International Journal of Computer Applications (0975 8887)

Volume 136 No.1, February 2016

Literature Review on Feature Selection Methods for

High-Dimensional Data
D. Asir Antony Gnana
Singh

S. Appavu alias
Balamurugan

Anna University, BIT Campus,

Tiruchirappalli, India.

K.L.N College of Information

Technology, Sivagangai, India.

ABSTRACT
Feature selection plays a significant role in improving the
performance of the machine learning algorithms in terms of
reducing the time to build the learning model and increasing
the accuracy in the learning process. Therefore, the
researchers pay more attention on the feature selection to
enhance the performance of the machine learning algorithms.
Identifying the suitable feature selection method is very
essential for a given machine learning task with highdimensional data. Hence, it is required to conduct the study on
the various feature selection methods for the research
community especially dedicated to develop the suitable
feature selection method for enhancing the performance of the
machine learning tasks on high-dimensional data. In order to
fulfill this objective, this paper devotes the complete literature
review on the various feature selection methods for highdimensional data.

General Terms
Literature review on feature selection methods, study on
feature selection, wrapper-based feature selection, embeddedbased feature selection, hybrid feature selection, filter-based
feature selection, feature subset-based feature selection,
feature ranking-based feature selection, attribute selection,
dimensionality reduction, variable selection, survey on feature
selection, feature selection for high-dimensional data,
introduction to variable and feature selection, feature selection
for classification.

Keywords
Introduction to variable and feature selection, information
gain-based feature selection, gain ratio-based feature
selection, symmetric uncertainty-based feature selection,
subset-based feature selection, ranking-based feature
selection, wrapper-based feature selection, embedded-based
feature selection, filter-based feature selection, hybrid feature
selection, selecting feature from high-dimensional data.

1. INTRODUCTION
In the digital era, handling the massive data is a challenging
task among the researchers since the data are accumulated
through various data acquisition techniques, methods, and
devices. These accumulated massive raw data degrade the
performance of the machine learning algorithms in terms of
causing overfitting, spending more time to develop the
machine learning modes and degrading their accuracy since
the raw data are noisy in nature and have more number of
features known as high-dimensional data. In general, the highdimensional data contains irrelevant and the redundant
features. The irrelevant features cannot involve in the learning
process and the redundant features contain same information
hence thy miss lead the learning process. Therefore, these
issues can be tackled by the feature selection. The feature

E. Jebamalar Leavline
Anna University, BIT Campus,
Tiruchirappalli, India.

selection is a process of removing the redundant and the

irrelevant features from a dataset to improve the performance
of the machine learning algorithms. The feature selection is
also known as variable selection or attribute selection. The
features are also known as variables or attributes. The
machine learning algorithms can be roughly classified into
two categories one is supervised learning algorithm and
another one is unsupervised learning algorithm. The
supervised learning algorithms learn the labeled data and
construct learning models that are known as classifiers. The
classifiers are employed for classification or prediction to
identify or predict the class-label of the unlabeled data. The
unsupervised learning algorithms lean the unlabeled data and
construct the learning models that known as clustering
models. The clustering models are employed to cluster or
categorize the given data for predicting or identifying their
group or cluster. Mostly, the feature selections are employed
for the supervised learning algorithms since they suffered by
the high-dimensional space. Therefore, this paper presents a
complete literature review on various feature selection
methods for high-dimensional data.
The rest of this paper is organized as follows: Section 2
describe the feature selection process. In Section 3, survey on
feature selection is conducted. Section 4 summarizes the
survey on feature section. Section 5 concludes this paper.

2. FEATURE SELECTION
Feature selection is a process of removing the irrelevant and
redundant features from a dataset in order to improve the
performance of the machine learning algorithms in terms of
accuracy and time to build the model. The process of feature
selection is classified into two categories namely feature
subset selection and feature ranking methods based on how
the features are combined for evaluation. The feature subset
selection approach generates the possible number of
combinations of the feature subsets using any one of the
searching strategies such as a greedy forward selection,
greedy backward elimination, etc. to evaluate the individual
feature subset with a feature selection metric such as
correlation, consistency, etc. In this method, space and the
computational complexity involved are more due to the subset
generation and evaluation [2].
In feature ranking method, each feature is ranked by a
selection metric such as information gain, symmetric
uncertainty, gain ratio, etc. and the top ranked features are
selected as relevant features by a pre-defined threshold value.
This approach is computationally cheaper and space
complexity is less compared to subset approach. However, it
does not deal with redundant features.
Further, the process of feature selection is classified into four
categories namely wrapper, embedded, filter, and hybrid

International Journal of Computer Applications (0975 8887)

Volume 136 No.1, February 2016
methods based on how the supervised learning algorithm is
employed in the feature selection process.

3.1 Feature Selection Based on Combining

the Features for Evaluation
This section reviews various methods of feature selection
based on how the features are combined for evaluation in
order to select the significant features from a dataset. They are
classified into feature subset-based and feature ranking-based
methods.

3.1.1 Feature subset-based methods

Figure 1 Feature selections with wrapper approach

Wrapper approach incorporates the supervised learning
algorithm for validating the generated feature subsets using
any one of the searching strategies as shown in Figure 1. It
yields high classification accuracy only for the particular
learning algorithm adopted. Hence, it does not possess a high
generality and the computational complexity is higher than
embedded and filter methods.
The embedded approach uses a part of supervised learning
algorithm for feature selection process and it produces better
accuracy only for the learning algorithm used in the selection
process. Hence, it does not have a high generality and it is
computationally expensive than the filter and lesser than the
wrapper method.

Figure 2 Feature selection with filter approach

The filter approach (Figure 2) selects the features without the
influence of any supervised learning algorithm. Hence, it
works for any classification algorithm and achieves more
generality with less computational complexity than the
wrapper and embedded methods. Therefore, it is suitable for
high-dimensional space. The combination of wrapper and
filter approach is known as hybrid method [1].

