Received January 9, 2021, accepted January 28, 2021, date of publication February 2, 2021, date of current version February

17, 2021.
Digital Object Identifier 10.1109/ACCESS.2021.3056407

Metaheuristic Algorithms on Feature Selection:

A Survey of One Decade of Research (2009-2019)
PRACHI AGRAWAL 1 , HATTAN F. ABUTARBOUSH 2, TALARI GANESH1 ,
AND ALI WAGDY MOHAMED 3,4
1 Department of Mathematics and Scientific Computing, National Institute of Technology at Hamirpur, Hamirpur 177005, India
2 College of Engineering, Electrical Engineering Department, Taibah University, Medina 42353, Saudi Arabia
3 Operations Research Department, Faculty of Graduate Studies for Statistical Research, Cairo University, Giza 12613, Egypt
4 Wireless Intelligent Networks Center (WINC), School of Engineering and Applied Sciences, Nile University, Giza 16453, Egypt

Corresponding author: Prachi Agrawal ([email protected])

The work of Prachi Agrawal was supported by the Ministry of Human Resource Development (MHRD Govt. of India).

ABSTRACT Feature selection is a critical and prominent task in machine learning. To reduce the dimension
of the feature set while maintaining the accuracy of the performance is the main aim of the feature selection
problem. Various methods have been developed to classify the datasets. However, metaheuristic algorithms
have achieved great attention in solving numerous optimization problem. Therefore, this paper presents an
extensive literature review on solving feature selection problem using metaheuristic algorithms which are
developed in the ten years (2009-2019). Further, metaheuristic algorithms have been classified into four
categories based on their behaviour. Moreover, a categorical list of more than a hundred metaheuristic algo-
rithms is presented. To solve the feature selection problem, only binary variants of metaheuristic algorithms
have been reviewed and corresponding to their categories, a detailed description of them explained. The
metaheuristic algorithms in solving feature selection problem are given with their binary classification, name
of the classifier used, datasets and the evaluation metrics. After reviewing the papers, challenges and issues
are also identified in obtaining the best feature subset using different metaheuristic algorithms. Finally, some
research gaps are also highlighted for the researchers who want to pursue their research in developing or
modifying metaheuristic algorithms for classification. For an application, a case study is presented in which
datasets are adopted from the UCI repository and numerous metaheuristic algorithms are employed to obtain
the optimal feature subset.

INDEX TERMS Binary variants, classification, feature selection, literature review, metaheuristic algorithms.

I. INTRODUCTION selects the relevant features from the original dataset. This
The real-world problems mostly include a large number paper focuses only on feature selection problems.
of data, and handling the data becomes a very complex A feature selection problem is one of the most challenging
and prominent task. A dataset contains a large number of tasks in machine learning. If a set contains n number of
attributes/features. Not always, all the features are necessary features, total 2n subsets are possible from which the best
to get useful information from the datasets. Some of the subset has to be picked. It will be most difficult when n tends
features may be irrelevant, redundant, which degrades the to a large number because it can not be possible to evaluate
performance of the model. Therefore, to reduce the size of the performance of the model at each subset. Hence, to handle
the original datasets while maintaining the accuracy of perfor- the situation, the various methodology has been proposed.
mance is the main aim of feature reduction problem. In fea- Exhaustive search, greedy search, random search etc. are
ture reduction, feature construction and feature selection take such techniques which have been applied to feature selection
part. The feature extraction or construction constructs a new problems to find the best subset. Most of the methods suf-
set of features from original datasets while feature selection fer from premature convergence, enormous complexity, high
computational cost. Therefore, metaheuristic algorithms get
The associate editor coordinating the review of this manuscript and much attention to deal with this type of conditions. They are
approving it for publication was Firooz B. Saghezchi . the most efficient and effective techniques and are capable

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
26766 VOLUME 9, 2021
P. Agrawal et al.: Metaheuristic Algorithms on Feature Selection: A Survey of One Decade of Research (2009-2019)

FIGURE 1. The number of papers published regarding the developing of metaheuristic algorithms and feature selection
problem.

of finding the best subset of features while maintaining the algorithms to feature selection problem, especially in med-
accuracy of the model. ical datasets. They provided a categorization of binary and
In the last three decades, several metaheuristic algorithms chaotic algorithms for different nature-inspired algorithms.
have been designed to solve various kinds of optimization Literature review has been presented in different fields such
problems. This study provides an extensive literature survey as sentiment analysis [5], bioinformatics [6], ensemble learn-
on metaheuristic algorithms which are developed in the last ing [7] using metaheuristic algorithms.
ten years (2009-2019) and applied to various applications The main contribution of presenting this study is given
of feature selection problems. Due to its various applica- as:
tions in different fields such as text mining, image process- (a) This paper presents the definitions and techniques of
ing, bioinformatics, industrial applications, computer vision feature selection problem, and basic concepts of meta-
etc. feature selection becomes an exciting research problem. heuristic algorithms are thoroughly explained.
There are lots of papers published in different publishers (b) The metaheuristic algorithms are classified, and a list of
regarding the developing metaheuristic algorithm and fea- metaheuristic algorithms is given.
ture selection problems that is shown in Figure 1. From (c) It presents an extensive literature of binary metaheuristic
the figure, it can be observed that Elsevier Publisher pub- algorithms for feature selection problem.
lishes more papers in top tier journals (Experts Systems with (d) The literature is represented with the vital factor of wrap-
Applications (IF = 5.452), Applied Soft Computing (IF = per feature selection techniques such as the description
5.472), Knowledge-Based Systems (IF = 5.921), Informa- of the classifier, name of the used datasets, evaluation
tion Sciences (IF = 5.910), Neurocomputing (IF = 3.317)). metrics etc.
In Springer publication, the papers are published in top tier (e) It explains the issues and challenges to develop an
journals Neural Computing and Applications (IF = 4.774), algorithm in solving feature selection problems. It also
Applied Intelligence (IF = 3.325), Soft Computing (IF = presents the evaluation metrics formula to investigate the
3.050) etc. The other publishers consist of IOS Press, Mas- performance.
sachusetts Institute of Technology (MIT) press, Citeseer, (f) Finally, the research gaps and the future work are also
ACM digital library, Hindawi publishing house, Multidisci- presented to enhance the research work.
plinary Digital Publishing Institute (MDPI). (g) Eight benchmark datasets have been considered from
Earlier, a literature survey has been found on feature selec- the UCI repository to show the application of feature
tion in which non-evolutionary algorithms have been con- selection using wrapper based techniques.
sidered. Xue et al. [1] presented a survey on evolutionary (h) Five metaheuristic algorithms are taken from the litera-
approaches which mainly focus on genetic algorithm, par- ture to implement on feature selection problem.
ticle swarm optimization, ant colony optimization, genetic (i) The results are compared with evaluation metrices i.e.
programming in detail. Lee et al. [2] studied feature selec- average fitness value, average classification accuracy,
tion in multimedia applications in which they provided an average number of feature selected and average compu-
extensive literature survey of seventy related papers from tational time.
2001 to 2017. Remeseuro and Bolon-Canedo [3] reviewed The organization of the paper as follows: Section II presents
feature selection methods on medical problems. Sharma and the preliminaries for the feature selection and metaheuristic
Kaur [4] presented a systematic review on nature-inspired algorithms. The extensive literature on feature selection using

VOLUME 9, 2021 26767

 P. Agrawal et al.: Metaheuristic Algorithms on Feature Selection: A Survey of One Decade of Research (2009-2019)

FIGURE 2. Overall feature selection process.

