Hybrid Information Gain Method and Bagging in Data Classification
using Support Vector Machine
Immanuel H. G. Manurung, Tulus and Poltak Sihombing
Departement of Computer Science and Information Technology, University of Sumatera Utara,
Jl. Dr. T. Mansur No.9, Medan, Indonesia
Keywords: Selection Attributes, SVM, Bagging, IG, Fold Cross-Validation
Abstract: The selection process is very influential attributes of dataset in SVM algorithm which tends to produce good
accuracy on the results of the classification (classifier) SVM is not optimal. To reduce the effect of selection
attributes on SVM classifiers, it is necessary to apply a combination of methods of feature selection algorithms
that are Bootstrapping Aggregation (Bagging) and methods of Information Gain (IG). The application of the
algorithm Bagging the feature selection is done to give weight to each feature are recommended, so that the
found feature is a strong classifier, whereas IG focuses on identifying attributes and evaluate the impact of a
beneficial features based on ranking the features that can be recommended to the classifier SVM in the process
classification. Experiments implementation of Information Gain feature selection techniques that use
attributes with election threshold level. The results showed that, the performance accuracy of SVM classifiers
in dataset by combining IG before bagging process, by setting the value thresold >= 0.02 and a 10-fold cross-
validation, show that with the implementation of information gain feature selection techniques can improve
the performance of machine learning classification algorithm.
1 INTRODUCTION
1.1 Background
Data mining technology is one tool for data mining in
large data bases and the specification level of
complexity that has been widely used in many
application domains such as banking, fraud detection
and telecommunications fields .Some research has
also been done using data mining techniques to gather
information from a database, such as to analyze the
performance (performance) of students in the
learning process (Kalles & Pierrakeas, 2006) as well
as to help teachers to manage classes diampunya
(Agathe & Kalina, 2005) and it is possible to analyze
and evaluate the academic data for know the quality
of higher education (AlRadaideh et al, 2006).
Selection of the right features and the classifier is
essential in improving the accuracy and computing in
the classification. Feature selection technique is done
to reduce the irrelevant features and reduce the
dimensions of the features in the data. A feature
selection techniques to reduce the dimension
attributes. The dimensional reduction is done to get
the attributes that are relevant and not excessive so as
to speed up the classification process and can improve
the accuracy of classification algorithms. (Arifin,
2015). Feature selection method used in this research
is the Information gain. The method will perform
computing process to obtain the attributes that most
influence on the dataset. Dinakaran et al implement
feature selection techniques Information Gain
method of ranking the best features and classification
Decision Tree algorithm J48 (Dinakaran et al, 2013).
Ramaswami and Bhaskaran conduct a comparative
study of five feature selection techniques and apply
four classification algorithms in data mining
education. The experimental results indicate that
Information Gain feature selection techniques
showed the best results (Ramaswami & Bhaskaran,
2009).
In this study, used Information Gain for feature
selection and bagging to improve the accuracy of
classification in machine learning. "Bagging is a
method that can improve the results of the
classification of machine learning algorithms
(Breimann, 1994)". "This method formulated by Leo
Breiman and the name is inferred from the phrase"
Bootstrap aggregating "(Breimann, 1994)". Bagging
194
Manurung, I., Tulus, . and Sihombing, P.
Hybrid Information Gain Method and Bagging in Data Classification using Support Vector Machine.
DOI: 10.5220/0010040401940201
In Proceedings of the 3rd International Conference of Computer, Environment, Agriculture, Social Science, Health Science, Engineering and Technology (ICEST 2018), pages 194-201
ISBN: 978-989-758-496-1
Copyright
c
2021 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved
is a method based on the ensemble method, several
studies applying meta-algorithm, these include the
research for planning and environmental changes
(Kang et al, 2016), research for the development of
systems to track visual (Chang and Hsu, 2015),
enhancing the performance related to research in the
field of pharmacy (Galvan, 2015), improved
performance in the field of voice processing (Chen,
2015),
Therefore, in this study the proposed Information
Gain for feature selection algorithms and meta
bagging to improve the classification performance of
SVM method. Two stages in building a hybrid
scheme, the first step is the selection of the features
considered relevant for classification and there
construct the second phase of the hybrid intelligent
model scheme. Further features that are considered
relevant will be used as input in the classification with
SVM.
1.2 Problem Formulation
Issues to be addressed in this study is how much the
value of accuracy obtained from the application of
Information Gain for feature selection and bagging to
improve the accuracy of classification in machine
learning using Support Vector Machine (SVM) in the
form of hybrid models.
1.3 Limitations
Boundary problem discussed in this research is to
implement a meta model of machine learning bagging
on classification Support Vector Machine (SVM)
with Information Gain feature selection.
1.4 Research Objectives
The purpose of this study is to model meta bagging
on classification machine learning Support Vector
Machine (SVM), to improve the accuracy of each
algorithm values and compare them.
1.5 Benefits Research
The benefits of this research are:
1. Can be used as reference material to expand
horizons in research specifically in the
classification of machine learning.
2. Can be used to provide the best information in the
selection of attributes to get the attributes that
most influence on the dataset.
2 LITERATURE
2.1 Data Mining
Data mining is a concept used to find hidden
knowledge in the database. Data mining is a semi-
automatic process that uses statistical techniques,
mathematics, artificial intelligence, and machine
learning to extract and identify potential knowledge
and useful information that is stored in large
databases. Data mining is part of the process of KDD
(Knowledge Discovery in Databases), which consists
of several stages such as the selection of data, pre-
processing, transformation, data mining, and
evaluating results. In terms of other data mining is
also
defined as the process to obtain useful
information from large data base warehouse.
Data mining is a series of processes for adding
additional value of a set of data in the form of
knowledge that had been unknown to them manually.
Said mining means that efforts to get a bit of valuables
from a large number of basic material. Data mining is
the process of finding patterns and relationships
hidden in large amounts of data for the purpose of
classification, estimation, prediction, association rule,
clustering, description and visualization. In a simple
data mining can be regarded as a process of filtering
or "mining" knowledge from large amounts of data.
The process and data filtering technique determines
the quality of knowledge and information that will be
obtained. Another term for data mining is Knowledge
Discovery in Databases (KDD).
Data cleaning
Cleaning of the data was performed to remove noise
and inconsistent data
Data integration.
Data integration process undertaken to
menggabungkandata from various sources.
Data Selection.
Selection data is done to retrieve the relevant data,
which will be used for process analysis in data
mining.
Data transformation.
This process is done to transform the data into the
proper form for in-mine.
Data Mining.
Data mining is the process for applying a
metodeuntuk extract patterns in the data.
Evaluation pattern.
Evaluation of the pattern is needed to identify some
interesting patterns representing knowledge.
Presentation of knowledge.
Represents the knowledge that had been dug up to the
user to visualize such knowledge.
Hybrid Information Gain Method and Bagging in Data Classification using Support Vector Machine
195
2.2 Support Vector Machine
Support Vector Machine(SVM) was developed by
Boser, Guyon, Vapnik, and was first presented at the
1992 Annual Workshop on Computational Learning
Theory. The basic concept of Support Vector
Machine (SVM) is actually a harmonious
combination of theories of computing has been
around for decades before, such as margin hyperplane
(Duda & Hart 1973, Cover 1965, Vapnik 1964, and
others), the kernel was introduced by Aronszajn 1950,
and likewise with the concepts supporting the other.
However, until 1992, there has never been an attempt
assembling these components.
Concept of Support Vector Machine (SVM) can
be explained simply as an attempt to find the best
hyperplane which serves as a separator of two classes
in the input space. pattern that is a member of two
classes: +1 and -1 and share alternative dividing lines
(boundaries discrimination). Margin is the distance
between the hyperplane to the nearest pattern of each
kelas.Pattern closest is called a support vector.
Attempts to locate this hyperplane is the core of the
learning process on SVM (Christianini, 2000).
Initialization of training data is used, given the
label in the form of

