Volume: 03, Issue: 02, December 2014

Pages: 430-443

International Journal of Data Mining Techniques and Applications

ISSN: 2278-2419

A Novel Class Imbalance Approach using

Cluster Disjuncts
Syed Ziaur Rahman 1 G. Samuel Vara Prasad Raju 2
1

Research Scholar, Department of CS&SE, University College of Engineering, Andhra University,

Visakhapatnam, A.P, India.

Professor, Department of CS&SE, University College of Engineering, Andhra University, Visakhapatnam,

A.P, India.
E-Mail- [email protected], [email protected]

ABSTRACT:
In Data mining and Knowledge Discovery hidden and valuable knowledge from the data sources is discovered.
The traditional algorithms used for knowledge discovery are bottle necked due to wide range of data sources
availability. Class imbalance is a one of the problem arises due to data source which provide unequal class i.e.
examples of one class in a training data set vastly outnumber examples of the other class(es). Researchers have
rigorously studied several techniques to alleviate the problem of class imbalance, including resampling
algorithms, and feature selection approaches to this problem. In this paper, we present a new hybrid frame work
dubbed as Cluster Disjunct Minority Oversampling Technique (CDMOTE) and Nave Bayes for Cluster
Disjunct (NBCD) for learning from skewed training data. These algorithms provide a simpler and faster
alternative by using cluster disjunct concept. We conduct experiments using fifteen UCI data sets from various
application domains using four algorithms for comparison on six evaluation metrics. The empirical study
suggests that CDMOTE and NBCD have been believed to be effective in addressing the class imbalance
problem.
Keywords Classification, class imbalance, cluster disjunct, CDMOTE, NBCD.
1. INTRODUCTION
A dataset is class imbalanced if the classification
categories are not approximately equally
represented. The level of imbalance (ratio of size of
the majority class to minority class) can be as huge
as 1:99 [1]. It is noteworthy that class imbalance is
emerging as an important issue in designing
classifiers [2], [3], [4]. Furthermore, the class with
the lowest number of instances is usually the class
of interest from the point of view of the learning
task [5]. This problem is of great interest because it
turns up in many real-world classification
problems, such as remote-sensing [6], pollution
detection [7], risk management [8], fraud detection
[9], and especially medical diagnosis [10][13].
There exist techniques to develop better performing
classifiers with imbalanced datasets, which are
generally called Class Imbalance Learning (CIL)
methods. These methods can be broadly divided
into two categories, namely, external methods and
internal methods. External methods involve
preprocessing of training datasets in order to make
them balanced, while internal methods deal with
modifications of the learning algorithms in order to
reduce their sensitiveness to class imbalance [14].
The main advantage of external methods as
previously pointed out, is that they are independent
of the underlying classifier.
Whenever a class in a classification task is under
represented (i.e., has a lower prior probability)

Integrated Intelligent Research (IIR)

compared to other classes, we consider the data as

imbalanced [15], [16]. The main problem in
imbalanced data is that the majority classes that are
represented by large numbers of patterns rule the
classifier decision boundaries at the expense of the
minority classes that are represented by small
numbers of patterns. This leads to high and low
accuracies in classifying the majority and minority
classes, respectively, which do not necessarily
reflect the true difficulty in classifying these
classes. Most common solutions to this problem
balance the number of patterns in the minority or
majority classes. The proposed framework which is
shown in Figure 1 addresses the above said issues
for class imbalance datasets.
Resampling techniques can be categorized into
three groups. Undersampling methods, which
create a subset of the original data-set by
eliminating instances (usually majority class
instances); oversampling methods, which create a
superset of the original data-set by replicating some
instances or creating new instances from existing
ones; and finally, hybrids methods that combine
both sampling methods. Among these categories,
there exist several different proposals; from this
point, we only center our attention in those that
have been used in under sampling. Either way,
balancing the data has been found to alleviate the
problem of imbalanced data and enhance accuracy
[15], [16], [17]. Data balancing is performed by,
e.g., oversampling patterns of minority classes

430

Volume: 03, Issue: 02, December 2014

Pages: 430-443

International Journal of Data Mining Techniques and Applications

ISSN: 2278-2419

either randomly or from areas close to the decision

boundaries. Interestingly, random oversampling is
found comparable to more sophisticated
oversampling methods
[17]. Alternatively,
undersampling is performed on majority classes
either randomly or from areas far away from the
decision boundaries. We note that random
undersampling may remove significant patterns and
random oversampling may lead to overfitting, so
random sampling should be performed with care.
We also note that, usually, oversampling of
minority classes is more accurate than under
sampling of majority classes [17]. In this paper, we
are laying more stress to propose an external class
imbalance learning method for solving the class
imbalance problem.
This paper is organized as follows. Section II
presets the problem of cluster disjucts. Section III
briefly reviews the Data Balancing problems and
its measures and in Section IV, we discuss the
proposed method of CDMOTE (Cluster Disjunct
Minority Oversampling technique) and NBCD
(Nave Bayes for Cluster Disjunct) for CIL. Section
V presents the imbalanced datasets used to validate
the proposed method, while In Section VI, we
present the experimental setting and In Section VII
discuss, in detail, the classification results obtained
by the proposed method and compare them with
the results obtained by different existing methods
and finally, in Section VIII we conclude the paper.
2 Problem of Cluster Disjunct
In Class Imbalance learning, the numbers of
instances in the majority class are outnumbered to
the number of instances in the minority class.
Furthermore, the minority concept
may
additionally contain a sub concept with limited
instances, amounting to diverging degrees of
classification difficulty [35-36]. This, in fact, is the
result of another form of imbalance, a within-class
imbalance, which concerns itself with the
distribution of representative data for sub concepts
within a class [37-39].
The existence of within-class imbalances is closely
intertwined with the problem of small disjuncts,
which has been shown to greatly depreciate
classification performance [37-40]. Briefly, the
problem of small disjuncts can be understood as
follows: A classifier will attempt to learn a concept
by creating multiple disjunct rules that describe the
main concept [35-36], [40]. In the case of
homogeneous concepts, the classifier will generally
create large disjuncts, i.e., rules that cover a large
portion (cluster) of examples pertaining to the main
concept. However, in the case of heterogeneous
concepts, small disjuncts, i.e., rules that cover a
small cluster of examples pertaining to the main
concept, arise as a direct result of underrepresented
sub concepts [35-36], [40]. Moreover, since

Integrated Intelligent Research (IIR)

classifiers attempt to learn both majority and

minority a concept, the problem of small disjuncts
is not only restricted to the minority concept. On
the contrary, small disjuncts of the majority class
can arise from noisy misclassified minority class
examples or underrepresented subconcepts.
However, because of the vast representation of
majority class data, this occurrence is infrequent. A
more common scenario is that noise may influence
disjuncts in the minority class. In this case, the
validity of the clusters corresponding to the small
disjuncts becomes an important issue, i.e., whether
these examples represent an actual subconcept or
are merely attributed to noise. To solve the above
problem of cluster disjuncts we propose the method
cluster disjunct minority oversampling technique
for class imbalance learning.
3 Literature Review
The different imbalance data learning approaches
are as follows [43]:
Table 1. Imbalanced Data learning Approaches
SAMPLING METHODS
BASIC SAMPLING METHODS
Under-Sampling
Over-Sampling
ADVANCED SAMPLING METHODS
Tomek Link
The SMOTE approach
Borderline-SMOTE
One-Sided Selection OSS
Neighbourhood Cleaning Rule (NCL)
Bootstrap-based Over-sampling (BootOS)