3. SURVEY ON FEATURE SELECTION

As the feature selection is employed in various machine
learning applications, it has remarkable literature records
made by the research community. Feature selection is a
preprocessing technique to select the significant features from
a dataset by removing the irrelevant and redundant features
for improving the performance of the machine learning
algorithms. The feature selection process can be categorized
into various methods based on how the features are combined
for evaluation in the feature selection process and how the
supervised learning algorithm is used to evaluate the features
in the features selection process. This paper reviews the
literature on various features selection methods and explores
their merits and demerits.

In the feature subset-based method, the features are combined

as possible combinations of feature subsets using any one of
the searching strategies. Then, the feature subsets are
evaluated using any one of the statistical measures or the
supervised learning algorithms to observe the significance of
each subset and the most significant subset is selected as the
significant feature subset for a given dataset. If the subset is
evaluated using the supervised learning algorithm, then this
method is known as wrapper method.
The best example for the feature subset-based method is
correlation-based feature subset selection (CRFS) developed
by Hall [3]. In this approach, two correlation measures are
considered; one is feature-class correlation and another one is
feature-feature correlation. Initially, N numbers of features are
combined as possible combinations of feature subsets using
heuristic-based best-first search, then each subset is evaluated
with the two correlation measures as mentioned above. The
subset that has lesser feature-feature correlation and higher
feature-class correlation compared to other feature subsets is
considered as the selected significant feature subset for the
classification task. Liu & Setiono [4] proposed a feature
subset-based feature selection method namely consistencybased feature subset selection (COFS). This method uses the
class consistency as an evaluation metric in order to select the
significant feature subset from the given dataset. These
methods are the filter-based methods since they do not use the
supervised learning algorithm to validate the subsets and they
use the statistical measure for evaluating the feature subsets.
In general, the exhaustive or complete search has to generate
2N number of subsets to produce the maximum number of
possible combinations of feature subsets from the N number
of features for evaluation. Therefore, this exhaustive
searching strategy is computationally quite expensive hence
the heuristic searching strategies such as simulated annealing
(SA), tabu searching (TS), ant colony optimization (ACO),
genetic algorithm (GA), particle swarm optimization (PSO),
etc. [5] are used by some of the researchers to get the optimal
solution by generating less number of feature subsets for
evaluation. In the heuristic searching, the heuristic function
obtains the prior knowledge to guide the search process to
generate the subsets and these subsets are evaluated using
supervised machine learning algorithm. These factors make
the feature subset-based methods computationally expensive
and also these methods seem to be the wrapper approach.
Some researchers used the simulated annealing search for
generating the feature subset for evaluations. For example,
Lin et al used the simulated annealing search to generate the
feature subsets and evaluated them by supervised learning
algorithm namely back-propagation network (BPN) to choose
the better feature subset [6]. Meiri & Zahavi used simulated
annealing-based feature selection for marketing application
[7]. In several feature selection methods, the tabu search is
used for subset generation such as Zhang & Sun developed a
tabu search-based feature selection. In this method, the
subsets generated by tabu search are evaluated using the
classification error criteria to find the better feature subset [8].

10

International Journal of Computer Applications (0975 8887)

Volume 136 No.1, February 2016
Tahir et al formed the feature subsets using tabu search then
these subsets are evaluated using K-nearest neighbor classifier
(kNN) with the classification error as evaluation criteria to
obtain the significant feature subset [9].
A number of feature selection processes used the ant colony
optimization as the searching criteria for subset generation.
Aghdam et al employed the ant colony optimization search to
form the feature subsets and they are validated by the nearest
neighbor classifier for text classification application [10].
Kanan & Faez proposed a feature selection method using ant
colony optimization for face recognition system. In this
approach, the nearest neighbor classifier is adopted for
evaluating the generated subset using ant colony optimizationbased learning [11]. Sivagaminathan & Ramakrishnan
developed an ant colony optimization-based feature selection
with artificial neural networks (ANN) for medical diagnosis
system. In this method, the generated feature subsets are
validated using ANN [12]. Sreeja & Sankar presented an ant
colony optimization-based feature selection with instancebased pattern matching-based classification (PMC) [13].
In certain feature selection research works, the genetic
algorithm is adopted to generate the feature subsets for
evaluation and the supervised machine learning algorithm is
used to evaluate the generated subsets. Welikala et al
presented a feature selection using genetic algorithm with
support vector machine (SVM) for mining the medical dataset
[14]. Erguzel et al used the genetic algorithm and artificial
neural network for electroencephalogram (EEG) signal
classification [15]. Oreski & Oreski proposed a feature
selection method based on genetic algorithm with neural
networks for credit risk assessment [16]. Li et al developed a
genetic algorithm with support vector machine for hyperspectral image classification [17]. Das et al formulated a
genetic algorithm with support vector machine-based feature
selection for handwritten digit recognition application [18].
Wang et al applied the genetic algorithm for subset generation
with support vector machine in feature selection process for
data classification applications [19].
In the literature, some researches employed the particle swam
optimization to generate the feature subsets and to validate
them by supervised machine learning algorithm to identify the
significant feature subset. Xue et al designed a particle swarm
optimization (PSO)-based feature selection for classification.
In this method, the feature subsets generated by PSO are
evaluated using supervised learning algorithm [20]. Chen et al
presented a feature selection method using particle swarm
optimization search for sleep disorder diagnosis system [21].
Yang et al developed a particle swarm optimization-based
feature selection for land cover classification [22].
From the subset-based feature selection literature, it is
observed that the exhaustive or complete search leads to high
computational complexity as it generates 2N number of
subsets from N number of features for evaluation. This
searching strategy cannot be a better choice for highdimensional space. The heuristic search methods also lead to
more computational complexity, because they need prior
knowledge and each generated subset needs to develop a
classification model for evaluating them to obtain the optimal
feature subset in an iterative manner, hence these searching
strategies are not suitable for high-dimensional space.
However, these heuristic search methods follow a wrapperbased approach. Therefore, these methods are computationally
expensive and they can only produce higher classification
accuracy for the specific classification algorithm used to
validate the subset, so they cannot achieve high generality.