metaheuristic algorithms is given in Section III. The issues Several feature selection methods are developed to obtain the
and challenges are presented in Section IV and Section V best subset of features. Generally, the techniques are classi-
illustrates a case study based on feature selection problem. fied into three categories filter, wrapper and embedded meth-
Section VI suggests the future work based on wrapper feature ods [12]–[15]. Filter methods are independent of learning
techniques. The concluding remarks are shown in Section VI. or classification algorithm. It always focuses on the general
characteristics of the data [16]. Wrapper methods always
II. BACKGROUND include the classification algorithm and interact with the
This section presents the detailed description of the feature classifier. These are computationally expensive methods than
selection problem with the mathematical model and the def- the filter and also provide more accurate results as compared
initions, concepts and the classifications of metaheuristic to filter methods. Embedded methods are a combination of
algorithms. filters and wrapper methods. In embedded methods, the fea-
ture selection is a part of the training process and training
A. FEATURE SELECTION process held with the classifier. Moreover, the embedded
Feature selection deals with inappropriate, irrelevant, methods use learning algorithm in its process, they will be
or unnecessary features. It is a process that extracts the best considered in wrapper approaches category [17].
features from the datasets [8]. Feature selection is one of the Wrapper approaches present better results in comparison
most critical and challenging problems in machine learning. with filter methods, but they are slower than filters meth-
The various applications of the feature selection problem ods. Wrapper methods depend on the modelling algorithm
can be demonstrated in different fields. There are some in which every subset is generated and then evaluated. Sub-
applications such as biomedical problems (to find the best set generation in wrapper methods is based on the differ-
gene from candidate gene) [9]; text mining (to find the best ent search strategy. Jovic et al. [18] differentiates search
terms word or phrases) [10]; image analysis (to select the techniques into three categories; exponential, sequential and
best visual contents pixels, colour) [11] etc. Mathematically, randomized selection strategy. In the exponential method,
a feature selection problem can be formulated in the following the number of evaluated features increases exponentially with
way: the size of features. Although this method shows accurate
Assume a dataset 0 S 0 contains 0 d 0 number of features. Then results, it is not practically possible to apply because of
the working mechanism of feature selection problem is to the high computational cost. The examples for exponential
select relevant features among 0 d 0 features. search strategy are exhaustive search, branch and bound
Given dataset S = {f1 , f2 , f3 , . . . , fd } method [19], [20]. Sequential algorithms include or remove
The objective is to select the best subsets of features from S. features sequentially. Once a feature is included or removed
Extract Subset D = {f1 , f2 , f3 , . . . , fn } where, n < d in the selected subset, it can not be further changed that
and f1 , f2 , f3 , . . . , fn represents the features/attributes of any leads to local optima. Some sequential algorithms are linear
dataset. Figure 2 depicts the working mechanism of the forward selection, floating forward or backward selection,
feature selection process. From the figure, it can be observed best first etc. Randomized algorithms include randomness to
that there are five main components of the feature selection explore the search space, which saves the algorithms from
process, i.e. original dataset, selection of feature subset, eval- trapping into local optima. Randomized algorithms are com-
uation of feature subset, selection criterion and validation. monly known as population-based approaches for example

26768 VOLUME 9, 2021

P. Agrawal et al.: Metaheuristic Algorithms on Feature Selection: A Survey of One Decade of Research (2009-2019)

FIGURE 3. Classification of feature selection methods.

simulated annealing, random generation, metaheuristic algo- generation), industrial fields (scheduling jobs, transportation,
rithms etc. [21]. vehicle routing problem, facility location problem), in civil
We do not present a detailed description of every method of engineering (to design the bridges, buildings), communica-
the feature selection process. The detailed explanation of each tion (radar design, networking), data mining (classification,
method can be found in [1]. The flow chart of categorization prediction, clustering, system modelling) etc.
of methods for solving feature selection is shown in Figure 3. Metaheuristic algorithms classify into the following two
In the figure, the dashed line box represents the methodology main categories;
of this paper which describes how we reach to metaheuristic (i) Single solution based metaheuristic algorithms:
algorithms. These techniques start their optimization process with
one solution, and their solution is updated during the
B. METAHEURISTIC ALGORITHMS iterations. It may lead to trapping into local optima and
Metaheuristic algorithms are optimization methods that also does not explore the search space thoroughly.
obtain the optimal (near-optimal) solution of optimization (ii) Population (multiple) solution based metaheuristic
problems. These algorithms are derivative-free techniques algorithms: Initially, these algorithms generate a pop-
and, have simplicity, flexibility and capability to avoid local ulation of solutions and start their optimization process.
optima [22]. The behaviour of metaheuristic algorithms are The population of solutions update with the number of
stochastic; they start their optimization process by generat- generations/iterations. The algorithms are beneficial for
ing random solutions. It does not require to calculate the avoiding local optima as multiple solutions assist each
derivative of search space like in gradient search techniques. other and have a great exploration of search space. They
The metaheuristic algorithms are flexible and straightforward also have the quality of jump towards the promising
due to the simple concept and easy implementation. The part of search space. Therefore, population-based algo-
algorithms can be modified easily according to the particular rithms use in solving most of the real-world problems.
problem. The main property of metaheuristic algorithms is Researchers pay great attention to metaheuristic algorithms
that they have a remarkable ability to prevent the algorithms because of their characteristics. Several algorithms have been
from premature convergence. Due to the stochastic behaviour designed and solved different types of problems. Based on
of algorithms, the techniques work as a black box and their behaviour, the metaheuristic algorithms can be divided
avoid local optima and explore the search space efficiently into four categories; evolution-based, swarm intelligence-
and effectively. The algorithms make a tradeoff between based, physics-based and human-related algorithms [25]. The
its two main essential aspects exploration and exploitation categorization of the algorithms is depicted in Figure 4.
[23], [24]. In the exploration phase, the algorithms investi- (1) Evolution based algorithms: It is inspired from the
gate the promising search space thoroughly, and exploitation natural evolution and start their process with randomly
comes for the local search of promising area(s) that are generated population of solutions. In these type of algo-
found in the exploration phase. They are successfully applied rithms, the best solutions are put togther to create new
to various engineering and sciences problems, e.g. in elec- individuals. The new individuals are formed using muta-
trical engineering (to find the optimal solution for power tion, crossover and select the best solution. The most

VOLUME 9, 2021 26769

 P. Agrawal et al.: Metaheuristic Algorithms on Feature Selection: A Survey of One Decade of Research (2009-2019)

FIGURE 4. Classification of metaheuristic algorithms.