N
i
ii
xy
1
,
, with

n
ii
Rxy ,1,1
using formulations C-SVM.
For the linear case, the algorithm is shown to find the
best hyperplane to separate the data by minimizing
the following function:

N
i
i
Cww
1
2
2
1
,
(2.1)
for

iii
bxwy
1
and
0
i
, Where C>
0 is a trade off of the constraints. Kernel function K
data point (x, z): Rn x Rn R is the result of inner
product of

zx
, Data mapping function

x
and

z
mentioned very difficult to find
values of high yield dimensional inner product value
is equal to K (x, z). Because of these difficulties, the
system of direct control of the kernel function is used.
Then to get the optimal hyperplane equation, can be
used the following equation:
 

N
i
iji
ij
jii
xxKyyW
1
,
2
1
(2.2)
for :
C
i
0
and
0
1
N
i
ii
y
(2.3)
the function of the decision is

xfsign
, Where :

bxxKyxf
m
i
iii
1
,
(2.4)
Note that Equation 3 only require access to the
kernel function alone, without having to perform
mapping data by function

.
, M is the number of
support vectors. So this allows one to solve
formulations in high-dimensional feature space with
a highly efficient, step is called the kernel trick. For
linear kernel, we can use the kernel functions
zxzxK
,
And function hyperplane
bxwxf
, Where the vector w can be
calculated by the formula
m
i
iii
xyw
1
and
xwxwb ..
2
1
, If in case of non-linear
vector w can be calculated by the formula
m
i
iii
xyw
1
and the constant b by the
formula

xwxwb
.
2
1
,
Additive selected specifically for the kernel to use the
formula
n
i
i
zxKzxK
1
),(,
and