ENSEMBLE LEARNING METHODS

BAGGING
Asymmetric bagging, SMOTE Bagging
Over Bagging, Under Bagging
Roughly balanced bagging
Lazy Bagging
Random features selection
BOOSTING
Adaboost
SMOTEBoost
DataBoost-IM
RANDOM FORESTS
Balanced Random Forest BRF
Weighted Random Forest WRF

COST-SENSITIVE LEARNING
Direct cost-sensitive learning methods
Methods for cost-sensitive meta-learning
Cost-sensitive meta-learning
Thresholding methods
MetCost
Cost-sensitive meta-learning sampling
methods

FEATURE SELECTION METHODS

Warpper
PREE (Prediction Risk based feature

431

Volume: 03, Issue: 02, December 2014

Pages: 430-443

International Journal of Data Mining Techniques and Applications

ISSN: 2278-2419

selection for Easy

Ensemble)

ALGORITHMS MODIFICATION
Proposal for new splitting criteria DKM
Adjusting the distribution reference in the
tree
Offset Entropy
In this section, we first review the major research
about clustering in class imbalance learning and
explain why we choose undersampling as our
technique in this paper.
Siti Khadijah Mohamad et al. [18] have conducted
a review to look into how the data mining was
tackled by previous scholars and the latest trends
on data mining in educational research. Hongzhou
Sha et al. [19] have proposed a method named
EPLogCleaner that can filter out plenty of
irrelevant items based on the common prefix of
their URLs.
M.S.B. PhridviRaj et al. [20] have proposed an
algorithm for finding frequent patterns from data
streams by performs only one time scan of the
database initially and uses the information to find
frequent patterns using frequent pattern generation
tree. Chumphol Bunkhumpornpat et al. [21] have
a new over-sampling technique called DBSMOTE
is proposed. DBSMOTE technique relies on a
density-based notion of clusters and is designed to
oversample an arbitrarily shaped cluster discovered
by DBSCAN. DBSMOTE generates synthetic
instances along a shortest path from each positive
instance to a pseudo centroid of a minority-class
cluster. Matas Di Martino et al. [22] have
presented a new classifier developed specially for
imbalanced problems, where maximum F-measure
instead of maximum accuracy guide the classifier
design.
V. Garcia et al. [23] have investigated the influence
of both the imbalance ratio and the classifier on the
performance of several resampling strategies to
deal with imbalanced data sets. The study focuses
on evaluating how learning is affected when
different resampling algorithms transform the
originally imbalanced data into artificially balanced
class distributions. Table 2 presents recent
algorithmic advances in class imbalance learning
available in the literature. Obviously, there are
many other algorithms which are not included in
this table. A profound comparison of the above
algorithms and many others can be gathered from
the references list.

Table 2
Recent advances in Class Imbalance Learning
_______________________________________________________________

ALGORITHM

DESCRIPTION

REFERENECE

_________________________________________
DCEID

Combining ensemble learning

[37]
with cost-sensitive learning.
---------------------------------------------------------------------

Integrated Intelligent Research (IIR)

RUSBoost A new hybrid sampling/boosting

[40]
Algorithm.
--------------------------------------------------------------------CO2RBFN A evolutionary cooperativecompetitive
model for the design of radial-basis
function networks which uses both
[42]
radial-basis function and the
evolutionary cooperative-competitive
technique.

-------------------------------------------------------------Improved Adapt the 2-tuples based genetic tuning

FRBCSs approach to classification problems
[45]
showing the good synergy between
this method and some FRBCSs.
--------------------------------------------------------------------BSVMs A model assessment of the interplay
between various classification
[49]
decisions using probability, corresponding
decision costs, and quadratic program
of optimal margin classifier.
______________________________________________

Mara Dolores Prez-Godoy et al. [24] have

proposed CO2RBFN, a evolutionary cooperative
competitive model for the design of radial-basis
function networks which uses both radial-basis
function and the evolutionary cooperativecompetitive technique on imbalanced domains.
CO2RBFN follows the evolutionary cooperative
competitive strategy, where each individual of the
population represents an RBF (Gaussian function
will be considered as RBF) and the entire
population is responsible for the definite solution.
This paradigm provides a framework where an
individual of the population represents only a part
of the solution, competing to survive (since it will
be eliminated if its performance is poor) but at the
same time cooperating in order to build the whole
RBFN, which adequately represents the knowledge
about the problem and achieves good
generalization for new patterns.
Der-Chiang Li et al. [25] have suggested a strategy
which over-samples the minority class and undersamples the majority one to balance the datasets.
For the majority class, they build up the Gaussian
type fuzzy membership function and a-cut to
reduce the data size; for the minority class, they
used the mega-trend diffusion membership function
to generate virtual samples for the class.
Furthermore, after balancing the data size of
classes, they extended the data attribute dimension
into a higher dimension space using classification
related information to enhance the classification
accuracy.
Enhong Che et al. [26] have described a unique
approach to improve text categorization under class
imbalance by exploiting the semantic context in
text documents. Specifically, they generate new
samples of rare classes (categories with relatively
small amount of training data) by using global
semantic information of classes represented by
probabilistic topic models. In this way, the numbers

432

Volume: 03, Issue: 02, December 2014

Pages: 430-443

International Journal of Data Mining Techniques and Applications

ISSN: 2278-2419

of samples in different categories can become more

balanced and the performance of text
categorization can be improved using this
transformed data set. Indeed, this method is
different from traditional re-sampling methods,
which try to balance the number of documents in
different classes by re-sampling the documents in
rare classes. Such re-sampling methods can cause
overfitting. Another benefit of this approach is the
effective handling of noisy samples. Since all the
new samples are generated by topic models, the
impact of noisy samples is dramatically reduced.
Alberto Fernndez et al. [27] have proposed an
improved version of fuzzy rule based classification
systems (FRBCSs) in the framework of imbalanced
data-sets by means of a tuning step. Specifically,
they adapt the 2-tuples based genetic tuning
approach to classification problems showing the
good synergy between this method and some
FRBCSs. The proposed algorithm uses two
learning methods in order to generate the RB for
the FRBCS. The first one is the method proposed
in [28], that they have named the Chi et al.s rule
generation. The second approach is defined by
Ishibuchi and Yamamoto in [29] and it consists of a
Fuzzy Hybrid Genetic Based Machine Learning
(FH-GBML) algorithm.
J. Burez et al. [30] have investigated how they can
better handle class imbalance in churn prediction.
Using more appropriate evaluation metrics (AUC,
lift), they investigated the increase in performance
of sampling (both random and advanced undersampling) and two specific modeling techniques
(gradient boosting and weighted random forests)
compared to some standard modeling techniques.
They have advised weighted random forests, as a
cost-sensitive learner, performs significantly better
compared to random forests.
Che-Chang Hsu et al. [31] have proposed a method
with a model assessment of the interplay between
various classification decisions using probability,
corresponding decision costs, and quadratic
program of optimal margin classifier called:
Bayesian Support Vector Machines (BSVMs)
learning strategy. The purpose of their learning
method is to lead an attractive pragmatic expansion
scheme of the Bayesian approach to assess how
well it is aligned with the class imbalance problem.
In the framework, they did modify in the objects
and conditions of primal problem to reproduce an
appropriate learning rule for an observation sample.
In [32] Alberto Fernndez et al. have proposed to
work with fuzzy rule based classification systems
using a preprocessing step in order to deal with the
class imbalance. Their aim is to analyze the
behavior of fuzzy rule based classification systems
in the framework of imbalanced data-sets by means
of the application of an adaptive inference system
with parametric conjunction operators. Jordan M.

