Content uploaded by K. Usha Rani
Author content
All content in this area was uploaded by K. Usha Rani on Apr 22, 2016
Content may be subject to copyright.
ANALYSIS OF FEATURE SELECTION
WITH CLASSFICATION: BREAST
CANCER DATASETS
D.Lavanya *
Department of Computer Science, Sri Padmavathi Mahila University
Tirupati, Andhra Pradesh, 517501, India
lav_dlr@yahoo.com
http://www.spmvv.ac.in
Dr.K.Usha Rani
Department of Ccomputer Science, Sri Padmavathi Mahila University
Tirupati, Andhra Pradesh, 517501, India
usharanikurubar@yahoo.co.in
http://www.spmvv.ac.in
Abstract
Classification, a data mining task is an effective method to classify the data in the process of Knowledge Data
Discovery. A Classification method, Decision tree algorithms are widely used in medical field to classify the
medical data for diagnosis. Feature Selection increases the accuracy of the Classifier because it eliminates
irrelevant attributes. This paper analyzes the performance of Decision tree classifier-CART with and without
feature selection in terms of accuracy, time to build a model and size of the tree on various Breast Cancer
Datasets. The results show that a particular feature selection using CART has enhanced the classification
accuracy of a particular dataset.
Keywords: Data Mining; Feature Selection Classification; Decision Tree; CART; Breast Cancer Datasets.
1. Introduction
Knowledge Data Discovery (KDD) is a process of deriving hidden knowledge from databases. KDD consists of
several phases like Data cleaning, Data integration, Data selection, Data transformation, Data mining, Pattern
evaluation, Knowledge representation. Data mining is one of the important phases of knowledge data discovery.
Data mining is a technique which is used to find new, hidden and useful patterns of knowledge from large
databases. There are several data mining functions such as Concept descriptions, Association Rules,
Classification, Prediction, Clustering and Sequence discovery to find the useful patterns.
Data preprocessing is applied before data mining to improve the quality of the data. Data preprocessing
includes data cleaning, data integration, data transformation and data reduction techniques. Cleaning is used to
remove noisy data and missing values. Integration is used to extract data from multiple sources and storing as a
single repository. Transformation transforms and normalizes the data in a consolidated form suitable for mining.
Reduction reduces the data by adopting various techniques i.e., aggregating the data, attribute subset selection,
dimensionality reduction, numerosity reduction and generation of concept hierarchies. The data reduction is also
called as feature selection. Feature selection or attribute selection identifies the relevant attributes which are
useful to the data mining task. Applying feature selection with data mining technique improves the quality of the
data by removing irrelevant attributes.
Classification is extensively used in various application domains: retail target marketing, fraud detection,
design of telecommunication service plans, Medical diagnosis, etc [Brachman.R et al., 1996], [K.U.M Fayyad et
al.,1996]. In the domain of medical diagnosis classification plays an important role. Since large volume of data
maintained in the medical field, classification is extensively used to make decisions for diagnosis and prognosis
of patient’s disease. Decision tree classifiers are used extensively for diagnosis of diseases such as breast cancer,
ovarian cancer and heart sound diagnosis and so on [Antonia Vlahou et al.,2003], [Kuowj et al.,2001], [Stasis
D.Lavanya et al./ Indian Journal of Computer Science and Engineering (IJCSE)
ISSN : 0976-5166
Vol. 2 No. 5 Oct-Nov 2011
756
A.C et al.,2003]. Feature selection with decision tree classification greatly enhances the quality of the data in
medical diagnosis.
In this study we considered three breast cancer datasets for experimental purpose to analyze the performance
of decision tree classifier, CART algorithm with various feature selection methods to find out whether the same
feature selection method may lead to best accuracy on various datasets of same domain. The paper has been
organized in six sections. Section 2 describes the related work to this study. Section 3 deals the fundamental
concepts of classification and decision trees. In Section 4 Feature Section mechanisms are presented. The
experimental results are presented with explanation in Section 5 followed by conclusions in Section 6.
2. Related Work
Classification is one of the most fundamental and important task in data mining and machine learning.