3.1.2

Feature ranking-based methods

In the feature-ranking based approach, each feature of a

dataset is weighted based on any one of the statistical or
information-theoretic measures and the features are ranked
based on their weight. Then the higher ranked features are
selected as the significant features using a predefined
threshold that determines the number of features to be
selected from a dataset. The best example for the feature
ranking-based method is chi-square-based feature selection
(CQFS). In this method, Liu & Setiono used the chi-square
statistic measure to weight the features in order to rank them
for selecting the significant features [24]. In the similar way,
the information-theoretic measures such as information gain,
symmetric uncertainty, gain ratio, etc. are employed to weight
the individual feature and rank them for selection.
Further, it is observed that the feature ranking-based methods
use the statistical measures or information-theoretic measures
to weight the individual feature only by observing the
relevancy between the individual feature and the target-class.
Hence, these methods take less runtime but fail to remove the
redundant features [2]. The feature ranking-based methods
follow a filter-based approach since these methods do not
involve the supervised learning algorithm to evaluate the
significance of the features. Consequently, these methods are
independent of the supervised learning algorithm hence they
achieve more generality and less computational complexity.
Thus, the feature ranking-based methods can be a good choice
for selecting the significant features from the highdimensional space with suitable redundancy analysis
mechanism.

3.2 Feature Selection Based on the

Supervised Learning Algorithm Used
This section reviews various methods of feature selection
based on the machine learning algorithm used. They are
categorized as wrapper, embedded, filter, and hybrid methods.

3.2.1

Wrapper-based methods

Wrapper-based approach generates the feature subsets using

any one of the searching techniques and evaluates these
subsets using the supervised learning algorithm in terms of
classification error or accuracy [25]. The wrapper method
seems to be a brute force method. This approach is
illustrated in Figure 1. Kohavi & John developed a wrapperbased feature selection method for selecting the significant
features from the dataset [26]. This method consists of search
engine for subset generation and classification algorithm to
evaluate the subset. Further, they compare the performance of
this method in terms of classification accuracy with hillclimbing and best-first searching strategies using decision tree
and nave Bayes classifiers. However, they observed that
wrapper method has the problems such as searching overhead,
overfitting, and increased runtime.
In wrapper approach, the searching is an overhead since the
searching technique does not have the domain knowledge. In
order to overcome the searching time overhead, Inza et al
used estimation of Bayesian network algorithm for feature
subset selection using naive Bayes and ID3 (Iterative
Dichotomiser 3) [27]. In general, the searching method may
lead to increase in computational complexity, since the
training data is split for evaluation. In order to overcome this
issue, Grimaldi et al used an aggregation principle with
sequential search [28]. Dy & Brodley developed a wrapperbased approach for unsupervised learning using order
identification (recognizing the number of clusters in the data)
with the expectation maximization (EM) clustering algorithm

11

International Journal of Computer Applications (0975 8887)

Volume 136 No.1, February 2016
using maximum likelihood (ML) criterion [29]. Aha &
Bankert presented a wrapper-based method with beam search
and IB1 classifier [30]. Also, they compared its performance
with the well known sequential search algorithms for feature
selection such as forward sequential selection (FSS) and
backward sequential selection (BSS). They observed that the
beam search outperforms the FSS and BSS.
The Maldonado & Weber developed a wrapper approachbased feature selection by combining support vector machine
(SVM) with kernel functions. This method uses the sequential
backward selection for feature subset generation and these
subsets are validated in terms of classification error to identify
the best subset [31]. In order to minimize the searching
overhead, Gtlein et al used the search algorithm namely
ORDERED-FS that orders the features in terms of
resubstitution error to identify their irrelevancy [32]. Kabir et
al developed a wrapper-based constructive approach for
feature selection (CAFS) using neural network (NN). In this
method, the correlation measure is used to remove the
redundancy in the searching strategy for improving the
performance of NN [33]. Stein et al proposed an ant colony
optimization-based feature selection with wrapper model. In
this approach, the ant colony optimization is used as a
searching method in order to reduce the searching overhead
such as blind search or forward selection or backward
elimination searching methods [34]. Furthermore, to minimize
the searching overhead, Zhuo et al presented a wrapper-based
feature selection using genetic algorithm with support vector
machine for classifying the hyper-spectral images [35].
In the wrapper approach, overfitting can be overcome by postpruning, jitter, and early stopping methods. Post-pruning is
carried out while developing the decision tree [36]. In jitter
method, the noisy data that make the learning process more
difficult are eliminated in order to fit the training data thereby
the overfitting is eliminated [37]. In early stopping method,
overfitting is eliminated using neural network by stopping the
training process when performance on a validation set starts to
deteriorate [38] [39]. The researchers have tried to reduce the
overfitting by early stopping method using genetic algorithmbased searching with early stopping (GAWES) [40].
Further, it is observed that the wrapper-based methods are
suffered by the searching overhead, overfitting [41] and have
more computational complexity with less generality since
they use the supervised learning algorithm for evaluating the
generated subsets by the searching method. Therefore, these
methods are not suitable choice for the high-dimensional
space.

3.2.2

Embedded-based methods

The embedded-based methods use a part of the learning

process of the supervised learning algorithm for feature
selection. Embedded-based methods reduce the computational
cost than the wrapper method [42]. This embedded method
can be roughly categorized into three namely pruning method,
built-in mechanism, and regularization models. In the
pruning-based method, initially all the features are taken into
the training process for building the classification model and
the features which have less correlation coefficient value are
removed recursively using the support vector machine (SVM)
[43]. In the built-in mechanism-based feature selection
method, a part of the training phase of the C4.5 [36] and ID3
[44] supervised learning algorithms are used to select the
features. In the regularization method, fitting errors are
minimized using the objective functions and the features with
near zero regression coefficients are eliminated [45] [46].