popular algorithm in this category is Genetic algo- TABLE 1. List of Evolution based algorithms from 2009-2019.
rithm (GA) that is based on Darwin evolution tech-
nique [26]. There are other algorithms such as evolution
strategy [27], genetic programming [28], tabu search
[29], differential evolution [30] etc.
(2) Swarm intelligence-based algorithms: These algo-
rithms are inspired by the social behaviours of insects,
animals, fishes or birds etc. The popular technique is Par-
ticle Swarm Optimization (PSO) developed by Kennedy
and Eberhart [31]. It is inspired by the behaviour of a sections, the algorithms which have been developed in ten
group of birds that fly throughout the search space and years span and applied to feature selection problems are
find their best location (position). Ant Colony optimiza- discussed.
tion [32], Honey bee swarm optimization algorithm [33],
monkey optimization [34] etc are the examples of swarm III. METAHEURISTIC ON FEATURE SELECTION
intelligence algorithms. It describes the metaheuristic algorithm, which has been
(3) Physics based algorithms: These are inspired by the used in solving the feature selection problem. Binary vectors
rules of physics in the universe. Simulated anneal- representations are considered to obtain the relevant feature.
ing [35], Harmony search [36] etc come under In the designed algorithm, a solution vector is represented
physics-based algorithms. by (10101100 . . . ..) this implies that 1 means that a par-
(4) Human behaviour related algorithms: These tech- ticular feature is selected and 0 means that feature is not
niques are purely inspired by human behaviour. Every selected in the subset. Hence, this section investigates all
human being has its way of doing activities that affect binary variants of metaheuristic algorithms in detail. The first
its performance. It motivates researchers to develop section describes the evolution-based algorithms; the sec-
the algorithms. The popular algorithms are Teaching ond describes the swarm intelligence based algorithms, third
learning-based optimization algorithm (TLBO) [37], demonstrates the physics-based algorithms, and the fourth
League Championship algorithm [38] etc. one is for the human-related algorithm. And the last section
is for the hybrid algorithms, which are a combination of two
It is worth mentioning here that there are many metaheuristic or more metaheuristic algorithms that have been used for
algorithms developed from 1966 to till now. In this paper, classification problems.
we present the literature of those algorithms which are devel-
oped or proposed since 2009 to 2019 (ten-year span). Accord- A. EVOLUTION BASED ALGORITHMS
ing to the category, the list of metaheuristic algorithms are From Table 1, it can be seen that there are very few algo-
presented in Table 1, 2, 3, 4. The first column of the tables rithms are developed in evolution based category from 2009-
present the abbreviation; the second column gives the name 2019. Gan and Duan [123] proposed a chaotic differen-
of the algorithms, which is followed by the developed year. tial search algorithm for image processing and it has been
These algorithms have been applied to solve many real-world combined with lateral inhibition to edge extraction and
applications but, this paper is restricted to present the applica- image enhancement. Negahbani et al. [124] used differen-
tion in feature selection problems. Therefore, in the following tial search algorithm for the diagnosis of coronary artery

26770 VOLUME 9, 2021

P. Agrawal et al.: Metaheuristic Algorithms on Feature Selection: A Survey of One Decade of Research (2009-2019)

TABLE 2. List of Swarm Intelligence based algorithms from 2009-2019. TABLE 3. List of Physics based algorithms from 2009-2019.

TABLE 4. List of Human behaviour related algorithms from 2009-2019.

disease with fuzzy c-means that was used as a classifier. The B. SWARM INTELLIGENCE BASED ALGORITHMS
performance of the proposed approach has been evaluated This section presents a detailed description of some
using accuracy, sensitivity and specificity measures. Zhang well-known algorithms which are based on the different
et al. [125] proposed binary backtracking algorithm for wind swarm behaviour and modified to solve feature selection
speed forecasting in which extreme learning machine was problems. We have given our best to present the review of
employed for feature selection. Binary backtracking algo- algorithms with their modifications, many datasets used and
rithm was developed using a sigmoidal function that trans- some other information which will be useful to provide a
forms the continuous variables to binary variables. To identify quick idea about the published research paper.
the Leukemia cancer symptoms, Dhal et al. [126] imple-
mented the stochastic fractal search algorithm to provide opti- 1) CUCKOO SEARCH
mal identification. The developed algorithm was compared Cuckoo search algorithm was developed by observing the
with other classical methods and achieved high accuracy. behaviour of cuckoo birds and their reproduction strategy.
Besides, a binary stochastic fractal search was developed It is very well known and popular algorithm and has achieved
to classify the galaxy colour images with extreme machine great success in solving various real-world problems. How-
learning [127]. ever, several binary versions of cuckoo search algorithm have

VOLUME 9, 2021 26771

 P. Agrawal et al.: Metaheuristic Algorithms on Feature Selection: A Survey of One Decade of Research (2009-2019)

been developed in solving binary optimization problems. of features, computational time and accuracy of the classi-
In 2012, Tiwari [128] used CS algorithm for face recognition fier were used. Naik et al. [144] used BBA with one-pass
problem. Firstly, features were extracted using discrete cosine generalized classifier neural network to estimate the number
transformation which worked as host egg in CS algorithm. of selected features. The fitness function was formed using
It proved its efficiency for finding out the most matching sensitivity and specificity with the classification accuracy.
image in face recognition. To obtain the optimal feature The proposed approaches were evaluated using six avail-
subset, Rodrigues et al. [129] proposed a binary CS algo- able microarray datasets, and the obtained results were com-
rithm (BCS) using a function which converts the continuous pared with the different type of classifiers. Moreover, several
variable to its binary form. BCS has successfully applied to modifications have been done to improve the performance
two datasets of theft detection in a power system with opti- of BBA and applied to various real-world classification
mum path forest classifier and obtained that BCS was the very problems [145]–[148].
suitable and fastest approach in solving feature selection for
industrial datasets. Gunavathi and Premalatha [130] used CS 3) FIREFLY ALGORITHM
algorithm for classification in microarray gene data of can- It imitates the mechanism of firefly mating and exchange
cer. The features were ranked according to T and F-statistics, of information using light flashes. Emary et al. [149] pro-
and KNN classifier was used as a fitness function. The posed first binary version of firefly algorithm (FFA) to
results showed that CS algorithm obtained 100% accuracy solve feature selection problems using a threshold value.
in most of the datasets. Sudha and Selvarajan [131] pro- The developed algorithm made a good exploration quality
posed an enhanced CS algorithm to find the optimal fea- which found a quick solution to the problem. It has been
tures for breast cancer classification. In ECS, classification applied to benchmark datasets of UCI with KNN classifier.
accuracy was used as an objective function and evaluated Kanimozhi and Latha [150] presented a technique for image
using KNN classifier. Salesi and Cosma [132] extended retrieval by using SVM classifier and FFA. The main aim
BCS algorithm in which pseudobinary mutation neighbour- was to increase the performance of the algorithm with opti-
hood search was designed to solve feature selection prob- mal features, and the algorithm has been tested over Corel
lems of biomedical datasets. Pandey et al. [133] introduced Caltech and Pascal database images. To predict the disease,
binary binomial cuckoo search algorithm to select the best cardiotocogram data has been used with SVM and FFA by
features and applied to fourteen benchmark datasets of Subha and Murugan [151]. Zhang et al. [152] proposed a
UCI [134]. Although lots of applications of machine learning return cost based FFA for a public dataset in which binary
have been solved by developing different versions of CS movement operator has been used to update the position of
algorithm [135]–[137]. fireflies. The proposed approach proved that it was very com-
petitive with other algorithms. A self-adaptive FFA has been
2) BAT ALGORITHM developed for feature selection using mutual information cri-
It is inspired by the behaviour of bats and very popular terion [153]. Two modifications have been done for getting
in solving various real-world problems. In solving feature rid of trapping into local optima. Twelve datasets has been
selection problems, firstly Nakamura et al. [138] developed a considered for evaluating the performance of the algorithm.
binary version of BA (BBA) with sigmoid function to restrict Xu et al. [154] combined binary FFA with opposition based
the position of bat’s to binary variables. The accuracy was learning algorithm in solving the feature selection problem
calculated using optimum path forest classifier, and BBA was and applied to ten datasets. For network intrusion detection,
applied to five datasets. Laamari and Kamel [139] improved FFA algorithm has been used with C4.5 and Bayesian net-
the performance of BBA by applying other V-shaped transfer works classifier and utilized for KDD CUP 99 datasets [155].
function with SVM classifier to intrusion detection systems. For more versions of FFA in various applications of feature
Rodrigues et al. [140] used different transfer functions for selection, interest researchers can be found in [156]–[159].
obtaining binary-based optimization techniques with opti-
mum path forest classifier. The binary BA presented best 4) FLOWER POLLINATION ALGORITHM
results with hyperbolic tangent transfer function in most of FPA is inspired from the pollination procedure of flowers.
the datasets. Enache and Sgârciu [141] improved the ver- Several binary variants of FPA were developed to solve fea-
sion of BBA with different classifiers such as SVM and ture selection problem. Firstly, Rodrigues et al. [160] pro-
C4.5 and applied to NSL-KDD datasets for intrusion detec- posed a binary constrained version of FPA (BFPA) in which
tion systems. Yang et al. [142] modified BA to improve the binary solutions were obtained after generating local polli-
diversity of the population of bats so that it made a good nations. The BFPA run with optimum path forest classifier,
balance between exploration and exploitation and solve the which calculated the performance accuracy. The obtained
feature selection problem. To classify the MR brain tumour results were compared with other state-of-the-art metaheuris-
image, Kaur et al. [143] modified BA by combining Fisher tic algorithms PSO, HS and FA and proved that BFPA is also
and parameter-free bat algorithm for good exploration. The suitable to get the optimal feature subsets. To improve the
dataset has been taken from UCI repository and applied with performance of BFPA, Zawbaa and Emary [161] used KNN
SVM classifier. The evaluation measure such as the number classifier with a modified binary variant of FPA. In binary