Many of the researchers performed experiments on medical datasets using decision tree classifier. Few are
summarized here:
In 2010, Asha Gowda Karegowda et al.[ Asha Gowda Karegowda et al.,2010] proposed a wrapper
approach with genetic algorithm for generation of subset of attributes with different classifiers such as
C4.5,Naïve Bayes, Bayes Networks and Radial basis functions. The above classifiers are experimented on
the datasets Diabetes, Breast cancer, Heart Statlog and Wisconsin Breast cancer.
Aboul Ella Hassanien [Hassaneian, 2003] in 2003 had experimented on breast cancer data using feature
selection technique to obtain reduced number of relevant attributes, further decision tree–ID3 algorithm is used
to classify the data.
In 2005, Kemal Polat et al. [Kemal Polat et al., 2005], proposed a new classification algorithm feature
selection-Artificial Immune Recognition System (FS-AIRS) on breast cancer data set. To reduce the data set
C4.5 decision tree algorithm is used as a feature selection method.
Deisy.C et al. in 2007 [Deisy. C et al., 2007] experimented breast cancer data using three feature selection
methods Fast correlation based feature selection, Multi thread based FCBF feature selection and Decision
dependent-decision independent correlation further the data is classified using C4.5 decision tree algorithm.
Mark A. Hall et al. [Mark A. Hall et.al., 1997] in 1997 have done experiments on various data sets using
Correlation based filter feature selection approach further the reduced data is classified using C4.5 decision tree
algorithm.
In 2011, D. Lavanya et al. [D.Lavanya et al., 2011] analyzed the performance of decision tree classifiers on
various medical datasets in terms of accuracy and time complexity.
3. Classification
Classification [ J.Han et al.,2000], a data mining task which assigns an object to one of several pre-defined
categories based on the attributes of the object. The input to the problem is a data-set called the training
set, which consists of a number of examples each having a number of attributes. The attributes are either
continuous, when the attribute values are ordered, or categorical when the attribute values are unordered. One
of th e categorical attributes is called the class label or the classifying attribute. The objective is to use the
training set to build a model of the class label based on the other attributes such that the model can
be used to classify new data not from the training data-set. Classification has been studied extensively in
statistics, machine learning, neural networks and expert systems over decades [Mitchell, 1997]. There are
several classification methods:
• Decision tree algorithms
• Bayesian algorithms
• Rule based algorithms
• Neural networks
• Support vector machines
• Associative classification
• Distance based methods
• Genetic Algorithms
D.Lavanya et al./ Indian Journal of Computer Science and Engineering (IJCSE)
ISSN : 0976-5166
Vol. 2 No. 5 Oct-Nov 2011
757
3.1 Decision Trees
Decision tree induction [J.Han et al., 2000] is a very popular and practical approach for pattern classification.
Decision tree is constructed generally in a greedy, top down recursive manner. The tree can be constructed in a
breadth first manner or depth first manner. Decision tree structure consists of a root node, internal nodes and
leaf nodes. The classification rules are derived from the decision tree in the form of - if then else. These rules are
used to classify the records with unknown value for class label. The decision tree is constructed in two phases:
Building Phase and Pruning Phase.
In Building phase of the tree the best attributes are selected based on attribute selection measures such as
Information gain, Gain Ratio, Gini Index, etc. Once the best attribute is selected then the tree is constructed with
that node as the root node and the distinct values of the attribute are denoted as branches. The process of
selecting best attribute and representing the distinct values as branches are repeated until all the instances in the
training set belong to the same class label.
In Pruning phase the sub trees are eliminated which may over fit the data. This enhances the accuracy of a
classification tree. Decision trees handle continuous and discrete attributes. Decision trees are widely used
because they provide human readable rules, easy to understand, construction of decision tree is fast and it yields
better accuracy.