xf
can
be written as

bxwxf
,
2.3 Meta-algorithm
Meta-algorithm is an algorithm that uses another
algorithm as a representative, and also an algorithm
that has a sub-algorithms as variables and parameters
can be changed. Examples include meta-algorithm is
boosting, simulated annealing, bootstrap aggregating,
AdaBoost, and random-restart hill climbing.
2.3.1 Algorithm boosting
Boosting is an meta-algorithm in machine learning
to perform supervised learning. Boosting theory
introduced by questions Kearns (1988): Can a set of
weak learner creates a strong unity learner? Weak
learner is a classifier that has little correlation with
actual classification, while strong learner is a
classifier which has a strong correlation with the
actual classification.
Most of boosting algorithms follow a plan.
Generally boosting occurs in iterations, incrementally
adding weak learner into a strong learner. At each
iteration, the weak learner to learn from a workout
data. Then, the weak learner was added to the strong
learner. After a weak learner is added, the data is then
converted their respective weights. The data were
misclassified will experience weight gain, and the
data were classified correctly will experience a
ICEST 2018 - 3rd International Conference of Computer, Environment, Agriculture, Social Science, Health Science, Engineering and
Technology
196
reduction in weight. Therefore, the weak learner on
the next iteration will be more focused on data
misclassified by the weak learner that before.
Of the many variations of boosting algorithm,
AdaBoost is the most famous in the history of its
development, is the first algorithm that can adapt to
the weak learner. However, there are some examples
of other boosting algorithms such as LPBoost,
TotalBoost, BrownBoost, MadaBoost, LogitBoost,
and others. Most boosting algorithm can be
incorporated into the framework AnyBoost.
2.3.2 Metode Bagging (Bootstrap
Aggregating)
This combined method is a method models each of
which is a combination of a lineup that includes a
number of k-models of the M
1
, M
2
..., M
k
, With the
aim to create an increase in the combined models, M
*. In general there are two methods of combination
of models, namely Bagging and Boosting. Both can
be used in the process of classification and
prediction (Han and Kamber, 2006). Illustrations
combined methods can be seen in Figure 2.1.
Figure 2.1: Illustration Model Combination Methods
Pada Figure 2, there are descriptions of the work
steps of the method combination of models, where in
the method aims to improve the accuracy of the model
consisting of Bagging and Boosting method. This
method produces a classification or prediction
models, M
1
, M
2
..., M
k
, Where voting strategy
against emerging choice is a combination of strategies
used to combine the predictions for the object that has
been given the unknown category.
Bagging method is the theory proposed by
Breiman (1996), which is based on the concept of a
bootstrap theory and aggregating that combine both
benefits into a single theory. Bootstrap is applied
based on the theory of random sampling with
replacement (Tibshirani and Efron, 1993). In effect,
the model in engineering classification (classifier)
that has been formed is likely to have a better
performance. Understanding of the process of
aggregating is incorporating some of the classifier.
Sometimes combined classifier gives better
results compared to the classifier only, due to the
incorporation of both the benefits of some of the
classifier at the end Bagging penyelesaiannya.Oleh
therefore helpful to help build or establish a better
classifier on the training data samples. Here are the
stages of implementation Bagging technique:
1. Perform as many Xb bootstrap replication of a
number n m training data. Repeat these steps for
1,2, ..., B. Where m is the number of data that is
taken from the training data, n is the sample size
of the training data and B is the number of
bootstrap replication is done.
2. By using a simple majority vote, been labeled the
most widely emerged from the assessment results
as a rule for making a final decision.
2.4 Hybrid Algorithm
Hybrid or Hybrid Algorithm algorithm is an
algorithm that combines two or more methods into
one or several models combined to produce items
according to the user's wishes (Burke, R., 2007).
