, pp. 326-334.
[17] Vladimir N. Vapnik. (1998) Statistical Learning Theory. New York: Wiley.
[18] Vladimir N. Vapnik. (1995) The Nature of Statistical Learning Theory, New York: Springer-Verlag,
[19] C. Campbell, N. Cristianini, and J.Shawe-Taylor, (1999) “Dynamically Adapting Kernels in Support
Vector Machines”, Advances in Neural Information Processing Systems,Vol. 11. MIT Press, 204-210.
[20] Corinna Cortes,Valdimir Vapnik, (1995) “Support-Vector Networks” Machine Learning, 20, 273-
297.
[21] M,Pontil ,A.Verri, (1998) “Support Vector Machines for 3 D object recognition.” IEEE T Pattern
Anal , 20(6):637-646.
[22] You et al, (2010) “A semi-supervised learning approach to predict synthetic genetic interactions by
combining functional and topological properties of functional gene network” BMC Bioinformatics,
11:343.

[23] Bernhard Schölkopf and Alex Smola, (2002) Learning with kernels. MIT Press, Cambridge, MA
[24] D.G. Altman, J.M. Bland, (1994) “Diagnostic tests Sensitivity and specificity”. BMJ 308
(6943):1552.
Authors
S. Aruna is a graduate in Siddha Medicine from Dr M.G.R Medical University,
Guindy, Chennai. Inspired by her software knowledge Prof Dr L.V.K.V Sarma
(Retd HOD, Maths, I.I.T. Chennai) insisted her to do higher studies in computer
science. The author completed her PGDCA as University first and gold medallist and
MCA with first class, distinction. The author is presently a research scholar in Dr
M.G.R University, Maduravoil, Chennai-95 in Dept of Computer Applications under
Dr S.P. Rajagopalan PhD. She is also School first and Science first in class X. Her
research interests include Semi supervised learning, SVM, Bayesian classifiers and
Feature selection techniques.
Dr S.P.Rajagopalan PhD is Professor Emeritus in Dr M.G.R University,
Maduravoil, Chennai-95, India. He was former Dean, College Development
council, Madras UniversiChennai, India. Fifteen scholars have obtained PhD degrees
under his supervision. One hundred and sixty papers have been published in National
and International journals. Currently 20 scholars are pursuing PhD under his
supervision and guidance.
Mr L.V. NandaKishore MSc, MPhil (Mathematics), Assistant Professor, Dept of
Mathematics, Dr M.G.R University, Maduravoil, Chennai-95, currently doing his
PhD in Madras University. He has many publications on Bayesian estimation, asset
pricing and cliquet options. His research interest includes stochastic processes, asset
pricing, fluid dynamics, Bayesian estimation and statistical models.

KNOWLEDGE BASED ANALYSIS OF VARIOUS STATISTICAL TOOLS IN DETECTING BREAST CANCER

More Related Content

What's hot (19)

Similar to KNOWLEDGE BASED ANALYSIS OF VARIOUS STATISTICAL TOOLS IN DETECTING BREAST CANCER (20)

More from cscpconf (20)

Recently uploaded (20)

KNOWLEDGE BASED ANALYSIS OF VARIOUS STATISTICAL TOOLS IN DETECTING BREAST CANCER