There are several algorithms to classify the data using decision trees. The frequently used decision tree
algorithms are ID3, C4.5 and CART [Matthew N Anyanwu et al,]. In this study the CART algorithm is chosen
to classify the breast cancer data because it provides better accuracy for medical data sets than ID3, C4.5
algorithms [D.Lavanya et al., 2011]. CART [Breiman et al., 1984] stands for Classification and Regression
Trees introduced by Breiman. It is based on Hunt’s algorithm. CART handles both categorical and continuous
attributes to build a decision tree. It also handles missing values. CART uses Gini Index as an attribute selection
measure to build a decision tree. Unlike ID3 and C4.5 algorithms, CART produces binary splits. Hence, it
produces binary trees. Gini Index measure does not use probabilistic assumptions like ID3 [Quinlan, 1986],
C4.5 [Quinlan, 1992]. CART uses cost complexity pruning to remove the unreliable branches from the decision
tree to improve the accuracy.
4. Feature Selection
Feature selection (FS) plays an important role in classification. This is one of the Preprocessing techniques in
data mining. Feature selection is extensively used in the fields of statistics, pattern recognition and medical
domain. Feature Selection means reducing the number of attributes. The attributes are reduced by removing
irrelevant and redundant attributes, which do not have significance in classification task. The feature selection
improves the performance of the classification techniques. The process of feature selection is
• Generation of candidate subsets of attributes from original feature set using searching
techniques.
• Evaluation of each candidate subset to determine the relevancy towards the
classification task using measures such as distance, dependency, information,
consistency, classifier error rate.
• Termination condition to determine the relevant subset or optimal feature subset.
• Validation to check the selected feature subset.
The feature selection process [Mark A.Hall et al., 1997] is represented in figure 1.
Fig. 1. Feature Selection Process
D.Lavanya et al./ Indian Journal of Computer Science and Engineering (IJCSE)
ISSN : 0976-5166
Vol. 2 No. 5 Oct-Nov 2011
758
Feature selection methods are classified as filter, wrapper and hybrid approaches. The filter approach applies
to the data before the classification. In this method features are evaluated by using heuristics based on general
characteristics of the data. In the wrapper approach, the features are evaluated using the classification
algorithms. In Hybrid approach features are evaluated using filter and wrapper approaches. Further, the reduced
dataset is considered for classification.
5. Experimental Results
In this experiment the medical data related to breast cancer is considered because the breast cancer is one of the
leading causes of death in women. The experiments are conducted using weka tool. In this study CART
algorithm is chosen to analyze the breast cancer datasets because it provides better accuracy for medical data
sets than the other two frequently used decision tree algorithms ID3 and C4.5 [D.Lavanya et al., 2011]. With an
intension to find out whether the same feature selection method may lead to best accuracy for various datasets of
same domain, various experiments are conducted on three different breast cancer datasets. Further to analyze
the importance of feature selection in decision tree classifier-CART, we considered three different Breast
Cancer Datasets: Breast Cancer, Breast Cancer Wisconsin (Original) and Breast Cancer Wisconsin (Diagnostic)
with different attribute types. The data is collected from UCI machine learning repository [www.ics.uci.edu ]
which is publicly available. The description of the datasets is given in Table 1.
Table 1: Description of Breast Cancer Datasets
For consistency the missing values in the datasets are replaced with mean value of the respective attribute.
The experiments are conducted on the above datasets with and without feature selection methods. And the
results are compared and analyzed. The performance of the classifier is analyzed in terms of accuracy, time
taken to execute to a model and tree size.
The performance of CART algorithm related to breast cancer datasets without feature selection is shown in
table 2.
Table 2: CART algorithm – without feature selection
Datasets Accuracy (%) Time (Secs) Tree Size
Breast Cancer 69.23 0.23 5
Breast Cancer Wisconsin (Original) 94.84 0.44 15
Breast Cancer Wisconsin (Diagnostic) 92.97 0.73 17
Further the experiments are conducted on 13 feature section methods. The supporting search techniques
vary form one FS method to another. Hence, by considering the entire supporting search techniques related to a
particular feature selection method various experiments are conducted and the results are shown in the tables 3 -
7. The search technique with best accuracy is highlighted for quick reference in those tables.
The highlighted best search technique corresponding to a particular feature selection method for all Breast
Cancer Datasets is posted in the tables 8-10. By comparing all the feature selection methods the
SVMAttributeEval method with accuracy of 73.03% is best for Breast Cancer Dataset,
PrincipalComponentsAttributeEval method is the best one with accuracy of 96.99% for Breast Cancer
Wisconsin (Original) Dataset and SymmetricUncertAttributesetEval method is best for Breast Cancer Wisconsin
(Diagnostic) Dataset with 94.72% accuracy .