Neumann et al developed an embedded-based feature

selection method for selecting the significant features from
synthetic and real world datasets. In their approach, linear and
non linear SVMs are employed in the selection process using
the deference of convex functions algorithm (DCA) [47].
Xiao et al proposed an embedded-based method to select the
significant features from audio signals for emotion
classification. This method was implemented based on the
principle of evidence theory with mass function and the
identified most relevant features are added incrementally for
classification [48]. Maldonado et al developed an embedded
method to select the significant features from imbalanced data
for classification with several objective functions [49].
Further, it is observed that the embedded methods are
computationally efficient than the wrapper methods and
computationally costlier than the filter methods hence they
cannot be suitable choice for high-dimensional space and they
have poor generality since the embedded methods use the
supervised learning algorithm.

3.2.3

Filter-based methods

The filter-based approaches are independent of the supervised

learning algorithm therefore offer more generality and they
are computationally cheaper than the wrapper and embedded
approaches. For processing the high-dimensional data, the
filter methods are suitable rather than the wrapper and
embedded methods.
Generally, the process of feature selection aimed at choosing
the relevant features. The best example is Relief [50] that was
developed with the distance-based metric function that
weights each feature based on their relevancy (correlation)
with the target-class. However, Relief is ineffective as it can
handle only the two-class problems and also does not deal
with redundant features. The modified version of the Relief
known as ReliefF [51] can handle the multi-class problems
and deal with incomplete and noisy datasets too. However, it
fails to remove the redundant features. Holte developed a rulebased attribute selection known as OneR which forms one
rule for each feature and selects the rule with the smallest
error [52]. Yang & Moody proposed a joint mutual
information-based approach (JMI) for classification. It
calculates the joint mutual information between the individual
feature and the target-class to identify the relevant features,
and a heuristic search is adopted for optimization when the
number of features is more. The features containing similar
information and lesser relevancy to the target-class are treated
as redundant features that are to be eliminated [53].
Peng et al proposed a mutual information-based maxrelevancy min-redundancy (MRMR) feature selection. To
identify the feature relevancy, the mutual information is
computed between the individual feature and target-class, and
to identify the redundant feature, the mutually exclusive
condition is applied [54]. Battiti developed a mutual
information-based feature selection method (MIFS). In this
method, mutual information measure is used to determine the
relevancy between the individual feature and the target-class.
The features having similar information are considered as
redundant features that are to be removed [55]. Fleuret
presented a feature selection scheme namely conditional
mutual info maximization (CMIM) that recursively chooses
the features that have maximum mutual information with the
target-class for classification [56].
Meyer & Bontempi proposed a filter-based approach that uses
double input symmetrical relevance (DISR) metric for feature
selection. This approach returns the selected features that

12

International Journal of Computer Applications (0975 8887)

Volume 136 No.1, February 2016
contain more information about the target-class than the
information about other features [57]. Lin & Tang introduced
an information theory-based conditional infomax feature
extraction (CIFE) algorithm to measure the class-relevancy
and redundancy for feature selection [58]. Brown et al used
the conditional redundancy (CondRed) metric for selecting
the significant features from the dataset [59].
In the recent past, the clustering technique is also adopted in
feature selection. Song et al developed a feature selection
framework and adopted the graph-based clustering technique
to identify the similarity among the features for removing the
redundant features [60]. Dhillon et al developed a feature
selection algorithm based on information theory for text
classification. In this approach, the hierarchical clustering is
used to cluster the features or terms of documents for
identifying their dependencies [61]. Li et al incorporated the
clustering algorithm with the chi-square statistical measure to
select the features from statistical data [62]. Cai et al
developed a spectral clustering-based feature selection
(MCFS) for selecting the significant features from the datasets
[63]. Chow & Huang employed the supervised clustering
technique and mutual information for identifying the salient
features from synthetic and real world datasets [64]. Mitra et
al presented a feature selection approach by adopting the
graph-based clustering approach to identify the similarity
among the features for redundancy analysis [65]. Sotoca &
Pla developed a feature selection method for classification
based on feature similarity with hierarchical clustering [66].
Further, it is observed that the filter-based methods are
computationally better than the wrapper [67] and embedded
[68] methods. Therefore, the filter-based methods can be a
suitable choice for high-dimensional space. The filter-based
methods achieve high generality since they do not use the
supervised learning algorithm.

3.2.3

Hybrid Methods

The hybrid methods are the combination of filter and

wrapper-based approaches [69]. In general, processing the
high-dimensional data is a difficult task with the wrapper
method therefore the authors Bermejo et al developed a
hybrid feature selection method known as filter-wrapper
approach. In this approach, they used a statistical measure to
rank the features based on their relevancy then the higher
ranked features are given to the wrapper method so that the
number of evaluations required for the wrapper method is
linear. Thus, the computational complexity is reduced using
hybrid method for medical data classification [70]. Ruiz et al
developed a gene (feature) selection algorithm for selecting
the significant genes for the medical diagnosis system. They
used a statistical ranking approach to filter the features from
high-dimensional space and the filtered features are fed into
the wrapper approach. This combination of the filter and
wrapper approach was used to distinguish the significant
genes causing cancer disease in the diagnosis process [71].
Xie et al developed a hybrid approach for diagnosing the
erythemato-squamous diseases. In this approach, F-score
measure is used to rank the features to identify the relevant
features (filter approach). The significant features are selected
from the ranked features with the sequential forward floating
search (SFFS) and SVM (wrapper method) [72]. Kannan &
Faez presented a hybrid feature selection framework. In this
approach, ant colony optimization (ACO)-based local search
(LS) is used with the symmetric uncertainty measure to rank
the features [73]. Xie et al designed a hybrid approach with
F-score to identify the relevant attributes from a disease