Broadly speaking, this algorithm has two stages. The
first stage is the algorithm generate node ordering and
second stages of the algorithm constructs a structure.
3 RESEARCH METHODOLOGY
All data is primary data collected from UCI Machine
Learning Repository, Research steps are generally
depicted in the flow diagram Figure 3.1
Figure 3.1 Block diagram of the stages of research
Hybrid Information Gain Method and Bagging in Data Classification using Support Vector Machine
197
There are several stages in this study, the dataset
collection, pre-processing, a single model, and hybrid
models. After data collection, pre-processing is done
data by eliminating the data contained missing value.
In the single phase model of the result data
preprocessing The classified using SVM. The final
stage is to establish a scheme hybrid intelligent model
to produce some hybrid combinations. At the stage of
the model development hybrid, there is a feature
selection to select features considered relevant as an
input for classification.
3.1 Variable Data and Research
This study will use data on Diabetes Diagnosis of UCI
Machine Learning Repository. Overall, the data
Diabetes Diagnosis contains 9 attributes described in
the following table.
Table 3.1.
No. Attribute
1.
p
re
g
nancies
2. PG Concentration
3. diastolic BP
4. Tri Fold Thic
k
5. serum Ins
6. BMI
7. DP Function
9 A
g
e
10. diagnosis
4 RESULTS AND DISCUSSION
In this section explored further on the implementation
of the proposed system architecture. In testing, the
evaluation techniques used cross validation models to
observe and analyze the performance measurement
results. Tests were conducted at four methods of
classification. In each method to do a comparison
between the performance of the classification by
application of meta-adaptive boosting algorithm and
without the application of this meta-algorithm. To
measure the performance of this classifier is used
seven sizes, namely accuracy, Kappa, MAE,
Precision, Recall, F-Measure, and the ROC.
4.1 Read Data Training
Training data used contains as many as 768 data line
containing the seven attributes that have been
described in the early chapters. To specify the data
that will be analyzed then the first step is to read
training data.The following are some of the training
data will be used.
4.2 Testing Single Model SVM Method
From a probability value above 768 will be tested as
much data as the data and completed using weka tools
so that the resulting classification result as in Figure
4.2.
Figure 4.4: Testing Results with SVM method
a. Accuracy obtained is 77.3438% of total
Correctly classified instances as many as 594.
b. Number of instances was incorrectly classified
as much as 174 or 22.6563%.
c. The results of the root of the mean squared
error is 0.476.
4.3 Testing Results with Information
Gain
From the calculation of the value of the determined
value overlapping provisions of the discriminant
(discriminant ability). Average - average value of the
provisions of the discriminant (discriminant ability)
of each feature on each - each class is divided by the
number of changes made reference in calculating the
suspension of each feature. Where the features with
the smallest score is the highest rating in ranking and
can be recommended in the classification process.
The following ranking of feature selection using the
IG shown in Table 4.2.
ICEST 2018 - 3rd International Conference of Computer, Environment, Agriculture, Social Science, Health Science, Engineering and
Technology
198
Table 4.2: Rating Each Attribute
Ranked Attribute Score
1 PG Concentration 0.1901
2 BMI 0.0749
3 Age 0.0725
4 serum Ins 0.0595
5 Tri Fold Thic
k
0.0443
6
p
re
g
nancies 0.0392
7 DP Function 0.0208
8 diastolic BP 0014
In the third experiment, started by applying feature
selection using the IG. The results of ranking the
features of the IG, the percentage of moderately by
70% the results of the second experiment had the
highest accuracy value will be recommended to the
bagging algorithm. From the selection by bagging
features that the Hx value equal to +1 is decent
features recommended and the value is equal to -1 Hx
is a feature that is not recommended in classification.
Here are the results of measurement with IG before
bagging shown in Table 4.3.
Figure 4.5: Selection Attributes with the Information Gain