Dataset No. of Attributes No. of Instances No. of Classes Missing values
Breast Cancer 10 286
2 yes
Breast Cancer Wisconsin (Original) 11 699 2 yes
Breast Cancer Wisconsin (Diagnostic) 32 569 2 no
D.Lavanya et al./ Indian Journal of Computer Science and Engineering (IJCSE)
ISSN : 0976-5166
Vol. 2 No. 5 Oct-Nov 2011
759
Table 3: FS Method - CfssSubsetEval.
Breast Cancer Dataset Breast Cancer Wisconsin (Original) Breast Cancer Wisconsin (Diagnostic)
Search
Technique
Reduced
No of
Attributes
Accuracy
(%) Time
(Sec) Tree
Size
Reduced
No of
Attributes
Accuracy
(%) Time Time
(Sec)
Reduced
No of
Attributes
Accuracy
(%) Time
(Sec)e Tree
Size
Best-First 5 70.27 0.14 5 9 94.84 0.17 15 12 92.97 0..3 15
Exhaustive 5 70.27 0.14 5 9 94.84 0.14
15 - - - -
Genetic 5 70.27 0.14 5
9 94.84 0.14
15 14 93.49 0.44 5
Greedy-
step wise 5 71.32 0.14
5 9 94.84 0.14
15 12 92.61 0.24 15
Linear
forward
selection 5 70.27 0.13 5 9 94.84 0.14
15 12 92.97 0..3 15
Random 6 70.27 0.28 5
9 94.84 0.14
15 - - - -
Rank 5 70.97 0.13 5
9 94.84 0.14
15 11 93.32 0..3 15
Scatter 5 70.27 0.11 5 9 94.84 0.27 15 12 92.26 0.28 15
Subsetsize
forward
selection 5 71.32 0.19 5 9 94.84 0.24 15 12 92.79 0.58 15
Table 4: FS Method - ClassifierSubsetEval
Breast Cancer Dataset Breast Cancer Wisconsin (Original) Breast Cancer Wisconsin (Diagnostic)
Search
Technique
Reduced
No of
Attributes
Accuracy
(%) Time
(Sec) Tree
Size
Reduced
No of
Attributes
Accuracy
(%) Time
(Sec) Tree
Size
Reduced
No of
Attributes
Accuracy
(%) Time
(Sec) Tree
Size
Best-First 6 67.83 6.25 29 6 93.99 3.31 19 7 93.32 48.34 27
Exhaustive 6 63.98 57.84 39 6 94.84 42.17 37
- - - -
Genetic 6 67.48 8.58 29 6 93.41 10.45 37
9 94.05 87.31 25
Greedy-
step wise 4 66.78 2.03 11 3 94.70 1.92 15 5 94.02 20.59 27
Linear
forward
selection 6 67.13 7.75 29 6 94.13 4.06 19 11 92.79 68.39 25
Race 5 70.62 19.61 7 3 94.99 5.2 9 7 92.99 50.19 15
Random 6 66.78 14.95 29 7 93.70 13.16 37 - - - -
Rank 3 70.62 5.52 9 6 95.13 1.23 17
9 93.76 30.05 15
Scatter 4 70.27 5.52 9 4 94.42 5.94 33 11 92.97 28.78 25
Subset size
forward
selection 4 67.48 2.25 11 3 94.13 2.42 15 7 93.32 28.56 15
Table 5: FS Method - ConsistencySubsetEval
Breast Cancer Dataset Breast Cancer Wisconsin (Original) Breast Cancer Wisconsin (Diagnostic)
Search
Technique
Reduced
No of
Attributes
Accuracy
(%) Time
(Sec) Tree
Size
Reduced
No of
Attributes
Accuracy
(%) Time
(Sec) Tree
Size
Reduced
No of
Attributes
Accuracy
(%) Time
(Sec) Tree
Size
Best-First 7 69.58 0.2 7 7 93.70 0.13 9 7 94.20 0.66 21
Exhaustive 7 69.23 0.61 7 6 95.13 0.41 9 - - - -
Genetic 7 69.23 0.2 7 6 94.84 0.17 37 9 92.44 0.66 15
Greedy-
step wise 7 69.23 0.17 7 7 93.99 0.16 9 7 94.20 0.27 21
Linear
forward
selection 8 70.27 0.2 7 7 93.84 0.14 9 7 93.14 0.36 23
Random 9 69.23 0.22 5 7 93.56 0.22 37 - - - -
Rank 9 69.23 0.2 5 9 94.84 0.13 15 21 92.97 0.45 17
Scatter 9 70.62 0.44 5 7 94.56 0.2 19
10 94.55 2.8 13
Subset size
forward