dataset. For feature subset generation from the relevant

features, the searching strategies such as sequential backward
floating search (SBFS), extended sequential forward search
(ESFS), and sequential forward floating search (SFFS) are
also employed [74]. Naseriparsa et al proposed a hybrid
method using information gain and genetic algorithm-based
searching method combined with a supervised learning
algorithm [75]. Huda et al developed a hybrid feature
selection method by combining the mutual information (MI)
and artificial neural network (ANN) [76]. Gunal presented a
hybrid feature selection method by combining filter and
wrapper method for text classification. In this method,
information gain measure is used for ranking the significant
features and the genetic algorithm is used as the searching
strategy with support vector machine [77].
DAlessandro et al proposed a hybrid approach for epileptic
seizure prediction, in which ranking with genetic algorithmbased wrapper approach was implemented [78]. Yang et al
developed a hybrid method for classifying the micro array
data. In this method, the information gain and correlation
metric are used for filter method and an improved binary
particle swarm optimization (BPSO) method is used with the
supervised learning algorithm as the wrapper method to
improve the performance of the classification algorithm. The
performance of this method is evaluated using kNN and SVM
classifiers [79]. To avoid the computational cost of the
wrapper method, Bermejo presented a hybrid method by
combining the filter and wrapper methods. In this method, the
GRASP meta-heuristic based on stochastic algorithm is used
as filter method for reducing the wrapper computation [80].
Foithong et al also designed a hybrid feature selection method
by combining the filter and the wrapper methods. In this
method, the mutual information criterion is used for filtering
the relevant features and the supervised learning algorithm is
adopted as the wrapper method for evaluating features
obtained from the filter method [81].
Further, it is observed that the hybrid methods are
computationally intensive than the filter methods since they
combine the wrapper and filter methods and have less
generality compared to the filter methods since they use the
supervised learning algorithm in feature selection process.
These hybrid methods take more computational time than the
filter-based methods.

4. SUMMARY
This section summarizes the feature selection methods that are
categorized based on how the features are combined in the
selection process namely feature subset-based and feature
ranking-based and based on how the supervised learning
algorithm used namely wrapper, embedded, hybrid, and filter.
The subset-based methods generate the feature subsets using
any one of the searching strategies for evaluation. The
exhaustive or complete search is used to generate the subset
that leads to high computational complexity since maximum
2N number of possible combination of the subsets to be
generated from the N number of features to evaluate them.
This is a brute force method so this is not suitable for highdimensional space. The heuristic search such as SA, TS,
ACO, GA, and PSO are employed to reduce the number of
feature subset generation for evaluation using the heuristic
function. The subset-based feature selection methods using
the heuristic search lead to more computational complexity,
because they need prior knowledge and each generated subset
need to develop a classification model to evaluate them.

13

International Journal of Computer Applications (0975 8887)

Volume 136 No.1, February 2016
However, this heuristic search methods follow a wrapperbased approach therefore these methods are computationally
expensive and they can only produce higher classification
accuracy for the specific classification algorithm used to
obtain the fitness or heuristic function. Therefore, these
methods cannot achieve high generality. The ranking-based
methods take less computation time and achieve high
generality since they do not use the supervised learning
algorithm. They cannot remove the redundant features since
they only compute the correlation or similarity between the
individual feature and the target-class. Therefore, they can be
a suitable choice for high dimensional space with a suitable
redundancy analysis mechanism.
The wrapper, embedded, and hybrid methods are
computationally inefficient than the filter approach. In
addition, they do not have high generality since they use the
supervised learning algorithm in feature selection process.
Therefore, the filter methods are the best choice for the highdimensional data. Further, the filter methods are preferred
because they can perform better with any classification
algorithm since they possess better generality and require less
computational complexity. The ranking-based approaches are
better than the feature subset-based methods since the subsetbased methods require more space and computational
complexity. Therefore, the ranking-based methods are the best
choice for selecting the relevant features from the highdimensional space.
In the feature selection literature, some researchers have
succeeded in effectively removing the irrelevant features, but
failed to handle the redundant features. On the other hand,
some other researchers dealt with removing the irrelevant
features and redundant features. Furthermore, the state-of-theart feature selection methods reported in literature use the
rule-based metric and nearest neighbor principles. Both of the
methods eliminate the irrelevant features, but fail to treat the
redundant features. Some of the methods use the informationtheoretic-based metric to calculate the relevancy between the
feature and the target-class for relevancy analysis and to
calculate the independency among features for redundancy
analysis.
In most of the information-theoretic-based approaches, the
same metric is used for both redundancy and irrelevancy
analysis. Some of these approaches perform pair wise analysis
to identify the independency among the features for
redundancy analysis, resulting in increased time complexity.
They do not have special mechanism for treating the
redundant features, yet they do moderate redundancy analysis.
Most of the clustering-based approaches use the hierarchical
clustering for feature selection and they deal with specific
types of datasets. However, the hierarchical clustering is
expensive for high-dimensional datasets and less effective in
high-dimensional space due to the dimensionality
phenomenon. Hence, simple, scalable, and faster K-means
clustering algorithm can be used [82] for relevancy analysis in
feature selection.

5. CONCLUSION
This paper analyzed several feature selection methods that are
proposed by various researchers. From the earlier research
works, it is observed that the feature ranking-based methods
are better than the subset-based methods in terms of memory
space and computational complexity and the ranking-based
methods do not reduce the redundancy. Further, the wrapper,
embedded, and hybrid methods are computationally
inefficient than the filter method and they have poor

generality. Therefore, the feature selection can be developed

for high-dimensional data using the filter approach with
ranking method for selecting the significant features from the
high-dimensional space. In addition, to overcome the
limitations of the ranking method the redundancy analysis
mechanism can be adopted with a suitable clustering
approach.