26772 VOLUME 9, 2021

P. Agrawal et al.: Metaheuristic Algorithms on Feature Selection: A Survey of One Decade of Research (2009-2019)

version, a threshold value was used to get a binary string UCI datasets. Moreover, in the initialization phase, small,
from continuous variables. Moreover, the proposed algorithm random and large initialization techniques were used for
was evaluated on a multi-objective function for classification good exploration. Sharma et al. [171] proposed a modified
and regression datasets. It showed that the algorithm outper- version of GWO for identifying the symptoms of Parkin-
formed PSO, GA and BA. son’s disease with random forest, KNN and decision tree
Rajamohana et al. [162] used different values of λ parame- classifier. Pathak et al. [172] proposed levy flight GWO
ter and profounded an adaptive scheme of BFPA i.e. ABFPA which was used to select the relevant features from the
and solved feature selection problem. For text classification, original datasets. It used random forest classifier for image
Majidpour et al. [163] used BFPA with ada-boost algorithm, steg analysis and applied to Bossbase ver 1.01 datasets. The
which worked as a classifier. Sayed et al. [164] considered obtained results showed its great performance for achieving
clonal selection algorithm for exploitation and FPA for explo- great convergence. Devanathan et al. [173] identified the
ration to form the binary clonal FPA. The proposed CFPA optimal features for Indirect Immunofluorescence image by
was applied to three datasets i.e. Australian, breast cancer presenting another version of binary GWO. To diagnosis
and german number and obtained the results. Yan et al. of cardiovascular disease, Al-Tashi et al. [174] used GWO
[165] improved the proposed version by considering a group algorithm for selecting the best features and SVM used as a
strategy to avoid local optima, adoptive transfer function was classifier. The proposed approach has been applied to Cleve-
used for binary encoding, and Gaussian mutation strategy land dataset, which is freely available and performed greatly.
was used for exploitation. Based on these modifications, Moreover, the author proposed a binary version (BMOGW-
improved version IBCFPA was applied to six biomedical S) by using a sigmoidal function for solving multi-objective
datasets with different classifiers (SVM, KNN and NB) and feature selection problem in which the artificial neural net-
obtained optimal feature subsets for every dataset. work was used for classification. BMOGW-S applied to fif-
teen benchmark datasets and compared with MOGWO with
5) KRILL HERD ALGORITHM tanh transfer function [175]. Hu et al. [176] proposed new
It is based on the movement of Antarctic krills to search transfer functions and new updating scheme for parameters of
their food and to increase the density. Rodrigues et al. [166] GWO. Advanced GWO (ABGWO) applied to twelve datasets
solved feature selection problem by introducing the binary of UCI and showed superior results as compared to other
variant of KH algorithm (BKH) that generated binary vectors algorithms. There are several versions of GWO are developed
by evaluating the transfer operator. Optimum path forest used for classification in different fields such as medical diagno-
as a classifier and the proposed algorithm BKH applied to six sis [177], cervical cancer [178], electromyography (EMG)
datasets and presented the results. Mohammadi et al. [167] signal [179], facial emotion recognition [180], text feature
considered breast cancer datasets for extracting the fuzzy selection [181] etc.
rules. They modified the KH algorithm as binary krill herd
fuzzy rule miner (BKH-FRM) that choose the best krill as
well as the local best krill among the population of krills. 7) ANT LION OPTIMIZER
The results obtained by BKH-FRM was compared with other ALO is a very popular algorithm and inspired from the hunt-
ten metaheuristic algorithms and presented high accuracy ing procedure of antlion insects and ants. It has been applied
among others. Rani and Ramyachitra [168] classified the to various real-world problems to find out the optimal (near-
cancer types using KH algorithm and random forest classifier. optimal) solutions. To solve the feature selection problem,
They modified the algorithm by using a horizontal crossover Zawbaa et al. [182] used a threshold value at continuous
and position mutation operator and applied to ten differ- variables to propose the binary variant of ALO algorithm.
ent gene microarray cancer datasets. For microarray data The proposed algorithm BALO with K-NN classifier was
classification, Zhang et al. [169] improved BKH algorithm applied to eighteen different datasets and compared with
named as IG-MBKH by using a hyperbolic tangent function well-known metaheuristic algorithms GA and PSO. They
and adaptive transfer function. Furthermore, to initialize the have used different evaluation criteria such as average classi-
population, information gain feature ranking was used to fication accuracy, the average number of the selected feature,
explore the search space efficiently. IG-MBKH was applied average Fisher score (F-score) etc. for calculating the perfor-
to high dimensional microarray datasets that generated rele- mance. In the next year, Emary et al. [183] proposed different
vant feature subsets. K-NN classifier used for calculating the variants of BALO in which each individual changed its posi-
accuracy of the selected features. tion according to the crossover operator between two binary
solutions. The binary solutions were obtained by applying
6) GREY WOLF OPTIMIZER S and V-shaped transfer functions or simply by using basic
It is based on the hunting process of a pack of grey wolves operators of ALO. Moreover, three initialization methods
in nature. In the first binary version of GWO, Emary et al. were adopted for a good exploration of the search space
[170] used the sigmoidal transfer function to get the binary and concluded that initialization process affected the search-
vectors (bGWO). To calculate the classification accuracy, ing quality and performance of the algorithm. Although,
KNN classifier was utilized and applied to eighteen different Mafarja et al. [184] proposed six different binary variants of