Integrated Intelligent Research (IIR)

Malof et al. [33] have empirically investigates how

class imbalance in the available set of training
cases can impact the performance of the resulting
classifier as well as properties of the selected set. In
this K-Nearest Neighbor (k-NN) classifier is used
which is a well-known classifier and has been used
in numerous case-based classification studies of
imbalance datasets.
The bottom line is that when studying problems
with imbalanced data, using the classifiers
produced by standard machine learning algorithms
without adjusting the output threshold may well be
a critical mistake. This skewness towards minority
class (positive) generally causes the generation of a
high number of false-negative predictions, which
lower the models performance on the positive
class compared with the performance on the
negative (majority) class.
4 Methodology
In this section, we follow a design decomposition
approach to systematically analyze the different
imbalanced domains. We first briefly introduce the
framework design for our proposed algorithm.
The working style of oversampling tries to generate
synthetic minority instances. Before performing
oversampling on the minority subset, the main
cluster disjuncts has to be identified and the
borderline and noise instances around the cluster
disjuncts are to be removed. The number of
instances eliminated will belong to the k cluster
disjuncts selected by visualization technique. The
remaining cluster disjunct instances have to be
oversampled
by
using
hybrid
synthetic
oversampling technique.
Here, the above said
routine is employed o every cluster disjunct, which
removes examples suffering from missing values at
first and then removes borderline examples and
examples of outlier category.

(a)

(b)

433

Volume: 03, Issue: 02, December 2014

Pages: 430-443

International Journal of Data Mining Techniques and Applications

ISSN: 2278-2419

(c)
(d)
After Applying CDMOTE: (a). Checking Status
(b). Duration (c). Credit History (d). Housing
The j value will change from one dataset to other,
and depending upon the unique properties of the
dataset the value of k can be equal to one also i.e
no cluster disjunct attributes can be identified after
applying visualization technique on the dataset.

(d)
Before (a). Checking Status (b). Duration (c).
Credit History (d). Housing
The algorithm 1: CDMOTE and NBCD can be
explained as follows, the inputs to the algorithm are
majority subclass p and minority class n with
the number of features j. The output of the
algorithm will be the average measures such as
AUC, Precision, F-measure, TP rate and TN rate
produced by the CDMOTE and NBCD methods.
The algorithm begins with initialization of k=1 and
j=1, where j is the number of cluster disjuncts
identified by applying visualization technique on
the subset n and k is the variable used for looping
of j cluster disjuncts.

In another case attributes related cluster disjunct

oversampling can also be performed to improve the
skewed dataset. In any case depending on the
amount of minority examples generated, the final
"strong set" can or cannot be balanced i;e number
of majority instances and minority instances in the
strong set will or will not be equal.
The presented CDMOTE and NBCD algorithms
are summarized as below.
Algorithm 1 CDMOTE or NBCD

Input: A set of major subclass examples P, a set

of minor subclass examples N, jPj < jNj,
and Fj, the feature set, j > 0.
Output: Average Measure { AUC, Precision,
F-Measure, TP Rate, TN Rate}
Phase I: Initial Phase:
1: begin

(a)

2: k 1,j1.
3: Apply Visualization Technique on subset N,
4: Identify cluster disjunct Cj from N, j= number
of cluster disjunct identified in visualization
5: repeat

(b)

6: k=k+1
7: Identify and remove the borderline and outlier
instances for the cluster disjunct Cj.
8: Until k = j
Phase II: Over sampling Phase
9: Apply Oversampling on Cj cluster disjunct

(c)

Integrated Intelligent Research (IIR)

from N,

434

Volume: 03, Issue: 02, December 2014

Pages: 430-443

International Journal of Data Mining Techniques and Applications

ISSN: 2278-2419

10: repeat
11: k=k+1
12: Generate Cj s synthetic positive

efficient and more valuable rules for proper

knowledge discovery.
4.4 Forming the strong dataset

13: Until k = j

The minority subset and majority subset is

combined to form a strong and balance dataset,
which is used for learning of a base algorithm. In
this case we have used C4.5 or Nave Bayes as the
base algorithm.

Phase III: Validating Phase

5 Evaluation Metrics

examples from the minority examples in each

cluster disjunct Cj.

14: Train and Learn A Base Classifier (C4.5) or

(Nave Bayes) using P and N
15: end
The different components of our new proposed
framework are elaborated in the next subsections.
4.1 Preparation of the Majority and Minority
subsets
The datasets is partitioned into majority and
minority subsets. As we are concentrating over
sampling, we will take minority data subset for
further visualization analysis to identify cluster
disjuncts.

To assess the classification results we count the

number of true positive (TP), true negative (TN),
false positive (FP) (actually negative, but classified
as positive) and false negative (FN) (actually
positive, but classified as negative) examples. It is
now well known that error rate is not an
appropriate evaluation criterion when there is class
imbalance or unequal costs. In this paper, we use
AUC, Precision, F-measure, TP Rate and TN Rate
as performance evaluation measures.
Let us define a few well known and widely used
measures:
The Area under Curve (AUC) measure is computed
by equation (1),

4.2 Initial phase of removing noisy and cluster

disjunct borderline instances
Minority subset can be further analyzed to find the
noisy or borderline instances so that we can
eliminate those. For finding the weak instances one
of the ways is that find most influencing attributes
or features and then remove ranges of the noisy or
weak attributes relating to that feature.
How to choose the noisy instances relating to that
cluster disjunct from the dataset set? We can find a
range where the number of samples are less can
give you a simple hint that those instances coming
in that range or very rare or noise. We will
intelligently detect and remove those instances
which are in narrow ranges of that particular cluster
disjunct. This process can be applied on all the
cluster disjuncts identified for each dataset.
4.3 Applying oversampling on cluster disjunct
The oversampling of the instances can be done on
the improved cluster disjuncts produced in the
earlier phase. The oversampling can be done as
follows:
Apply resampling supervised filter on the cluster
disjunct for generating synthetic instances. The
synthetic minority instances generated can have a
percentage of instances which can be replica of the
pure instances and reaming percentage of instances
are of the hybrid quality of synthetic instances
generated by combing two or more instances from
the pure minority sunset. Perform oversampling on
cluster disjunct can help so as to form strong,