6. REFERENCES
[1] Saeys, Y, Inza, I & Larraaga, P 2007, A review of
feature selection techniques in bioinformatics.
Bioinformatics, vol. 23, no. 19, pp.2507-2517
[2] Boln-Canedo, V, Snchez-Maroo, N & AlonsoBetanzos, A, 2013, A review of feature selection
methods on synthetic data, Knowledge and information
systems, vol. 34, no.3, pp.483-519.
[3] Hall, MA 1999, Correlation-based feature selection for
machine learning, Ph.D. thesis, The University of
Waikato, NewZealand.
[4] Liu, H & Setiono, R 1996, A probabilistic approach to
feature selection-a filter solution, Proceedings of
Eighteenth International Conference on Machine
Learning, Italy, pp. 319-327.
[5] Lisnianski, A, Frenkel, I & Ding, Y, 2010, Multi-state
system reliability analysis
and optimization for
engineers and industrial managers, Springer, New York.
[6] Lin, S.W, Tseng, TY, Chou, SY & Chen, SC 2008, A
simulated-annealing-based approach for simultaneous
parameter optimization and feature selection of backpropagation
networks,
Expert
Systems
with
Applications, vol. 34, no.2, pp.1491-1499.
[7] Meiri, R & Zahavi, J 2006, Using simulated annealing
to optimize the feature selection problem in marketing
applications, European Journal of Operational Research,
vol.171, no.3, pp.842-858.
[8] Zhang, H & Sun, G 2002, Feature selection using tabu
search method, Pattern recognition, vol. 35, no.3,
pp.701-711.
[9] Tahir, MA, Bouridane, A & Kurugollu, F 2007,
Simultaneous feature selection and feature weighting
using Hybrid Tabu Search/K-nearest neighbor classifier
Pattern Recognition Letters, vol. 28, no.4, pp.438-446.
[10] Aghdam, MH, Ghasem-Aghaee, N & Basiri, ME 2009,
Text feature selection using ant colony optimization,
Expert systems with applications, vol. 36, no.3, pp.68436853.
[11] Kanan, HR & Faez, K 2008, An improved feature
selection method based on ant colony optimization
(ACO) evaluated on face recognition system, Applied
Mathematics and Computation, vol. 205, no.2, pp.716725.
[12] Sivagaminathan, RK & Ramakrishnan, S 2007, A
hybrid approach for feature subset selection using neural
networks and ant colony optimization, Expert systems
with applications, vol. 33, no.1, pp.49-60.
[13] Sreeja, NK & Sankar, A 2015, Pattern Matching based
Classification using Ant Colony Optimization based
Feature Selection, Applied Soft Computing, vol. 31,
pp.91-102.

14

International Journal of Computer Applications (0975 8887)

Volume 136 No.1, February 2016
[14] Welikala, R.A, Fraz, MM, Dehmeshki, J, Hoppe, A, Tah,
V, Mann, S, Williamson, TH & Barman, SA 2015,
Genetic algorithm based feature selection combined
with dual classification for the automated detection of
proliferative diabetic retinopathy, Computerized
Medical Imaging and Graphics, vol. 43, pp.64-77.
[15] Erguzel, TT, Ozekes, S, Tan, O & Gultekin, S 2015,
Feature
Selection
and
Classification
of
Electroencephalographic Signals an Artificial Neural
Network and Genetic Algorithm Based Approach,
Clinical EEG and Neuroscience, vol. 46, no.4, pp.321326.
[16] Oreski, S & Oreski, G 2014, Genetic algorithm-based
heuristic for feature selection in credit risk assessment,
Expert systems with applications, vol. 41, no.4, pp.20522064.
[17] Li, S, Wu, H, Wan, D & Zhu, J, 2011, An effective
feature selection method for hyperspectral image
classification based on genetic algorithm and support
vector machine, Knowledge-Based Systems, vol. 24,
no.1, pp.40-48.
[18] Das, N, Sarkar, R, Basu, S, Kundu, M, Nasipuri, M &
Basu, DK 2012, A genetic algorithm based region
sampling for selection of local features in handwritten
digit recognition application, Applied Soft Computing,
vol.12, no.5, pp.1592-1606.

[26] Kohavi, R & John, GH 1997, Wrappers for feature

subset selection, Artificial intelligence, vol. 97, no.1,
pp.273-324.
[27] Inza, I, Larraaga, P, Etxeberria, R & Sierra, B 2000,
Feature subset selection by Bayesian network-based
optimization, Artificial intelligence, vol. 123, no. 1,
pp.157-184.
[28] Grimaldi, M, Cunningham, P & Kokaram, A 2003, An
evaluation of alternative feature selection strategies and
ensemble techniques for classifying music, Proceedings
of
Fourteenth European Conference on Machine
Learning and the Seventh European Conference on
Principles and Practice of Knowledge Discovery in
Databases, Dubrovnik, Croatia
[29] Dy, JG & Brodley, CE 2000, Feature subset selection
and order identification for unsupervised learning,
proceedings In Proceedings of the Seventeenth
International Conference on Machine Learning, p. 247
254.
[30] Aha, DW & Bankert, RL 1996, A Comparative
Evaluation of Sequential Feature Selection Algorithms,
Springer, New York.
[31] Maldonado, S & Weber, R 2009, A wrapper method for
feature selection using support vector machines,
Information Sciences, 179(13), pp.2208-2217.

[19] Wang, Y, Chen, X, Jiang, W, Li, L, Li, W, Yang, L,

Liao, M, Lian, B, Lv, Y, Wang, S & Wang, S 2011,
Predicting human microRNA precursors based on an
optimized feature subset generated by GASVM,
Genomics, vol. 98, no.2, pp.73-78.

[32] Gtlein, M, Frank, E, Hall, M & Karwath, A 2009,

March. Large-scale attribute selection using wrappers,
Proceeding of IEEE Symposium on Computational
Intelligence and Data Mining, Nashville, TN, USA, pp.
332-339.

[20] Xue, B, Zhang, M & Browne, WN 2013, Particle swarm

optimization for feature selection in classification: A
multi-objective approach, IEEE Transactions on
Cybernetics, vol. 43, no.6, pp.1656-1671.

[33] Kabir, MM, Islam, MM & Murase, K 2010, A new

wrapper feature selection approach using neural
network, Neurocomputing, vol. 73, no. 16, pp.32733283.

[21] Chen, LF, Su, CT, Chen, KH & Wang, PC 2012,

Particle swarm optimization for feature selection with
application in obstructive sleep apnea diagnosis, Neural
Computing and Applications, vol. 2, no. 8, pp.20872096.

[34] Stein, G, Chen, B, Wu, AS & Hua, KA 2005, March.