VOLUME 9, 2021 26773

 P. Agrawal et al.: Metaheuristic Algorithms on Feature Selection: A Survey of One Decade of Research (2009-2019)

ALO using S and V-shaped transfer functions and obtained 9) WHALE OPTIMIZATION ALGORITHM
optimal feature subsets. WOA is based on the special behaviour of the hunting method
To improve the efficiency of proposed binary versions, of humpback whales. To solve the binary optimization prob-
Zawbaa et al. [185] applied chaos theory to some parameter lems, Hussien et al. [199], [200] utilized S and V-shaped
of ALO algorithm. Considering of chaotic maps in ALO transfer function in conventional WOA and solved feature
algorithm, CALO avoided the local optima and made a selection problem with eleven UCI datasets in 2017. For
proper balance between exploration and exploitation qual- classification, KNN classifier was used, which ensured the
ity. CALO algorithm applied to ten biological datasets, and selected features for their relevancy. The proposed approach
eight datasets form other categories. The obtained solu- bWOA showed its capability for obtaining the maximum
tions proved its robustness and outperformed ALO, PSO accuracy and the minimum number of selected features.
and GA. To enhance the proposed approach, Sayed et al. [201] pre-
To save the algorithm from trapping into local optima, sented a chaotic whale optimization algorithm (CWOA) with
Emary and Zawbaa proposed a new approach of a binary ten chaotic maps. The chaotic maps were used in place ran-
variant of ALO [186]. Levy flight random walk was used dom parameters that made a better tradeoff between two main
to generate solutions, and five different initialization meth- important properties of algorithm exploration and exploita-
ods were utilized to generate the initial solutions. Other tion. Tubishat et al. [202] classified Arabic datasets for senti-
than, Mafarja and Mirjalili [187] embedded two rough ment analysis by proposing improved WOA (IWOA). IWOA
set filter approach Quick Reduct and CEBARKCC with incorporated the evolutionary operators such as crossover,
the ALO algorithm, which improved the initial popula- mutation and selection as in differential evolution. IWOA
tion as well as the final optimal solution. For hyper spec- applied to four publicly available datasets and compared with
tral image classification, a modified version of ALO algo- other techniques. Mafarja and Mirjalili [203] proposed two
rithm (MALO) was proposed with wavelet SVM (WSVM) binary variants of WOA by embedding crossover and muta-
classifier [188]. MALO with WSVM performed better in tion operator and by using tournament and roulette wheel
most of the datasets and proved that it was beneficial in solv- selection in WOA. Twenty benchmark datasets have been
ing the feature selection problems. Azar et al. [189] combined utilized in this approach. Agrawal et al. [204] proposed a
rough set theory with ALO algorithm and applied to different new version of WOA that is based on quantum concepts in
datasets. which, quantum bit representation was used for all individ-
uals. And the new version was applied to fourteen datasets.
There are some other versions of WOA [205]–[207] which
8) DRAGONFLY ALGORITHM are presented in solving feature selection problems.
It is inspired by the behaviour of dragonflies in nature and
applied to various problems. Medjahed et al. [190] pro-
10) GRASSHOPPER OPTIMIZATION ALGORITHM
posed a complete diagnosis procedure of cancer using binary
GOA is based on the lives of grasshopper how their behaviour
dragonfly (BDF) algorithm with SVM. In this, SVM-RFE
of living changes. Ibrahim et al. [208] proposed GOA with
(SVM-recursive feature elimination) used to extract the gene
SVM classifier (GOA+SVM) in which the parameters of
from the datasets, and BDF was used to enhance the perfor-
SVM were also optimized by using GOA. GOA+SVM
mance of SVM-RFE. The proposed algorithm was applied to
applied to biomedical datasets for Iraqui cancer patients and
six microarray datasets and presented high accuracy results.
obtained the results. To deal with immature convergence of
Mafarja et al. [191] proposed a binary version of DA using
GOA, Mafarja et al. [209] proposed GOA_EPD algorithm
a transfer function and solved the feature selection problem
by using evolutionary population dynamic, roulette wheel
with several datasets. Moreover, they also introduced a binary
and tournament selection for guiding the agent. Twenty-two
variant of dragonfly algorithm using time-varying transfer
benchmark datasets were considered for evaluating the per-
functions which make a proper balance between exploration
formance of the proposed approach. Hichem et al. [210]
and exploitation. These approaches have been applied to
introduced a new transfer function Hamming distance which
UCI datasets and compared with other state-of-the-art meta-
converted continuous variables into a binary vector. The new
heuristic algorithms [192]. Karizaki and Tavassoli [193] used
version of GOA (NBGOA) utilized for 20 standard datasets
filter and wrapper approaches simultaneously in which BDA
and compared with other versions of GOA. The presented
algorithm was used to find optimal subset of features and
results showed the ability to achieve great performance of
ReliefF algorithm was used as a filter approach. It has been
NBGOA. Sigmoidal and V-shaped transfer functions were
applied to five datasets and obtained the results. The BDA
used with mutation operator to enhance the exploration qual-
algorithm was applied to different learning algorithm such
ity of BGOA by Mafarja et al. [211].
as Naik et al. [194] used BDA with radial basis neural net-
work function and selected the features from microarray gene
data. Several other binary versions of DA have been pro- 11) SALP SWARM ALGORITHM
posed to solve features selection problem that can be found SSA is inspired by the salps swarming attitude. In 2017,
in [195]–[198]. Ibrahim et al. [212] first time used SSA for solving feature