Integrated Intelligent Research (IIR)

AUC

1 TPRATE FPRATE
2

________

(1)

The Precision measure is computed by equation

(2),
Pr ecision

TP

TP FP

________ (2)

The F-measure Value is computed by equation (3),

2 Pr ecision Re call
____(3)
Pr ecision Re call
The True Positive Rate measure is computed by
equation (4),
F measure

TruePositi veRate

TP

TP FN

________ (4)

The True Negative Rate measure is computed by

equation (5),
TrueNegati veRate

TN

TN FP

________ (5)

6 Experimental framework
In this study CDMOTE and NBCD are applied to
fifteen binary data sets from the UCI repository
[34] with different imbalance ratio (IR). Table 3
summarizes the data selected in this study and
shows, for each data set, the number of examples
(#Ex.), number of attributes (#Atts.), class name of
each class (minority and majority) and IR.

435

Volume: 03, Issue: 02, December 2014

Pages: 430-443

International Journal of Data Mining Techniques and Applications

ISSN: 2278-2419

In order to estimate different measure (AUC,

precision, F-measure, TP rate and TN rate) we use
a tenfold cross validation approach, that is ten
partitions for training and test sets, 90% for training
and 10% for testing, where the ten test partitions
form the whole set. For each data set we consider
the average results of the ten partitions.
To validate the proposed CDMOTE and NBCD
algorithms, we compared it with the traditional
Support Vector Machines (SVM), C4.5, Functional
Trees (FT), and SMOTE (Synthetic Minority
Oversampling TEchnique).
Table 3 Summary of benchmark imbalanced
datasets
_________________________________________
S.no Datasets # Ex. # Atts.

Class (_,+) IR

_________________________________________
1.Breast 268 9 (recurrence; no-recurrence) 2.37
2. Breast_w 699 9 (benign; malignant)

1.90

3.Colic

(yes; no)

1.71

21 (good; bad)

2.33

368

4.Credit-g

22

1000

5.Diabetes 768 8 (tested-potv; tested-negtv) 1.87

6.Heart-c

303

14 (<50,>50_1)

7.Heart-h

294 14

(<50,>50_1)

1.19
1.77

8.Heart-stat

270 14 (absent, present)

1.25

9.Hepatitis

155 19 (die; live)

3.85

10.Ionosphere 351 34

(b;g)

1.79

11. Kr-vs-kp

3196 37 (won; nowin)

1.09

12. Labor

56

1.85

16

(bad ; good )

13. Mushroom 8124 23 (e ; p )

14. Sick

1.08

15. Sonar
1.15

208 60 (rock ; mine )

_________________________________________
________
7 Results
For all experiments, we use existing prototypes
present in Weka [42]. We compare the following
domain adaptation methods:
We compared proposed methods CDMOTE and
NBCD with the SVM, C4.5 [44], FT, and SMOTE
[41] state-of -the-art learning algorithms. In all the
experiments we estimate AUC, Precision, Fmeasure, TP rate and TN rate using 10-fold crossvalidation. We experimented with 15 standard
datasets for UCI repository; these datasets are
standard benchmarks used in the context of highdimensional imbalance learning. Experiments on
these datasets have 2 goals. First, we study the
class imbalance properties of the datasets using
proposed CDMOTE and NBCD learning
algorithms. Second, we compare the classification
performance of our proposed CDMOTE and
NBCD algorithms with the traditional and class
imbalance learning methods based on all datasets.
Following, we analyze the performance of the
method considering the entire original algorithms,
without pre-processing, data sets for SVM, C4.5
and FT. we also analyze a pre-processing method
SMOTE for performance evaluation of CDMOTE
and NBCD. The complete table of results for all the
algorithms used in this study is shown in Table 4 to
9, where the reader can observe the full test results,
of performance of each approach with their
associated standard deviation. We must emphasize
the good results achieved by CDMOTE and
NBCD, as it obtains the highest value among all
algorithms.

3772 29 (negative ; sick ) 15.32

Table 9 Summary of tenfold cross validation performance for Accuracy on all the datasets

__________________________________________________________________________________________
Datasets

SVM

C4.5

FT

SMOTE

CDMOTE

NBCD

__________________________________________________________________________________________
Breast

67.217.28

74.286.05

Breast_w

96.752.00

95.012.73

Colic

79.786.57

85.165.91

79.11 6.51

Credit-g

68.914.46

71.253.17

71.883.68

Diabetes

76.554.67

74.495.27

Heart-c

81.027.25

Heart-h

81.816.20

Heart-stat

82.076.88

69.837.77

70.235.91

96.162.06

96.581.79

97.9711.503

87.924.70

90.6414.640

76.503.38

75.063.89

76.8444.494

70.62 4.67

76.084.04

81.754.08

79.3334.137

76.946.59

76.066.84

82.994.98

80.576.55

83.0526.371

80.227.95

78.337.54

85.655.46

83.565.81

85.1785.143

83.895.05

80.317.75

81.8727.342

78.157.42

Integrated Intelligent Research (IIR)

68.587.52
95.452.52

76.158.46

88.534.10

73.3566.603

436

Volume: 03, Issue: 02, December 2014

Pages: 430-443
Hepatitis

International Journal of Data Mining Techniques and Applications

ISSN: 2278-2419

81.908.38

79.229.57

Ionosphere 90.264.97

89.744.38

87.105.12

90.284.73

94.643.74

94.4113.590

Kv-rs-kp

99.020.54

99.440.37

90.611.65

99.660.27

99.450.42

98.1031.636

Labor

92.4011.07

78.6016.58

84.3016.24

80.2711.94

88.3311.09

95.9057.259

Mushroom 100.00.00

100.00.00

100.00.000

100.00.00

100.00.00

100.00.00

99.070.50

98.3790.691

86.238.31

86.178.187

Sick

99.260.04

Sonar

75.469.92

81.408.55

98.720.55

96.100.92

73.619.34

86.178.45

78.359.09

83.599.65

97.610.68
82.427.25

89.5298.001

__________________________________________________________________________________________
Table 4 Summary of tenfold cross validation performance for AUC on all the datasets
__________________________________________________________________________________________
Datasets
SVM
C4.5
FT
SMOTE
CDMOTE
NBCD
__________________________________________________________________________________________
Breast
Breast_w
Colic