Decision tree classifier for network intrusion detection
with GA-based feature selection Proceedings of the
forty-third ACM Annual Southeast regional conference,
Kennesaw, GA, USA, vol. 2, pp. 136-141.

[22] Yang, H, Du, Q & Chen, G 2012, Particle swarm

optimization-based
hyperspectral
dimensionality
reduction for urban land cover classification, IEEE
Journal of Selected Topics in Applied Earth
Observations and Remote Sensing, vol. 5 no.2, pp.544554.

[35] Zhuo, L, Zheng, J, Li, X, Wang, F, Ai, B & Qian, J 2008,

A genetic algorithm based wrapper feature selection
method for classification of hyperspectral images using
support vector machine Proceedings of Geoinformatics
and Joint Conference on GIS and Built Environment:
Classification of Remote Sensing Images, pp. 71471J71471J.

[23] Lin, SW, Ying, KC, Chen, SC & Lee, ZJ 2008, Particle
swarm optimization for parameter determination and
feature selection of support vector machines, Expert
systems with applications, vol. 35, no. 4, pp.1817-1824.

[36] Quinlan JR 2014, C4.5: programs for machine learning,

Morgan Kaufmann publishers, San Mateo, California.

[24] Liu, H & Setiono, R 1995, Chi2: Feature selection and

discretization of numeric attributes, Proceedings of the
IEEE Seventh International Conference on Tools with
Artificial Intelligence, Washington DC, USA, pp. 388391.
[25] Dash, M & Liu, H 1997, Feature selection for
classification, Intelligent data analysis, vol. 1, no.1,
pp.131-156.

[37] Koistinen, P & Holmstrm, L 1991, Kernel regression

and backpropagation training with noise, Proceedings of
IEEE International Joint Conference on Neural
Networks, pp. 367-372.
[38] Baluja, S 1994, Population-based incremental learning a
method for integrating genetic search based function
optimization and competitive learning, Technical Report
No. CMU-CS-94-163, Carnegie Mellon University,
Pittsburgh, Pa.
[39] Buntine, W 1991, Theory refinement on Bayesian
networks Proceedings of the Seventh conference on

15

International Journal of Computer Applications (0975 8887)

Volume 136 No.1, February 2016
Uncertainty in Artificial Intelligence, Barcelona, Spain,
pp. 52-60.
[40] Loughrey, J & Cunningham, P 2005, Overfitting in
wrapper-based feature subset selection: The harder you
try the worse it gets Proceedings of Research and
Development in Intelligent Systems, Springer London.
pp. 33-43.

[55] Battiti, R 1994, Using mutual information for selecting

features in supervised neural net learning, IEEE
Transactions on Neural Networks, vol. 5, no. 4, pp.537550.
[56] Fleuret, F 2004, Fast binary feature selection with
conditional mutual information, The Journal of Machine
Learning Research, vol. 5, pp.1531-1555.

[41] Freeman, C, Kuli, D & Basir, O 2015, An evaluation of

classifier-specific filter measure performance for feature
selection, Pattern Recognition, vol.48, no.5, pp.18121826.

[57] Meyer, PE & Bontempi, G 2006, On the use of variable

complementarity for feature selection in cancer
classification, Applications of Evolutionary Computing.
pp. 91-102.

[42] Chandrashekar G & Sahin, F 2014, A survey on feature

selection methods, Computers & Electrical Engineering,
vol. 40, no.1, pp.16-28.

[58] Lin, D & Tang, X 2006, Conditional infomax learning:

an integrated framework for feature extraction and
fusion, Proceeding of ninth European Conference on
Computer Vision, Graz, pp. 68-82.

[43] Guyon, I, Weston, J, Barnhill, S & Vapnik, V, 2002,

Gene selection for cancer classification using support
vector machines, Machine learning, vol. 46, no. 1-3,
pp.389-422.
[44] Quinlan, JR, 1986, Induction of decision trees,
Machine learning, vol. 1, no.1, pp.81-106.
[45] Tibshirani, R, Saunders, M, Rosset, S, Zhu, J & Knight,
K 2005, Sparsity and smoothness via the fused lasso,
Journal of the Royal Statistical Society: Series B
(Statistical Methodology), vol. 67, no. 1, pp. 91-108.

[59] Brown, G, Pocock, A, Zhao, MJ & Lujn, M 2012,

Conditional likelihood maximisation: a unifying
framework for information theoretic feature selection,
The Journal of Machine Learning Research, vol.13, no.1,
pp.27-66.
[60] Song, Q, Ni, J & Wang, G 2013, A fast clustering-based
feature subset selection algorithm for high-dimensional
data, IEEE Transactions on Knowledge and Data
Engineering, vol. 25, no.1, pp.1-14.

[46] Ma, S & Huang, J 2008, Penalized feature selection and

classification in bioinformatics, Briefings in
bioinformatics, vol. 9, no. 5, pp.392-403.

[61] Dhillon, IS, Mallela, S & Kumar, R 2003, A divisive

information theoretic feature clustering algorithm for text
classification, The Journal of Machine Learning
Research, vol. 3, pp.1265-1287.

[47] Neumann, J, Schnrr, C & Steidl, G 2004, SVM-based

feature selection by direct objective minimisation,
Proceeding of the twenty-sixth DAGM Symposium on
Pattern Recognition, Germany, pp. 212-219.

[62] Li, Y, Luo, C, & Chung, SM 2008, Text clustering with

feature selection by using statistical data. IEEE
Transactions on Knowledge and Data Engineering, vol.
20, no.5, pp.641-652.

[48] Xiao, Z, Dellandrea, E, Dou, W & Chen, L 2008, ESFS:

A new embedded feature selection method based on
SFS, Rapports de recherch.

[63] Cai, D, Zhang, C & He, X 2010, Unsupervised feature

selection for multi-cluster data, Proceedings of the
sixteenth ACM SIGKDD international conference on
Knowledge discovery and data mining, Washington, pp.
333-342.