selection 8 70.27 0.3 5 7 94.27 0.25 9 7 93.14 0.78 23
Table 6(Continued): FS Method - FilteredSubsetEval
Breast Cancer Dataset Breast Cancer Wisconsin (Original) Breast Cancer Wisconsin (Diagnostic)
Search
Technique
Reduced
No of
Attributes
Accuracy
(%) Time
(Sec) Tree
Size
Reduced
No of
Attributes
Accuracy
(%) Time
(Sec) Tree
Size
Reduced
No of
Attributes
Accuracy
(%) Time
(Sec) Tree
Size
Best-First 5 70.27 0.13 5 9 94.84 0.11 15
9 92.61 0.25 7
D.Lavanya et al./ Indian Journal of Computer Science and Engineering (IJCSE)
ISSN : 0976-5166
Vol. 2 No. 5 Oct-Nov 2011
760
Exhaustive 5 70.27 0.14 5 9 94.84 0.13 15 - - - -
Genetic 5 71.32 0.14 5 9 94.84 0.13 15 12 93.49 0.28 19
Greedy-step wise 5 71.32 0.13 5
9 94.84 0.16 15 9 92.26 0.2 7
Linear forward selection 5 70.27 0.14 5 9 94.84 0.14 15 9 92.61 0.39 7
Random 6 70.27 0.14 5 9 94.84 0.14 15 - - - -
Rank 5 70.27 0.16 5 9 94.84 0.13 15 7 92.97 0.39 5
Scatter 5 70.27 0.13 5 9 94.84 0.13 15 9 92.26 0.27 7
Subset size forward selection 5 71.32 0.19 5 9 94.84 0.24 15 9 92.44 0.24 7
Table 7: Other Feature Selection methods
Breast Cancer Dataset Breast Cancer Wisconsin (Original) Breast Cancer Wisconsin
(Diagnostic)
Feature Selection Method Search
Tech-
nique
Reduced
No of
Attri-
butes
Accu-
racy
(%)
Time
(Sec) Tree
Size
Reduced
No of
Attri-
butes
Accu-
racy
(%)
Time
(Sec) Tree
Size
Reduced
No of
Attri-
butes
Accu-
Racy
(%)
Time
(Sec) Size
ChiSquaredAtrributeEval Ranker 9 69.23 0.22 9 9 94.56 0.09 15 31 92.61 0.66 17
FilteredAttributeEval Ranker 9 69.23 0.39 9 9 94.56 0.13 15 31 92.61 0.81 17
InfoGainAttributeEval Ranker 9 69.23 0.22 9 9 94.56 0.13 15 31 92.61 0.47 17
GainRatioAttributeEval Ranker 9 69.23 0.22 9 9 94.42 0.13 15 31 92.26 0.59 17
ReliefFAttributeEval Ranker 9 69.23 0.3 9 9 94.56 0.83 15 31 92.79 1.78 17
PrincipalComponentsAttributeEval Ranker 9 70.63 0.47 9 9 96.99 0.19 3 11 92.09 0.41 21
SVMAttributeEval Ranker 9 73.07 19.3 9 9 94.56 1.41 15 9 94.56 1.41 15
SymmetricUncertAttributeEval Ranker 9 69.23 0.22 9 9 94.42 0.14 15 31 92.26 0.72 17
SymmetricUncertAttributesetEval FCBF 2 66.78 0.19 2 8 93.99 0.13 9 8 94.72 0.27 19
Table 8: Result of all feature selection methods - Breast Cancer Dataset
Feature selection Technique Reduced No. of Attributes Accuracy (%) Time (Sec) Tree size
CfssSubsetEval 5 71.32 0.14 5
ChiSquaredAtrributeEval 9 69.23 0.22 5
ClassifierSubsetEval 6 95.13 1.23 17
ConsistencySubsetEval 9 70.62 0.44 5
FilteredAttributeEval 9 69.23 0.39 5
FilteredSubsetEval 5 71.32 0.13 5
GainRatioAttributeEval 9 69.23 0.22 5
InfoGainAttributeEval 9 69.23 0.22 5
ReliefFAttributeEval 9 69.23 0.3 5
PrincipalComponentsAttributeEval 9 70.63 0.47 5
SVMAttributeEval 9 73.07 19.3 5
SymmetricUncertAttributeEval 9 69.23 0.22 5
SymmetricUncertAttributesetEval 2 66.78 0.19 1
D.Lavanya et al./ Indian Journal of Computer Science and Engineering (IJCSE)
ISSN : 0976-5166
Vol. 2 No. 5 Oct-Nov 2011
761