0.5860.102
0.9770.017
0.8020.073

0.6060.087

0.6040.082

0.9570.034
0.8430.070

0.7170.084

0.9490.030

0.7050.082

0.9670.025

0.9730.018

0.7990.074
0.9910.009

0.7770.072

0.9080.040

0.9000.042

0.9580.029

Credit-g

0.6500.075

0.6470.062

0.6550.044

0.7780.041

0.7880.041

0.8470.043

Diabetes

0.7930.072

0.7510.070

0.6680.051

0.7910.041

0.8360.046

0.8490.040

Heart-c

0.8430.084

0.7690.082

0.7570.069

0.8300.077

0.8220.077

0.9130.052

Heart-h

0.8520.078

0.7750.089

0.7630.082

0.9040.054

0.8690.065

0.9230.043

Heart-stat

0.8640.075

0.7860.094

0.7600.085

0.8320.062

0.8220.076

0.8700.068

Hepatitis

0.7570.195

0.6680.184

0.6780.139

0.7980.112

0.8480.136

0.9520.056

Ionosphere 0.9000.060

0.8910.060

0.9490.041

0.9610.032

Kr-vs-kp

0.9960.005

0.9980.003

0.9060.017

0.9990.001

0.9980.002

0.9950.004

Labor

0.9710.075

0.7260.224

0.8440.162

0.8330.127

0.8700.126

0.9950.024

1.0000.00

1.0000.00

1.0000.00

1.0000.00

0.7950.053

0.9620.025

0.9920.012

0.9790.019

0.8140.090

0.8540.086

0.9240.063

Mushroom 1.0000.00

1.0000.00

Sick

0.9900.014

0.9520.040

Sonar

0.7710.103

0.7530.113

0.8310.067

0.9040.053

0.8590.086

__________________________________________________________________________________________
1(a)
100

1(c)

80
60
Sonar

SVM
C4.5
FT
SMOTE
CDMOTE

1(b)

Diabetes

SVM
C4.5
FT
SMOTE

Sick

SVM
C4.5
FT
SMOTE
CDMOTE
NBCD

1(d)

100

SVM

100

90

C4.5

95

80

FT

90

Ionosphere

Fig. 1(a) (d) Test results on AUC on C4.5, CART, FT, REP, SMOTE and CDMOTE for Sonar,
Ionosphere, Diabetes and Sick Datasets.

Integrated Intelligent Research (IIR)

437

Volume: 03, Issue: 02, December 2014

Pages: 430-443

International Journal of Data Mining Techniques and Applications

ISSN: 2278-2419

Table 5 Summary of tenfold cross validation performance for Precision on all the datasets
__________________________________________________________________________________________
Datasets

SVM

C4.5

FT

SMOTE

CDMOTE

NBCD

__________________________________________________________________________________________
Breast

0.7450.051

0.7530.042

0.7620.051

0.7100.075

0.7130.059

0.7700.062

Breast_w

0.9880.019

0.9650.026

0.9640.026

0.9740.025

0.9860.021

0.9960.011

Colic

0.8450.060

0.8510.055

0.8390.062

0.8530.057

0.8640.059

0.9250.058

Credit-g

0.7760.033

0.7670.025

0.7910.027

0.7680.034

0.7990.044

0.8050.052

Diabetes

0.7930.037

0.7970.045

0.7640.036

0.7810.064

0.8620.050

0.8260.054

Heart-c

0.8250.080

0.7830.076

0.7760.068

0.7790.082

0.8080.087

0.8310.084

Heart-h

0.8490.058

0.8240.071

0.8300.063

0.8780.076

0.8940.072

0.8960.070

Heart-stat

0.8330.078

0.7990.051

0.7960.085

0.7910.081

0.8210.094

0.8280.084

Hepatitis

0.6040.271

0.5100.371

0.5460.333

0.7090.165

0.7390.200

0.7910.151

Ionosphere 0.9060.080

0.8950.084

0.9380.073

0.9340.049

0.9450.047

0.9440.051

Kr-vs-kp

0.9910.008

0.9940.006

0.9050.021

0.9960.005

0.9940.005

0.9780.023

Labor

0.9150.197

0.6960.359

0.8020.250

0.8710.151

0.9210.148

0.9380.122

Mushroom 1.0000.000

1.0000.000

1.0000.000

1.0000.000

1.0000.000

1.0000.000

Sick

0.9970.003

0.9920.005

0.9750.007

0.9830.007

0.9960.004

0.9900.005

Sonar

0.7640.119

0.7280.121

0.8830.100

0.8630.068

0.8510.090

0.8580.092

__________________________________________________________________________________________
Table 6 Summary of tenfold cross validation performance for F-measure on all the datasets
__________________________________________________________________________________________
Datasets

SVM

C4.5

FT

SMOTE

CDMOTE

NBCD

__________________________________________________________________________________________
Breast

0.7810.059

0.8380.040

0.7760.057

0.7300.076

0.7750.049

0.7820.056

Breast_w

0.9650.019

0.9620.021

0.9750.016

0.9600.022

0.9670.018

0.9800.015

Colic

0.8330.055

0.8880.044

0.8380.054

0.8800.042

0.8870.043

0.9080.045

Credit-g

0.8020.027

0.8050.022

0.7790.034

0.7870.034

Diabetes

0.7780.037

0.8060.044

0.8270.038

0.7410.046

Heart-c

0.7820.064

0.7920.059

0.8270.069

0.7720.070

0.8000.069

0.8270.065

Heart-h

0.8300.063

0.8510.061

0.8590.052

0.8410.061

0.8290.066

0.8500.054

Heart-stat

0.7810.083

0.8060.069

0.8410.061

0.7910.072

Hepatitis

0.4690.265

0.4090.272

0.5570.207

Ionosphere 0.7870.098

0.8500.066

Kv-rs-kp

0.9110.016

0.9950.004

Labor

0.7940.211

0.6360.312

Mushroom 1.0000.000

1.0000.000

0.7630.039
0.8080.047

0.7840.041
0.7860.044

0.8020.076

0.8190.077

0.6770.138

0.6930.192

0.8300.129

0.8550.079

0.9050.048

0.9440.039

0.9420.037

0.9910.005

0.9950.004

0.9940.004

0.9810.016

0.7930.132

0.8420.157

0.9540.082

1.0000.000

1.0000.000

1.0000.000

0.9950.003

0.9910.004

0.8790.195
1.0000.000

Sick

0.9790.005

0.9930.003

0.9960.003 0.9870.004

Sonar

0.8440.099

0.7160.105

0.7530.102

0.8610.061

0.8670.082

0.8660.080

__________________________________________________________________________________________

Integrated Intelligent Research (IIR)

438

Volume: 03, Issue: 02, December 2014

Pages: 430-443

International Journal of Data Mining Techniques and Applications

ISSN: 2278-2419

Table 7 Summary of tenfold cross validation performance for TP Rate (Recall) (Sensitivity) on all the
datasets
__________________________________________________________________________________________
Datasets