26774 VOLUME 9, 2021

P. Agrawal et al.: Metaheuristic Algorithms on Feature Selection: A Survey of One Decade of Research (2009-2019)

selection problem (SSA-FS) by applying a threshold value datasets, evaluation metrics (the performance of the proposed
of 0.5 to build binary vectors. SSA-FS applied to medical algorithm is compared with these measures), name of the
datasets of breast, bladder and colon cancer datasets and compared techniques and some other information regarding
compared with other algorithms. Sayed et al. [213] proposed the proposed approaches. We find the following algorithms
chaotic SSA with ten chaotic maps and transfer function. such as multi-verse optimizer [222], sine-cosine algorithm,
KNN classifier has been used to evaluate the classifica- gravitational search algorithm etc. which are developed to
tion accuracy and applied to twenty benchmark datasets. obtain the optimal subset of features on different datasets.
Faris et al. [214] developed an efficient binary SSA using S
and V-shaped transfer functions in which a crossover operator D. HUMAN RELATED ALGORITHMS
was embedded in place of the average operator to enhance It gives a summary of the human-related algorithms in
the exploration quality. The proposed approach has been used solving feature selection problems. In the following descrip-
with KNN classifier and applied to twenty-two well known tion, we have discussed three algorithm brain storm opti-
UCI datasets and obtained that S-shaped transfer function mization, teaching-learning optimization and gaining sharing
provided best results. To get rid of trapping into local optima knowledge-based algorithm.
and enhance the exploration and exploitation of SSA, salp’s
position was updated using the singer’s chaotic map and 1) BRAIN STORM OPTIMIZATION
used local search algorithm by Tubisat et al. [215]. It has The algorithm works on the mechanism of human brainstorm-
been applied to twenty benchmark datasets and three Hadith ing. BSO algorithm was also applied in data classification.
datasets. Hegazy et al. [216] improved SSA (ISSA) by insert- Papa et al. [223] integrated binary BSO using different S and
ing weight to adjust the presented best solution and classi- V-shaped transfer functions. The proposed approach ran over
fied by KNN classifier. ISSA was utilized to twenty-three optimum path forest classifier and tested over several Arizona
UCI datasets and compared with basic SSA and four other State University’s datasets. Pourpanah et al. [224] used fuzzy
metaheuristic algorithms. Other versions of SSA was used for min-max neural network learning model with binary BSO
feature selection problems that can be found in [217], [218]. algorithm and applied to a real-world dataset. They also intro-
duced a fuzzy ARTMAP model with BSO algorithm [225].
12) EMPEROR PENGUINS OPTIMIZER Tuba et al. [226] used BSO algorithm with SVM classifier
EPO is based on the huddling behaviour of emperor penguins. for medical classification. Moreover, the parameters of SVM
Baliarsingh et al. [219] proposed multi-objective binary EPO were optimized using BSO algorithm. Oliva and Elaziz [227]
with chaos to apply on high-dimensional biomedical datasets. proposed a new version of BSO algorithm for its good
The proposed methodology used for cancer classification and exploration quality. They have introduced chaotic maps and
obtaining the optimal feature subset. The obtained results opposition based learning algorithm for initialization of the
show the efficiency of the proposed algorithm in terms of solution. Disruptor operator was used to updating the initial
accuracy, specificity and F-score. Baliarsingh and Vipsita population. The modified version was used for classification
[220] proposed an intelligence hybrid technique for gene and in order to obtain the optimal features, eight datasets have
classification in which extreme learning machine is used with been considered from UCI repository.
chaos EPO. The proposed hybrid technique is employed on
seven well known microarray datasets and the experimental 2) TEACHING BASED LEARNING OPTIMIZATION
results show the efficacy of the hybrid technique. To obtain TLBO algorithm is based on the influence of the teacher on
the optimal feature subset, Dhiman et al. [221] developed the students in the class. Krishna and Vishwakarma [228]
binary EPO (BEPO) with S and V-shaped transfer functions proposed an improved version of TLBO algorithm with
which convert the continuous search space to binary space. wavelet transform function for fingerprint recognition. Jain
The BEPO is applied to different feature selection datasets and Bhadauria [229] selected optimal features using TLBO
and the obtained results show better results as compared to algorithm and SVM classifier for image retrieval datasets.
others. Kiziloz et al. [230] presented a multi-objective TLBO algo-
rithm to select the features in binary classification problems.
C. PHYSICS-BASED ALGORITHMS The algorithm was tested over well-known datasets of UCI
Various algorithms have been developed that are based on with three supervised learning algorithm logistic regression,
the rules of physics. The binary versions of physics-based SVM, and extreme learning machine. Among all three classi-
algorithms which have been applied to feature selection fication model, logistic regression with TLBO presented best
problems are discussed in the following Table 10 and 11. results in most of the datasets. Allam and Nandhini [231]
In Tables 10 and 11, the first column represents the name developed a binary TLBO (BTLBO) with a threshold value to
of modified algorithms, the second column gives details of restrict variables into binary form. They have used different
the transfer function that used for deciding the binary vari- classifier and used for the classification of breast cancer
ables and the third column shows the name of the classifier datasets. The proposed approach showed its high accuracy
which operated as a learning algorithm in the optimiza- with the minimum number of features. The TLBO algorithm
tion process. Furthermore, the table describes the considered has been applied to chronic kidney disease dataset with the

VOLUME 9, 2021 26775

 P. Agrawal et al.: Metaheuristic Algorithms on Feature Selection: A Survey of One Decade of Research (2009-2019)

improved version [232]. The fitness function was evaluated Hence, to enhance the performance of whale optimization
using Chebyshev distance formula and obtained the results. algorithm, Mafarja and Mirjalili [237] embedded SA into
WOA. It boosted the exploitation of WOA by improving
3) JAYA ALGORITHM the best solution found after each iteration. The performance
Jaya algorithm is based on the framework of TLBO algorithm of the hybrid algorithm WOA-SA was tested over eighteen
with only one phase. It has been successfully applied to datasets with KNN classifier. Arora et al. [238] used the posi-
benchmark functions. Das et al. [233] modified JA to find tion updation quality of crow search algorithm in grey wolf
the optimal feature subset by using a search technique which optimizer to make a good balance between exploration and
updates the worst features. The proposed approach has been exploitation. It hybridized the algorithm as GWOCSA which
tested over ten benchmark datasets with other optimizer for applied to twenty-one well-known datasets of UCI repository.
comparison. The results show its efficacy to find the optimal THe GWOCSA algorithm restricted to binary search space
feature subset. Using the S-shaped transfer function with JA, using the S-shaped transfer function. The accuracy of con-
Awadallah et al. [234] developed a binary JA in which adap- sidered KNN classifier was compared with other state-of-
tive mutation rate has been used. The adaptive mutation rate the-art metaheuristic algorithms. To get rid of local optima
controls the diversification in the search space. The proposed in sine cosine algorithm, Abd Elaziz et al. [239] proposed
approach BJAM is applied to twenty two benchmark datasets a hybrid algorithm by employing the local search method
with KNN classifier and obtained the optimal feature subset. of differential evolution algorithm. The enhanced version of
sine cosine algorithm were tested over eight UCI datasets
4) GAINING SHARING KNOWLEDGE BASED ALGORITHM and presented better results in case of performance measures
It is based on the concept of gaining and sharing knowledge and statistical analysis. Tawhid and Dsouza [240] developed a
among humans. Agrawal et al. [235] proposed the first novel hybrid algorithm using bat algorithm and enhanced version of
binary version of GSK algorithm for feature selection prob- PSO algorithm for solving feature selection problem in binary
lem (FS-NBGSK) by introducing binary junior and senior space. To transform the position of bats in binary space, a V-
gaining and sharing stages. The FS-NBGSK algorithm was shaped transfer function was used, and in the same way,
tested over 23 benchmark datasets of UCI repository with an S-shaped transfer function was employed to get the binary
KNN classifier. The approach showed the best results among position of a particle in PSO algorithm. Hybrid algorithm
the compared algorithm in terms of accuracy and a minimum HBBEPSO combined good exploration of bat algorithm and
number of selected features. convergence characteristic of the PSO algorithm that obtained
the optimal features of twenty standard datasets. The obtained
E. HYBRID METAHEURISTIC ALGORITHMS results proved its ability to have high accuracy among all the
Hybrid metaheuristic algorithms mean to combine the compared algorithms. To select the gene for preprocessing
best operators from different metaheuristic algorithms and technique in cancer classification, binary black hole algo-
develop a new enhanced algorithm. In recent years, hybrid rithm was employed with PSO algorithm to enhance the
algorithms have been achieved great attention in solving exploration and exploitation efficiently and effectively [241].
optimization problems. In particular, for feature selection Different classifiers have been used to evaluate the perfor-
problem, many hybrid metaheuristic algorithms have been mance accuracy and obtained the results of two standard
developed to obtain the relevant and optimal feature subset and three clinical microarray datasets. Neggaz et al. [242]
from the original dataset. A lot of possibilities occur to pro- presented a hybrid algorithm to boost the salp swarm algo-
duce a more enhanced algorithm which finds the solution rithm with the help of sine cosine algorithm in solving the
optimally. The enhance algorithms help to get rid of trapping feature selection problem. Baliarsingh [243] embedded social
into local optima, free from premature convergence, explore engineering optimizer in EPO to enhance the performance of
the search space efficiently and effectively and make good EPO. The SVM classifier is modified with memetic algorithm
exploitation. Moreover, the enhanced algorithms obtain the and it is applied with the proposed hybrid approach to medical
optimal or near-optimal solution and make a better tradeoff datasets. The proposed hybrid algorithm is compared with
between exploration and exploitation quality of an algorithm. other well known metaheuristic algorithms and outperforms
There are some algorithms described in which the best qual- the other algorithms. Another hybrid approach of EPO is
ities of different algorithms were combined to develop a new proposed with the cultural algorithm for face recognition
one. [244]. The proposed approach enhances the capability of
Hafez et al. [236] proposed MAKHA algorithm in which the existing one and is applied with SVM classifier for face
the jump process was taken from monkey algorithm and recognition. It presents the best results in terms of conver-
evolutionary operators (mutation, crossover) were used from gence, robustness etc. With the use of sine cosine algorithm,
krill herd algorithm to find the optimal solution quickly. the exploration phase was enhanced and also avoided to get
The MAKHA algorithm was checked over eighteen datasets into premature convergence. To get the optimal gene from
of UCI with KNN classifier and obtained the classifica- gene expression data, Shukla et al. [245] combined teaching
tion accuracy. Simulated annealing (SA) is the most popular learning based optimization with SA algorithm. SA algorithm
and very promising algorithm from physics-based category. utilized to enhance the solution quality of TLBO algorithm