Table 9: Result of all feature selection methods - Breast Cancer Wisconsin (Original) Dataset.
Feature selection Technique Reduced No. of attributes Accuracy (%) Time (Sec) Tree size
CfssSubsetEval 9 94.84 0.14 15
ChiSquaredAtrributeEval 9 94.56 0.09 15
ClassifierSubsetEval 6 95.13 1.23 17
ConsistencySubsetEval 6 95.13 0.41 9
FilteredAttributeEval 9 94.56 0.13 15
FilteredSubsetEval 9 94.84 0.11 15
GainRatioAttributeEval 9 94.42 0.13 15
InfoGainAttributeEval 9 94.56 0.13 15
ReliefFAttributeEval 9 94.56 0.83 15
PrincipalComponentsAttributeEval 9 96.99 0.19 3
SVMAttributeEval 9 94.56 1.41 15
SymmetricUncertAttributeEval 9 94.56 1.41 15
SymmetricUncertAttributesetEval 8 93.99 0.13 9
Table 10: Result of all feature selection methods - Breast Cancer Wisconsin (Diagnostic) Dataset
Feature selection Technique Reduced No. of Attributes Accuracy (%) Time (Sec) Tree size
CfssSubsetEval 14 93.49 0.44 5
ChiSquaredAtrributeEval 31 92.61 0.66 17
ClassifierSubsetEval 9 94.05 87.31 25
ConsistencySubsetEval 10 94.55 2.8 13
FilteredAttributeEval 31 92.61 0.81 17
FilteredSubsetEval 12 93.49 0.28 19
GainRatioAttributeEval 31 92.26 0.59 17
InfoGainAttributeEval 31 92.61 0.47 17
ReliefFAttributeEval 31 92.79 1.78 17
PrincipalComponentsAttributeEval 11 92.09 0.41 21
SVMAttributeEval 9 94.56 1.41 15
SymmetricUncertAttributeEval 31 92.26 0.72 17
SymmetricUncertAttributesetEval 8 94.72 0.27 19
6. Conclusion
Accuracy is most important in the field of medical diagnosis to diagnose the patient’s disease. Experimental
results show that Feature Selection, a Preprocessing technique greatly enhances the accuracy of classification.
We also conclude that the classifier accuracy has been surely enhanced by the use of any of Feature selection
method than the classifier accuracy without feature selection. With an intension to find out whether the same
feature selection method may lead to best accuracy for various datasets of same domain, various experiments are
conducted on three different breast cancer datasets. The performance of Decision tree classifier-CART with and
without feature selection in terms of accuracy, time to build a model and size of the tree on various Breast
Cancer Datasets are observed. From the results it is clear that, though we considered only breast cancer
datasets, a specific feature selection may not lead to the best accuracy for all Breast Cancer Datasets. The best
feature selection method for a particular dataset depends on the number of attributes, attribute type and
instances. Hence, whenever another dataset is considered, one has to experiment on that with various feature
selection methods to identify the best one to enhance the classifier accuracy instead of simply considering the
previously proved one related to the same domain. Once the best feature selection method is identified for a
particular dataset the same can be used to enhance the classifier accuracy.