SVM

C4.5

FT

SMOTE

CDMOTE

NBCD

__________________________________________________________________________________________
Breast

0.8060.091

0.9470.060

0.8150.095

0.7630.117

0.8610.101

.8000.085

Breast_w

0.9670.025

0.9590.033

0.9620.029

0.9470.035

0.9500.033

0.9650.026

Colic

0.8320.075

0.9310.053

0.8350.077

0.9130.058

0.9150.058

0.8960.063

Credit-g

0.8150.041

0.8470.036

0.7830.052

0.8100.058

0.7330.057

0.7670.051

Diabetes

0.7950.054

0.8210.073

0.8680.065

0.7120.089

0.7630.070

0.7530.061

Heart-c

0.7950.095

0.8080.085

0.8370.100

0.7770.110

0.8020.102

0.8310.092

Heart-h

0.8350.093

0.8850.081

0.8760.089

0.8150.084

0.7830.107

0.8160.088

Heart-stat

0.7750.113

0.8240.104

0.8570.090

0.8030.110

0.7940.102

0.8170.102

Hepatitis

0.4480.273

0.3740.256

0.5730.248

0.6810.188

0.7000.247

0.8920.149

Ionosphere 0.6890.131

0.8210.107

0.8200.114

0.8810.071

0.9460.054

0.9430.053

Kv-rs-kp

0.9160.021

0.9950.005

0.9900.007

0.9950.006

0.9950.006

0.9850.012

Labor

0.8450.243

0.6400.349

0.8850.234

0.7650.194

0.8230.227

0.9830.073

Mushroom 1.0000.000

1.0000.000

1.0000.000

1.0000.000

1.0000.000

1.0000.000

Sick

0.9840.006

0.9950.004

0.9950.004

0.9900.005

0.9930.004

0.9920.005

Sonar

0.8200.131

0.7210.140

0.7570.136

0.8650.090

0.8930.109

0.8830.105

__________________________________________________________________________________________
Table 8 Summary of tenfold cross validation performance for TN Rate (Specificity) on all the datasets
__________________________________________________________________________________________
Datasets
SVM
C4.5
FT
SMOTE
CDMOTE
NBCD
__________________________________________________________________________________________
Breast

0.2600.141

0.3350.166

0.1510.164

Breast_w

0.9320.052

0.9770.037

0.9310.060

Colic

0.7170.119

0.7340.118

0.7310.121

Credit-g

0.3980.085

0.4690.098

0.3710.105

Diabetes

0.6030.111

0.4640.169

0.6340.128

0.9840.025

0.9960.012

0.8620.063

0.8410.080

0.9180.069

0.7130.056

0.7720.063

0.7710.073

0.8730.054

0.8340.063

Heart-c

0.7230.119

0.7790.117

0.7170.119

0.8610.068

0.8090.099

0.8300.097

Heart-h

0.6550.158

0.7140.131

0.6360.152

0.8940.074

0.8930.079

0.8910.085

Heart-stat

0.7280.131

0.7750.123

0.6770.152

0.8620.064

0.8120.115

0.8200.098

Hepatitis

0.9000.097

0.8820.092

0.9420.093

0.8370.109

0.8880.097

0.8960.090

Ionosphere

0.9400.055

0.9490.046

0.9330.063

0.9280.057

Kv-rs-kp

0.9930.007

Labor

0.8650.197

0.9450.131

0.8430.214

Mushroom

1.0000.000

1.0000.000

1.0000.000

Sick

0.8750.071

Sonar

0.7490.134

0.5740.095

0.9900.009

0.9740.026
0.7520.148

0.5670.105

0.9870.010

0.8460.080
0.7620.145

0.6220.137
0.9750.024

0.8070.077

0.9980.003

0.9470.047

0.9450.054

0.9940.006

0.9770.025

0.9280.138

0.9460.106

1.0000.000

1.0000.000

1.0000.000

0.8720.053

0.9700.031

0.9190.045

0.8310.113

0.8390.120

0.8470.187

0.7520.113

__________________________________________________________________________________________

Integrated Intelligent Research (IIR)

439

Volume: 03, Issue: 02, December 2014

Pages: 430-443

International Journal of Data Mining Techniques and Applications

ISSN: 2278-2419

Fig.1(a)-(d) shows the average AUC computed for

all approaches, where we can observe that
CDMOTE and NBCD has obtained the best AUC
and accuracy values in the comparison and
therefore it is clearly given the indication of its
supremacy. Table 4, 5, 6, 7, 8 and 9 reports the
results of AUC, Precision, F-measure, TP Rate, TN
Rate and accuracy respectively for fifteen UCI
datasets.

Ionosphere 0.922 0.0710.9450.047 0.9440.051

The bold indicates a win of that algorithm o other

compared algorithms. For each row top two
algorithms are marked as bold. The results in the
tables show that CDMOTE and NBCD have given
a good improvement on all the measures of class
imbalance learning. This level of analysis is
enough for overall projection of advantages and
disadvantages of CDMOTE and NBCD. A twotailed corrected resampled paired t-test is used in
this paper to determine whether the results of the
cross-validation show that there is a difference
between the two algorithms is significant or not.
Difference in accuracy is considered significant
when the p-value is less than 0.05 (confidence level
is greater than 95%). In discussion of results, if one
algorithm is stated to be better or worse than
another then it is significantly better or worse at the
0.05 level.

Breast_w 0.984 0.014 0.9670.018 0.9800.015

Table 9 Summary of comparison of our

proposed approach with a recent published
algorithm CILIUS

Labor 0.754 0.337 0.9210.148 0.9380.122

Sick

0.990 0.006 0.9960.004 0.9900.005

Sonar 0.759 0.112 0.8510.090 0.8580.092

F-measure
Breast 0.812 0.046 0.7750.049 0.7820.056

Colic 0.827 0.073 0.8870.043 0.9080.045

Diabetes 0.836 0.040 0.8080.047 0.7860.044
Hepatitis 0.556 0.238 0.6930.192 0.8300.129
Ionosphere 0.881 0.065 0.9440.039 .9420.037
Labor 0.697 0.307 0.8420.157 0.9540.082
Sick

0.991 0.004 0.9950.003

Sonar

0.9910.004

0.752 0.103 0.8670.082 0.8660.080

TP Rate

Breast

0.325 0.156 0.8610.101 0.8000.085

Breast_w 0.978 0.030 0.9500.033 0.9650.026

Colic

0.765 0.122 0.9150.058 0.8960.063

Diabetes 0.696 0.096 0.7630.070 0.7530.061

Hepatitis 0.920 0.092 0.7000.247 0.8920.149
Ionosphere 0.948 0.052 0.9460.0540.9430.053

_________________________________________

Labor

0.865 0.207 0.8230.227 0.9830.073

Datasets
CILIUS [22] CDMOTE NBCD
_________________________________________

Sick

0.903 0.060 0.9930.004 0.9920.005

Sonar

0.743 0.138 0.8930.109 .8830.105

AUC
Breast

0.637 0.110 0.7050.082 0.7990.074

_________________________________________

Colic 0.873 0.082 0.9000.042 0.9580.029

We can make a global analysis of results

combining the results offered by Tables from 49
and Fig. 1(a)-(d):

Diabetes 0.826 0.056 0.8360.046 .8490.040

Breast_w 0.987 0.016 0.9730.018 0.9910.009

Hepatitis 0.714 0.166 0.8480.136 0.9520.056

Ionosphere 0.917 0.048 0.9490.0410.9610.032
Labor 0.765 0.217 0.8700.126 0.9950.024
Sick