Information gain feature selection techniques applied
by using perangkingan attributes by using the selected
threshold level features by using threshold >= 0,02.
4.4 Testing with Bagging - SVM
In the fourth experiment begins by applying feature
selection using bagging to perform weighting on
seven features recommended so that a strong
classifier and further evaluation of the impact of a
beneficial features based on ranking features using
IG. The measurement results of ranking the fourth
feature selection in the experiment are shown in Table
4.5.
Figure 4.6: Testing Results with bagging-SVM models
In the figure indicates that the value of the highest
accuracy by applying feature selection IG after
bagging in a 10-fold cross-validation is equal to
82.14%. From the results of measurements performed
some experiments can be concluded that the increase
in performance accuracy, precision and sensitivity
(recall) SVM models, are affected by the application
of feature selection in the SVM classifier and the
determination of the number-fold cross validation.
That is, if the value-fold cross validation the greater
the value of accuracy, precision and sensitivity
(recall) tends to increase and will get better reliability
SVM models.
4.5 Results Discussion
In the experimental study, there are two models:
Single classification results using SVM for hybrid
models while there are 2 classification accuracy
results, respectively for hybrid models with bagging
classifier. For hybrid models bagging scheme can
improve accuracy. Accuracy using only one classifier
was 77.34% and increased accuracy using hybrid
scheme with bagging-SVM amounted to 81.14%.
Because the hybrid scheme has fewer features in
addition to mengkatkan accuracy, computation
process will be faster. This shows that by using more
features does not improve accuracy, but by choosing
the features that are considered relevant so as to
reduce the size of features with the right method and
the right classifier will improve accuracy and speed
up the computation.
Hybrid Information Gain Method and Bagging in Data Classification using Support Vector Machine
199
5 CONCLUSION
5.1 Conclusion
Based on the results of experimental studies hybrid
model with bagging method for classifying a number
of conclusions as follows:
1. Bagging highly adaptive method with the
features, because each iteration in bagging an
election classifier that has the smallest error.
bagging choose the best feature in every
iteration. So with many or few features that are
used or with any data, bagging would classify
properly.
2. The research concluded that the hybrid scheme
with classifier bagging on classification has
been proven to improve accuracy and speed up
the computation. By using a single classifier
accuracy of 77.34% increased by 81.14% using
a hybrid scheme.
5.2 Suggestions
For further research, bagging metaalgoritme will be
developed that are not too sensitive to outliers (data
outliers), so that optimal performance of
metaalgoritme over again.
REFERENCES
Agathe, and Kalina, 2005, the Educational Data Mining: A
Case Study, Pole UniversitaireLéonard de Vinci,
France
Al-Radaideh, QA, Al-Shawakfa, EM, and AlNajjar, MI,
2006, the Student DataUsing Mining Decision Trees,
The 2006 International Arab Conference on
Information Technology (ACIT'2006).
Birant, D. 2011.Comparison of Decision Tree Algorithms
for Predicting Potential Air Pollutant Emissions
Models with Data Mining. Journal of Environmental
Informatics.Vol. 17 Issue 1, p46-53. 8p.
Christianini, N. & Taylor, SJ 2000.An Introduction to
Support Vector Machine and other Kernel-Based
Learning methods, Cambridge University Press.
Chang and Hsu, "Development of a Visual Compressive
Trackng System Enhanced by Adaptive Boosting," in
41st Annual Conference of the IEEE Industrial
Electronics Society (IECON 2015), pp. 3678-3682,
2015.
Chen, Y. Li, and X. Xu, "Rotating Target Classification
base on Micro-doppler Features Using a Modified
Adaptive Boosting Algorithm," International
Conference on Computers, Communications, adn
Systems, pp. 236-240, 2015.
Dekker, W. Gerben., Et.al. (2009). Predicting Students
Drop Out: ACase Study. Proceedings of the 2 nd
International Conference on EducationalData mining.
41-50.
Deb, AK, Member, Student, Member, senior, Gopal, M., &
Chandra, S. (2007). SVM-Based Tree-Type Neural