[49] Maldonado, S, Weber, R & Famili, F 2014, Feature

selection for high-dimensional class-imbalanced data sets
using Support Vector Machines, Information Sciences,
vol. 286, pp.228-246.
[50] Kira, K & Rendell, LA 1992, A practical approach to
feature selection, Proceedings of the ninth international
workshop on Machine learning, Aberdeen, Scotland, UK
(pp. 249-256).
[51] Kononenko, I 1994, Estimating attributes: analysis and
extensions of RELIEF. Proceeding of European
Conference on Machine Learning, Catania, Italy, pp.
171-182.
[52] Holte, RC 1993, Very simple classification rules
perform well on most commonly used datasets, Machine
learning, vol.11, no.1, pp.63-90.
[53] Yang, HH & Moody, JE 1999, Data Visualization and
Feature Selection: New Algorithms for Nongaussian
Data, Advances in Neural Information Processing
Systems, vol. 99, pp. 687693.
[54] Peng, H, Long, F & Ding C 2005, Feature selection
based on mutual information criteria of max-dependency,
max-relevance, and min-redundancy, IEEE Transactions
on Pattern Analysis and Machine Intelligence, vol. 27,
no.8, pp.1226-1238.

[64] Chow, TW & Huang, D 2005, Estimating optimal

feature subsets using efficient estimation of highdimensional mutual information, IEEE Transactions on
Neural Networks, vol.16, no.1, pp.213-224.
[65] Mitra S, & Acharya T 2005, Data mining: multimedia,
soft computing, and bioinformatics John Wiley & Sons,
New Jersey.
[66] Sotoca, JM & Pla, F 2010, Supervised feature selection
by clustering using conditional mutual information-based
distances, Pattern Recognition, vol. 43, no.6, pp.20682081
[67] Freeman, C, Kuli, D & Basir, O 2015, An evaluation of
classifier-specific filter measure performance for feature
selection, Pattern Recognition, vol.48, no.5, pp.18121826.
[68] Frnay, B, Doquire, G & Verleysen, M 2014,
Estimating mutual information for feature selection in
the presence of label noise, Computational Statistics &
Data Analysis, vol. 71, pp.832-848.
[69] Tabakhi, S, Moradi, P & Akhlaghian, F 2014, An
unsupervised feature selection algorithm based on ant

16

International Journal of Computer Applications (0975 8887)

Volume 136 No.1, February 2016
colony optimization, Engineering Applications of
Artificial Intelligence, vol. 32, pp.112-123.
[70] Bermejo, P, Gmez, J & Puerta, J 2008, On incremental
wrapper-based attribute selection: experimental analysis
of the relevance criteria, Proceedings of International
Conference on Information Processing and Management
of Uncertainty in Knowledge-Based Systems, France,
pp.638-645.
[71] Ruiz, R, Riquelme, JC & Aguilar-Ruiz, JS 2006,
Incremental wrapper-based gene selection from
microarray data for cancer classification Pattern
Recognition, vol. 39, no. 12, pp.2383-2392.
[72] Xie, J, Xie, W, Wang, C & Gao, X 2010, A Novel
Hybrid Feature Selection Method Based on IFSFFS and
SVM for the Diagnosis of Erythemato-Squamous
Diseases, Proceedings of Workshop on Applications of
Pattern Analysis, Cumberland Lodge, Windsor, UK, pp.
142-151.

using maximum relevance-minimum redundancy and

artificial neural network input gain measurement
approximation (ANNIGMA), Proceedings of the ThirtyFourth Australasian Computer Science Conference,
Australia, vol. 113, pp. 43-52.
[77] Gunal, S 2012, Hybrid feature selection for text
classification, Turkish Journal of Electrical Engineering
and Computer Sciences, vol. 20, no.2, pp.1296-1311.
[78] D'Alessandro, M, Esteller, R, Vachtsevanos, G, Hinson,
A, Echauz, J & Litt, B 2003, Epileptic seizure prediction
using hybrid feature selection over multiple intracranial
EEG electrode contacts: a report of four patients, IEEE
Transactions on Biomedical Engineering, vol. 50, no.5,
pp.603-615.
[79] Yang CS, Chuang LY, Ke CH, & Yang CH, 2008,A
hybrid feature selection method for microarray
classification, IAENG International Journal of Computer
Science, vol. 35, no. 3, pp. 1-3.

[73] Kannan, SS & Ramaraj, N 2010, A novel hybrid feature

selection via Symmetrical Uncertainty ranking based
local memetic search algorithm, Knowledge-Based
Systems, vol. 23, no. 6, pp.580-585.

[80] Bermejo, P, Gmez, JA & Puerta, JM 2011, A GRASP

algorithm for fast hybrid (filter-wrapper) feature subset
selection in high-dimensional datasets, Pattern
Recognition Letters, vol. 32, no.5, pp.701-711.

[74] Xie, J, Lei, J, Xie, W, Shi, Y & Liu, X 2013, Two-stage

hybrid feature selection algorithms for diagnosing
erythemato-squamous diseases, Health Information
Science and Systems, vol.1, no.10, pp.2-14.

[81] Foithong, S, Pinngern, O & Attachoo, B 2012, Feature

subset selection wrapper based on mutual information
and rough sets, Expert Systems with Applications, vol.
39, no.1, pp.574-584.

[75] Naseriparsa, M, Bidgoli, AM & Varaee, T 2013, A

Hybrid Feature Selection method to improve
performance of a group of classification algorithms,
International Journal of Computer Applications, vol. 69,
no. 17, pp. 0975 8887.

[82] Coates A, Ng AY (2012) Learning Feature

Representations with K-Means. In: Montavon, G., Orr,
G.B., and Mller, K.-R. (eds.) Neural Networks: Tricks
of the Trade. pp. 561580. Springer Berlin
Heidelberg.doi: 10.1007/978-3-642-35289-8_30

[76] Huda, S, Yearwood, J & Stranieri, A 2011, Hybrid

wrapper-filter approaches for input feature selection

IJCATM : www.ijcaonline.org

17