0.950 0.035 0.9920.012 0.9790.019

Sonar 0.774 0.114 0.8540.086 0.9240.063

Precision
Breast 0.736 0.050 0.7130.059 0.7700.062
Breast_w 0.986 0.020 0.9860.021 0.9960.011
Colic 0.787 0.090 0.8640.059

0.9250.058

Diabetes 0.810 0.048 0.8620.050 0.8260.054

Hepatitis 0.698 0.305 0.7390.200 0.7910.151

Integrated Intelligent Research (IIR)

Our proposals, CDMOTE and NBCD are the

best performing one when the data sets are no
preprocessed. It outperforms the preprocessing SMOTE methods and this
hypothesis is confirmed by including standard
deviation variations. We have considered a
complete competitive set of methods and an
improvement of results is expected in the
benchmark algorithms i;e SVM, C4.5, and FT.
However, they are not able to outperform
CDMOTE and NBCD. In this sense, the
competitive edge of CDMOTE and NBCD can
be seen.
Considering that CDMOTE and NBCD
behaves similarly or not effective than
SMOTE shows the unique properties of the
datasets where there is scope of improvement
in majority subset and not in minority subset.
Our CDMOTE and NBCD can only consider

440

Volume: 03, Issue: 02, December 2014

Pages: 430-443

International Journal of Data Mining Techniques and Applications

ISSN: 2278-2419

improvements in minority subset which is not

effective for some unique property datasets.
The contributions of this work are twofold:
A general strategy to handle class imbalance
problem: This is scalable, flexible, and modular,
allowing the many existing supervised methods to
be as a base algorithm. The method achieves
competitive or better results compared to state-ofthe-art baselines.
Table 9 reports the comparison of our proposed
approach with a recent published algorithm
CILIUS [22] and our proposed algorithm has
performed well. A specific methods based on
cluster disjoints for coping up with class imbalance
are proposed. This method naturally captures ad
removes the noise and weak instances for
performing over sampling, which has not been
previously examined in the context of supervised
approach.
We emphasize that our approach is learnerindependent: visualization can be used in
conjunction with many of the existing algorithms in
the literature. Furthermore, the fact that we select
samples in the model space, as opposed to the
feature space, is novel and sets it apart from many
previous approaches to transfer learning (for both
classification and ranking). This allows us to
capture the functional change assumption and
incorporate labeled information in the transfer
learning process.
Finally, we can say that CDMOTE and NBCD are
one of the best alternatives to handle class
imbalance problems effectively. This experimental
study supports the conclusion that a cluster disjunct
approach for cluster detections and elimination can
improve the class imbalance learning behavior
when dealing with imbalanced data-sets, as it has
helped the CDMOTE and NBCD methods to be the
best performing algorithms when compared with
four classical and well-known algorithms: SVM,
C4.5 and FT and a well-established pre-processing
technique SMOTE.

CDMOTE and NBCD are much less volatile than

C4.5.
In our future work, we will apply CDMOTE and
NBCD to more learning tasks, especially high
dimensional feature learning tasks. Another
variation of our approach in future work is to
analyze the influence of different base classifier
effect on the quality of synthetic minority instances
generated.
References:

1. J. Wu, S. C. Brubaker, M. D. Mullin, and J. M.

2.
3.
4.
5.

6.

7.

8.

9.

8. Conclusion
Class imbalance problem have given a scope for a
new paradigm of algorithms in data mining. The
traditional and benchmark algorithms are
worthwhile for discovering hidden knowledge from
the data sources, meanwhile class imbalance
learning methods can improve the results which are
very much critical in real world applications. In this
paper we present the class imbalance problem
paradigm, which exploits the cluster disjunct
concept in the supervised learning research area,
and implement it with C4.5 and NB as its base
learners. Experimental results show that CDMOTE
and NBCD have performed well in the case of
multi class imbalance datasets. Furthermore,

Integrated Intelligent Research (IIR)

10.

11.

Rehg, Fast asymmetric learning for cascade

face detection, IEEE Trans. Pattern Anal.
Mach. Intell., vol. 30, no. 3, pp. 369382, Mar.
2008.
N. V. Chawla, N. Japkowicz, and A. Kotcz,
Eds., Proc. ICML Workshop Learn.
Imbalanced Data Sets, 2003.
N. Japkowicz, Ed., Proc. AAAI Workshop
Learn. Imbalanced Data Sets, 2000.\
G. M.Weiss, Mining with rarity: A unifying
framework, ACM SIGKDD Explor. Newslett.,
vol. 6, no. 1, pp. 719, Jun. 2004.
N. V. Chawla, N. Japkowicz, and A. Kolcz,
Eds., Special Issue Learning Imbalanced
Datasets, SIGKDD Explor. Newsl., vol. 6, no.
1, 2004.
W.-Z. Lu and D.Wang, Ground-level ozone
prediction by support vector machine approach
with a cost-sensitive classification scheme,
Sci. Total. Enviro., vol. 395, no. 2-3, pp. 109
116, 2008.
Y.-M. Huang, C.-M. Hung, and H. C. Jiau,
Evaluation of neural networks and data
mining methods on a credit assessment task for
class imbalance problem, Nonlinear Anal. R.
World Appl., vol. 7, no. 4, pp. 720747, 2006.
D. Cieslak, N. Chawla, and A. Striegel,
Combating imbalance in network intrusion
datasets, in IEEE Int. Conf. Granular
Comput., 2006, pp. 732737.
M. A. Mazurowski, P. A. Habas, J. M. Zurada,
J. Y. Lo, J. A. Baker, and G. D. Tourassi,
Training neural network classifiers for
medical decision making: The effects of
imbalanced
datasets
on
classification
performance, Neural Netw., vol. 21, no. 23,
pp. 427436, 2008.
A. Freitas, A. Costa-Pereira, and P. Brazdil,
Cost-sensitive decision trees applied to
medical data, in Data Warehousing Knowl.
Discov. (Lecture Notes Series in Computer
Science), I. Song, J. Eder, and T. Nguyen,
Eds.,
K.Kilic,O zgeUncu and I. B. Turksen,
Comparison of different strategies of utilizing
fuzzy clustering in structure identification,

441

Volume: 03, Issue: 02, December 2014

Pages: 430-443

12.

13.

14.

15.

16.

17.

18.

19.

20.

21.

22.

23.