Networks as a Critic in Adaptive Critic Designs for
Control. IEEE, 18 (4), 1016-1030.
DPK Muhammad Yunus, Prediction Model Design
Graduate Student With Decision Tree algorithm,
Matrix, vol. 2, no. 13, pp. 1-5, 2015.
Erdogan, SZ, Timor, M. (2005) .A Data Mining
Applications in Student Database. Journal of
Aeronautics andSpace Technologies. Vol 2 (2) .53-57.
Galvan, "Integrase Inhibition Using Differential Evolution-
Binary Particle Swarm Optimization and Non-Linear
AdaptiveBoosting Random Forest Regression," 16th
International Conference on Information Reuse and
Integration, pp. 485-490, 2015.
Han, J., et al. Data Mining: Concepts and Techniques 2nd
Edition, Morgan Kaufmann Publishers, 2006.
Han J, Kamber M. 2001. Data Mining: Concepts and
Techniques. Simon Fraser University, Morgan
Kaufmann Publishers.
Han J and Kamber M. Data Mining: Concept and
Techniques. New York: Morgan Kaufmann Publishers;
2006.
Hassan, "Biomedical Signal Processing and Control
Computeraided obstructive sleep apnea detection using
inverse Gaussian normal parameters and adaptive
boosting," Biomed. Signal Process. Control, vol. 29, pp.
22-30, 2016.
Huang, Z., & Shyu, M.-ling. (2010) .k-NN LS-SVM Based
Framework for Long-Term Time Series Prediction.
System, 4-6, 69-74.
Jingbo Yuan and Ding Shunlin,
"Research And
Improvement On Association Rule Algorithm Base On
FP-Growth," 2012.
Kalles, D., & Pierrakeas, C. 2006, Analyzing Student
Performance in Distance Learning with Genetic
Algorithms and decsion Trees, Hellenic Open
University.
Kusumadewi Sri, Classification of Nutritional Status Using
Naïve Bayesian Classification, COMMIT, vol.3 no. 1,
pp. 6-11, 2009.
Kraipeerapun, P., & Fung, CC (2008) Performance of
Interval Neutrosophic. Comparing Sets and Neural
Networks with Binary Support Vector Machines for
Classification Problems. Learning, 34-37.
Larose, Daniel T. 2005. Discovering Knowledge in Data:
An Introduction to Data Mining. John Willey & Sons,
Inc.
Lassibille, G., Gomez, LN (2007). Why Do Higher
EducationStudents Drop Out? Evidencefrom
Spain.Education Economics.Vol 16 (1) .89-105.
Luan, J. (2002). Data Mining and Its Applications in Higher
Education. New Directions for Institutional
Research.Vol 133.17-36.
ICEST 2018 - 3rd International Conference of Computer, Environment, Agriculture, Social Science, Health Science, Engineering and
Technology
200
Maimon, O. and Last, M. 2000. Knowledge Discovery and
Data Mining, The Info-FuzzyNetwork (IFN)
Methodology. Dordrecht: Kluwer Academic.
Mark A. Hall and Geoffrey Holmes.Benchmarking
Attribute Discrete Class Selection Techniques for Data
Mining. IEEE Transactions on Knowledge and Data
Engineering, 15, no. 3, June 2003
Mujib Ridwan, Suryono Hadi, M.Sarosa, Application of
Data Mining for Evaluation of Student Academic
Performance Algorithm Naïve Bayes, EECCIS, vol 7
no. 1, pp. 59-64, 2013.
Nugroho, Satrio, A. 2003.Support Vector Machine Theory
and Applications in Bioinformatics.
Pintowati, W., & Otok, W. (2012) .Pemodelan Poverty in
East Java with Multivariate Adaptive Approach, 1 (1),
2-7.
Sartono, B. et al. 2003. Analysis of Multiple Variables.
Bogor Agricultural University, Bogor.
Sri Wahyunigsih, D. (2012). Modeling Farmer On Crops
With Transfer Functions and Multivariete Pendekaran
Regresion Adaptive Multi Variate. ITS, 1-11.
Turban, R., Rainer, R. and Potter, R 2005.'Introduction to
Information Technology '. USA: John Wiley & Sons,
Inc.
Tan, P, et al. 2006. Introduction to Data Mining. Boston:
Pearsion Education
Thomas E. 2004. Data Mining: Definition and Decision
Tree Examples, e-book
Wang, T. Lu, and Z. Xiong, "Adaptive Boosting for Image
Denoising: Beyond Low-Rank Representation and
Sparse Coding," 23rd International Conference on
Pattern Recognition (ICPR), 2016.
Xia, L., & Pan, H. (2010). Inferential Estimation of
Polypropylene Melt Index Stacked Using Neural
Networks Based on Absolute Error Criteria. Control,
216-218.
Dinakaran, S. Dr. PRJ Thangaiah, "Role of Attribute
Selection in Classification Algorithms", International
Journal of Scientific & Engineering Research, vol. 4,
issue 6, pp. 67-71. June, 2013.
Ramaswami.M, Bhaskaran.R, "A Study on Feature
Selection in Educational Data Mining Techniques",
Journal of Computing, vol. 1, Issue 1, pp. 7-11.
December 2009.
Hybrid Information Gain Method and Bagging in Data Classification using Support Vector Machine
201