References
[1] Matthew N.Anyanwu, Sajjan G.Shiva, “Comparative Analysis of Serial Decision Tree Classification Algorithms”, International
Journal of Computer Science and Security, volume 3.
[2] R. Brachman, T. Khabaza, W.Kloesgan, G.Piatetsky-Shapiro and E. Simoudis, “Mining Business Databases”, Comm. ACM, Vol. 39,
no. 11, pp. 42-48, 1996.
[3] Breiman, Friedman, Olshen, and Stone. “Classification and Regression Trees”, Wadsworth, 1984, Mezzovico, Switzerland.
[4] Deisy.C, Subbulakshmi.B, Baskar S, Ramaraj.N, Efficient Dimensionality Reduction Approaches for Feature Selection, Conference on
Computational Intelligence and Multimedia Applications, 2007.
[5] K U.M. Fayyad, G. Piatetsky-Shapiro and P. Smyth, “From Data Mining to knowledge Discovery in Databases”, AI Magazine, vol 17,
pp. 37-54, 1996.
[6] Mark A. Hall, Lloyd A. Smith, Feature Subset Selection: A Correlation Based Filter Approach, In 1997 International Conference
on Neural Information Processing and Intelligent Information Systems (1997), pp. 855-858.
[7] Aboul Ella Hassaneian, Classification and feature selection of breast cancer data based on decision tree algorithm, Studies and
Informatics Control ,vol12, no1,March 2003.
[8] J. Han and M. Kamber, “Data Mining; Concepts and Techniques, Morgan Kaufmann Publishers”, 2000.
D.Lavanya et al./ Indian Journal of Computer Science and Engineering (IJCSE)
ISSN : 0976-5166
Vol. 2 No. 5 Oct-Nov 2011
762
[9] Asha Gowda Karegowda, M.A.Jayaram, A.S. Manjunath, Feature Subset Selection Problem using Wrapper Approach in Supervised
Learning, International Journal of Computer Applications 1(7):13–17, February 2010.
[10] Kuowj, Chang RF,Chen DR and Lee CC,” Data Mining with decision trees for diagnosis of breast tumor in medical ultrasonic
images” ,March 2001.
[11] D.Lavanya, Dr.K.Usha Rani, Performance Evaluation of Decision Tree Classifiers on Medical Datasets. International Journal of
Computer Applications 26(4):1-4, July 2011.
[12] T. Mitchell, “Machine Learning”, McGraw Hill, 1997.
[13] Kemal Polat, Seral Sahan, Halife Kodaz and Salih Günes, A New Classification Method for Breast Cancer Diagnosis: Feature
Selection Artificial Immune Recognition System (FS-AIRS), In Proceedings of ICNC (2)'2005. pp.830~838
[14] J.R.Quinlan,”c4.5: Programs for Machine Learning”, Morgan Kaufmann Publishers, Inc, 1992.
[15] Quinlan, J.R, “Induction of decision trees”. Journal of Machine Learning 1(1986) 81-106.
[16] Stasis, A.C. Loukis, E.N. Pavlopoulos, S.A. Koutsouris, D. “Using decision tree algorithms as a basis for a heart sound diagnosis
decision support system”, Information Technology Applications in Biomedicine, 2003. 4th International IEEE EMBS Special Topic
Conference, April 2003.
[17] Antonia Vlahou, John O. Schorge, Betsy W.Gregory and Robert L. Coleman, “Diagnosis of Ovarian Cancer Using Decision Tree
Classification of Mass Spectral Data”, Journal of Biomedicine and Biotechnology • 2003:5 (2003) 308–314.
[18] www.ics.uci.edu/~mlearn/MLRepository.html
D.Lavanya et al./ Indian Journal of Computer Science and Engineering (IJCSE)
ISSN : 0976-5166
Vol. 2 No. 5 Oct-Nov 2011
763