International Journal of Data Mining Techniques and Applications

ISSN: 2278-2419

Inf. Sci., vol. 177, no. 23, pp. 51535162,

2007.
M. E. Celebi, H. A. Kingravi, B. Uddin, H.
Iyatomi, Y. A. Aslandogan, W. V. Stoecker,
and R. H. Moss, A methodological approach
to the classification of dermoscopy images,
Comput.Med. Imag. Grap., vol. 31, no. 6, pp.
362373, 2007.
X. Peng and I. King, Robust BMPM training
based on second-order cone programming and
its application in medical diagnosis, Neural
Netw., vol. 21, no. 23, pp. 450457,
2008.Berlin/Heidelberg, Germany: Springer,
2007, vol. 4654, pp. 303312.
Rukshan Batuwita and Vasile Palade (2010)
FSVM-CIL: Fuzzy Support Vector Machines
for Class Imbalance Learning, IEEE
TRANSACTIONS ON FUZZY SYSTEMS,
VOL. 18, NO. 3, JUNE 2010, pp no:558-571.
N. Japkowicz and S. Stephen, The Class
Imbalance Problem: A Systematic Study,
Intelligent Data Analysis, vol. 6, pp. 429-450,
2002.
M. Kubat and S. Matwin, Addressing the
Curse of Imbalanced Training Sets: One-Sided
Selection, Proc. 14th Intl Conf. Machine
Learning, pp. 179-186, 1997.
G.E.A.P.A. Batista, R.C. Prati, and M.C.
Monard, A Study of the Behavior of Several
Methods for Balancing Machine Learning
Training Data, SIGKDD Explorations, vol. 6,
pp. 20-29, 2004.1.
Siti Khadijah Mohamada, Zaidatun Tasir.
Educational data mining: A review, Procedia
- Social and Behavioral Sciences 97 ( 2013 )
320 324.
Hongzhou Sha, Tingwen Liu, Peng Qin, Yong
Sun, Qingyun Liu. EPLogCleaner: Improving
Data Quality of Enterprise Proxy Logs for
Efficient Web Usage Mining Procedia
Computer Science 17 ( 2013 ) 812 818.
M.S.B. PhridviRaj, C.V. GuruRao. Data
mining past, present and future a typical
survey on data Streams, Procedia Technology
12 ( 2014 ) 255 263.
Chumphol
Bunkhumpornpat,
Krung
Sinapiromsaran, Chidchanok Lursinsap.
DBSMOTE:
Density-Based
Synthetic
Minority Over-sampling Technique Appl
Intell (2012) 36:664684.
Matas Di Martino, Alicia Fernndez, Pablo
Iturralde, Federico Lecumberry. Novel
classifier scheme for imbalanced problems,
Pattern Recognition Letters 34 (2013) 1146
1151.
V. Garcia, J.S. Sanchez , R.A. Mollineda, On
the effectiveness of preprocessing methods
when dealing with different levels of class

Integrated Intelligent Research (IIR)

24.

25.

26.

27.

28.
29.

30.

31.

32.

33.

34.

imbalance, Knowledge-Based Systems 25

(2012) 1321.
Mara
Dolores
Prez-Godoy,
Alberto
Fernndez, Antonio Jess Rivera, Mara Jos
del Jesus, Analysis of an evolutionary RBFN
design algorithm, CO2RBFN, for imbalanced
data sets, Pattern Recognition Letters 31
(2010) 23752388.
Der-Chiang Li, Chiao-WenLiu, SusanC.Hu,
A learning method for the class imbalance
problem with medical data sets, Computers in
Biology and Medicine 40 (2010) 509518.
Enhong Che, Yanggang Lin, Hui Xiong,
Qiming Luo, Haiping Ma, Exploiting
probabilistic topic models to improve text
categorization under class imbalance,
Information Processing and Management 47
(2011) 202214.
Alberto Fernndez, Mara Jos del Jesus,
Francisco Herrera, On the 2-tuples based
genetic tuning performance for fuzzy rule
based classification systems in imbalanced
data-sets, Information Sciences 180 (2010)
12681291.
Z. Chi, H. Yan, T. Pham, Fuzzy Algorithms
with Applications to Image Processing and
Pattern Recognition, World Scientific, 1996.
H. Ishibuchi, T. Yamamoto, T. Nakashima,
Hybridization of fuzzy GBML approaches for
pattern
classification
problems,
IEEE
Transactions on System, Man and Cybernetics
B 35 (2) (2005) 359365.
J. Burez, D. Van den Poel, Handling class
imbalance in customer churn prediction,
Expert Systems with Applications 36 (2009)
46264636.
Che-Chang Hsu, Kuo-Shong Wang, ShihHsing Chang, Bayesian decision theory for
support
vector
machines:
Imbalance
measurement and feature optimization, Expert
Systems with Applications 38 (2011) 4698
4704.
Alberto Fernndez, Mara Jos del Jesus,
Francisco Herrera, On the influence of an
adaptive inference system in fuzzy rule based
classification systems for imbalanced datasets, Expert Systems with Applications 36
(2009) 98059812.
Jordan M. Malof, Maciej A. Mazurowski,
Georgia D. Tourassi, The effect of class
imbalance on case selection for case-based
classifiers: An empirical study in the context
of medical decision support, Neural Networks
25 (2012) 141145.
A. Asuncion D. Newman. (2007). UCI
Repository of Machine Learning Database
(School of Information and Computer
Science), Irvine, CA: Univ. of California
[Online].
Available:

442

Volume: 03, Issue: 02, December 2014

Pages: 430-443

35.
36.

37.

38.
39.

International Journal of Data Mining Techniques and Applications

ISSN: 2278-2419

http://www.ics.uci.edu/mlearn/MLRepositor
y.html
J.R. Quinlan, Induction of Decision Trees,
Machine Learning, vol. 1, no. 1, pp. 81-106,
1986.
T. Jo and N. Japkowicz, Class Imbalances
versus Small Disjuncts, ACM SIGKDD
Explorations Newsletter, vol. 6, no. 1, pp. 4049, 2004.
N. Japkowicz, Class Imbalances: Are We
Focusing on the Right Issue? Proc. Intl Conf.
Machine Learning, Workshop Learning from
Imbalanced Data Sets II, 2003.
R.C. Prati, G.E.A.P.A. Batista, and M.C.
Monard, Class Imbalances versus Class
Overlapping: An Analysis of a Learning
System Behavior, Proc. Mexican Intl Conf.
Artificial Intelligence, pp. 312-321, 2004.

Integrated Intelligent Research (IIR)

40. G.M. Weiss, Mining with Rarity: A Unifying

41.

42.

43.

44.

Framework, ACM SIGKDD Explorations

Newsletter, vol. 6, no. 1, pp. 7-19, 2004.
N. Chawla, K. Bowyer, and P. Kegelmeyer,
SMOTE: Synthetic minority over-sampling
technique, J. Artif. Intell. Res., vol. 16, pp.
321357, 2002.
Witten, I.H. and Frank, E. (2005) Data Mining:
Practical machine learning tools and
techniques. 2nd edition Morgan Kaufmann,
San Francisco.
Mohamed Bekkar and Dr. Taklit Akrouf
Alitouche, 2013. Imbalanced Data Learning
Approaches Review. International Journal of
Data Mining & Knowledge Management
Process (IJDKP) Vol.3, No.4, July 2013
J. R. Quinlan, C4.5: Programs for Machine
Learning, 1st ed. San Mateo, CA: Morgan
Kaufmann Publishers, 1993.

443