26776 VOLUME 9, 2021

P. Agrawal et al.: Metaheuristic Algorithms on Feature Selection: A Survey of One Decade of Research (2009-2019)

FIGURE 5. Role of different classifiers in solving feature selection problems.

and helped to find the relevant genes for the detection of subset of features for a different sample of datasets. For trying
cancer. Moreover, a new V-shaped transfer function was to find the best classification, feature selection algorithm
proposed to convert the variables into the binary variables. becomes unstable in most of the cases. Instability comes
The classification accuracy was evaluated with SVM clas- when there is a high correlation in the features, and they are
sifier and also tested over ten sets of microarray datasets. removed because of obtaining the best classification accu-
Several combinations of different metaheuristic algorithms racy. Therefore, stability is as important as the classification
have been developed in solving different application of fea- accuracy. The possible solution to make the algorithm stable
ture selection problem. Jaya algorithm is used with forest for feature selection problem can be found in [252], [253].
optimization algorithm in gene selection [246]. The use of
enhanced JA is to optimize the two parameters of forest opti- B. CHOICE OF CLASSIFIER
mization algorithm. This hybrid approach has been employed
To design a wrapper feature selection algorithm, the choice of
to the microarray datasets and outperformed the other opti-
a classifier has a great impact on the quality of the obtained
mizers. In the text feature selection, grey wolf optimizer and
solution. In solving feature selection problem using meta-
grasshopper optimization algorithm were employed [247],
heuristic algorithm, there are different types of classifiers
for industrial foam injection processes, PSO algorithm and
have been used such as K-nearest neighbour (KNN), Sup-
gravitational search algorithm were used [248]. A combina-
port Vector machine (SVM), Optimum Path Forest (OPF),
tion of the grey wolf and stochastic fractal search algorithm
Naive Bayesian (NB), Random Forest (RF), Artifical Neural
[249] and grasshopper and cat swarm optimization algorithm
Network (ANN), ID3, C4.5, Fuzzy rule based (FR), Kernel
are used for feature selection [250].
Extreme Learning Machine (KLM). The role of classifiers
used in feature selection problems is shown in Fig. 5. KNN is
IV. ISSUES AND CHALLANGES
most commonly used classifier with different datasets of UCI
Despite achieving great success of metaheuristic algorithms
repository. In contrast, the SVM classifier is used frequently
in solving feature selection problems, some challenges and
in intrusion detection systems and medical field datasets
issues occur that will be described in the following sections:
such as in cancer detection, artery disease etc. Additionally,
A. SCALABILITY AND STABILITY
Xue et al. [1] investigated that on medium-size datasets of
features SVM algorithm was adopted to present the promi-
In real-world problems, a dataset contains thousands or even
nent features with the time constraints. However, KNN clas-
millions of features. To handle the large datasets in feature
sifier is the most used classifier among all, and its benefits to
selection problem, the designed algorithm must be scalable.
applying for large dimensional datasets.
The designed algorithm must have a good scalable classifier
which handles large dataset [251]. Therefore, scalability is an
essential task for developing an algorithm to solve the feature C. CONSTRUCTION OF OBJECTIVE FUNCTION
selection problem. To select the best feature subset, a wrapper feature selection
Another important issue for designing an algorithm to algorithm optimizes a given objective function. The con-
solve the feature selection problem is stability. An algorithm struction of an objective function for feature selection varies
is said to be stable for feature selection if it finds the same according to the classification problem. Earlier, an objective

VOLUME 9, 2021 26777

 P. Agrawal et al.: Metaheuristic Algorithms on Feature Selection: A Survey of One Decade of Research (2009-2019)

function was formulated which contains either maximization (viii) Precision / Positive predictive value [256], [258]:
of classification accuracy, or the minimization of number Ratio of true positive observations and total predicted
of selected features. Besides, to combine both conflicting positive observations. i.e.
objectives, the multi-objective function was constructed in tP
solving the feature selection problem. The problem with the Precision = (4)
tP + fP
multi-objective function was converted into a single objective
by applying weights to both the objectives and performed (ix) F-score [257], [259]: It is a combination of Recall and
the learning algorithm. Many researchers [170], [185], [186], precision measures which provides a single score.
[191], [203], [235] have used multi-objective function and F-score is defined as a Harmonic mean of Recall and
obtained the best feature subset. Precision measures that can be formulated as
Moreover, the use of multi-objective function was very Precision ∗ Recall
F − Score = 2 (5)
effective and efficient to optimize the fitness function and find Precision + Recall
the best feature subset from the given datasets of features. (x) Matthew’s correlation coefficients (MCC) [260],
[261]: It describes the quality of binary classification
D. EVALUATION CRITERIA TO CHECK THE PERFORMANCE and mostly used in the field of bio-informatics. Based
In literature, there are lots of evaluation metrics that have on the above measures value, it can be formulated as
been used to investigate the performance of the wrapper fea-
ture selection algorithm. For example, sensitivity and speci- MCC
(tP × tN ) − (fP × fN )
ficity commonly used for medical classification, precision = √ (6)
and recall considered in computer science data classification, (tP + fP)(tP + fN )(tN + fP)(tN + fN )
the area under the curve used in radar signals. In general, there The value of MCC lies between −1 and 1, the value
are some other measures used to evaluate the performance 1 represents the perfect classification, 0 denotes for
of the algorithms. The most popular evaluation metrics are random prediction, and −1 represents the total dis-
reviewed and presented in detail. agreement of prediction and observations.
(i) True positive (tP): The actual observations are from (xi) Some general measure metrics are also used for
positive class and are estimated to be positive. checking the performance such as average fitness
(ii) True negative (tN): The actual observations are from value (objective function value), Worst and best fit-
negative class and are estimated by the model to be ness value, standard deviation of fitness values, aver-
negative. age number of selected features from the original
(iii) False positive (fP): When the model incorrectly esti- datasets. These performance measures are used in
mates observations in the positive class. [144], [262]–[269] to evaluate the performance.
(iv) False negative (fN): When the model incorrectly esti- • Averge fitness value: Assume Fi∗ be the optimal
mates the observations in the negative class. fitness value at ith run, then the average fitness
These above measures are used in [254], [255]. value represents the mean value of the fitness over
(v) Sensitivity/ True positive rate/ Recall [144], [256], total number of runs (Truns ). It can be formulated
[257]: It is the ratio of observations that are true mathematically as
positive and the total number of observations that are TX
runs
actually positive i.e. 1
AvgFitness = Fi∗ (7)
Truns
tP i=1
Recall = (1)
tP + fN • Average number of selected features: The selection
size of features is the ratio of number of selected
(vi) Specificity / True Negative rate (TNR) [144], [257], feauters and the total number of features in the orig-
[258]: It is the ration of the observations that are true inal datast. Mathematically, it can be represented as
negative and the total number of observation that are
TX
actually negative i.e. 1 runs
length(f )∗i
AvgFeature = (8)
tN Truns |S|
i=1
Specificity = (2)
tN + fP • Average computational time: Computational time
(vii) False positive rate (FPR) [255]: Ratio of false positive presents the running time to perform the k th algo-
observations and total predicted negative observa- rithm. The average of the running time is the mean
tions i.e. value of the time over total number of runs. It is
presented for the K th algorithms as
fP
FPR = (3) TX
runs
tN + fN 1
AvgkTime = Timeki (9)
Truns
It can also be defined as FPR = 1 − Specificity. i=1

26778 VOLUME 9, 2021

P. Agrawal et al.: Metaheuristic Algorithms on Feature Selection: A Survey of One Decade of Research (2009-2019)

V. CASE STUDY feature subset, the algorithms are run on the same platform
This section presents applications of feature selection in for which the following assumptions are made:
different datasets of machine learning. It describes the per-
Number of population size
formance of metaheuristic algorithms on feature selection (
datasets. 50, if Dimension < 20
=
1000, if Dimension ≥ 20
A. DATASETS Number(of function evaluations
To show the performance of metaheuristic algorithms on 5000, if Dimension < 20
feature selection datasets, eight datasets are adopted from the =
20000, if Dimension ≥ 20
UCI repository [134]. The datasets are of different dimen-
sions (number of features in a dataset). The datasets are Number(of runs
considered in which dimensions vary from 9 to 856 and 25, if Dimension < 20
=
number of instances are from 32 to 1593. As KNN is the most 10, if Dimension ≥ 20
preferred classifier, therefore, we consider KNN classifier to
evaluate the accuracy of selected feature subset. The three The four evaluation metrices are adopted to compare the per-
equal portion of a dataset is taken for training, testing and formance of the algorithms i.e. average fitness value, average
validation in a cross validation manner with K = 5. The classification accuracy, average number of selected features
description of datasets is shown in Table 5. and average computational time. The obtained results are
presented in Table 6-9. Table 6 presents the average fit-
TABLE 5. Datasets for the case study.
ness values of all algorithms for all datasets. Table 6 shows
that FS-NBGSK algorithm obtains minimum average fitness
values in most of the datasets. Specially, for large dimen-
sional datasets i.e. D8 , it shows the commandable results.

TABLE 6. Average fitness value of all algorithms.

B. RESULTS AND DISCUSSION

To evaluate the performance of metaheuristic algorithms on
the considered datasets, we considered five metaheuristic
algorithms i.e. binary particle swarm optimization (BPSO)
[270], binary differential evolution (BDE) [271], binary ant TABLE 7. Average accuracy of all algorithms.
lion optimizer (BALO) [183], binary grey wolf optimizer
(bGWO) [170] and binary gaining sharing knowledge based
algorithm (FS-NBGSK) [235]. The detailed methodologies
of the considered algorithms can be read from the correspond-
ing references. The main aim of feature selection problem is
to maximize the accuracy of the performance and minimize
the number of selected features. Therefore, a fitness function
is formed using the both criteria as
Number of selected features
min Z = ξ1 (Accuracy) + ξ2
Total number of features
TABLE 8. Average number of selected features from all algorithms.
(10)
where ξ1 and ξ2 are the coefficient of each criteria in which
ξ2 = 1 − ξ1 and ξ1 ∈ [0, 1]. The value of ξ1 is taken as
0.99 [183], [272]. The values of the parameters used in the
algorithms are as: In BDE, crossover probability = 0.95, c1
cognitive factor = 2, c2 social factor = 2; In BPSO, Wmax
maximum bound on inertia weight = 0.6, Wmin minimum
bound on inertia weight = 0.2; In FS-NBGSK, p probabil-
ity = 0.1, kr knowledge ratio = 0.95. To obtain the optimal

VOLUME 9, 2021 26779

 P. Agrawal et al.: Metaheuristic Algorithms on Feature Selection: A Survey of One Decade of Research (2009-2019)

FIGURE 6. Convergence graph of all algorithms.

TABLE 9. Average computational time taken by all algorithms. values of accuracy lies from 0 to 1. Table 7 presents the
average accuracy of all the considered algorithms. The sec-
ond main objective is to minimize the number of selected
features with maximum accuracy. Thus, the average number
of selected features are described in Table 8. It shows that
FS-NBGSK algorithm performs better than other algorithms.
To do the fair comparison, computational time is also con-
sidered as an evaluation metric. The average computational
time (in sec) taken by all algorithms is given in Table 9.
From the Table, it can be observed that FS-NBGSK
algorithm takes less computational time as compared to
The average accuracy of the algorithm is one of the most others.
important metrices which describes that how much accu- Moreover, convergence graph of all algorithms are
rate an algorithm performs over the selected features. The also drawn in Figure 6 for fitness values over the number of

26780 VOLUME 9, 2021

P. Agrawal et al.: Metaheuristic Algorithms on Feature Selection: A Survey of One Decade of Research (2009-2019)

TABLE 10. Details of physics based algorithm used for feature selection.

VOLUME 9, 2021 26781

 P. Agrawal et al.: Metaheuristic Algorithms on Feature Selection: A Survey of One Decade of Research (2009-2019)

TABLE 11. Details of physics based algorithm used for feature selection.

26782 VOLUME 9, 2021

P. Agrawal et al.: Metaheuristic Algorithms on Feature Selection: A Survey of One Decade of Research (2009-2019)

function evaluations. The figure shows the convergence VII. CONCLUSION

ability of the algorithms towards the optimal solution. This paper presents a comprehensive survey on metaheuristic
FS-NBGSK algorithm converges to the minimum fitness algorithms that are developed from 2009 to the 2019 year
value than the other algorithms. All other algorithms have and their binary variants, which have been applied to fea-
premature convergence or trap into local optima. ture selection problem. A detailed description and mathe-
matical model of feature selection problem are given that
VI. FUTURE WORKS could help researchers to understand the problem properly.
Based on the presented literature of metaheuristic algorithms Moreover, the techniques of solving feature selection prob-
and their binary versions for feature selection, the following lems are presented. Additionally, metaheuristic algorithms
research gaps are found. Interested researchers who are will- are considered in solving the feature selection problem.
ing to pursue their research in this field may consider these Therefore, basic definition, importance and the classifica-
findings: tion of metaheuristic algorithms are given. The evolution-
(1) Firstly, there are very less number of evolution based based, swarm-based, physics-based category, human rlated
and human behaviour related algorithms developed. New algorithms have been developed and applied to feature selec-
algorithms can be developed based on natural evolution tion problems. However, metaheuristic algorithms have some
and human activity. following drawbacks:
(2) No research has been done to propose the binary version
• They suffer from slow convergence rate due to random
of the following algorithms:
generation movement.
From swarm-based algorithms-bumblebees algorithm
• They explore the search space without knowing the
[43], paddy field algorithm [44], eagle strategy [51],
search direction.
Hierarchical swarm optimization [52], bird mating opti-
• They can trap into local optima, or they have some
mizer [58], Japanese tree frogs calling algorithm [61],
premature convergence.
the great salmon run algorithm [64], Egyptian vulture
• The values of the parameters used in the metaheuristic
optimization algorithm [67], animal migration optimiza-
algorithms have to be adjusted, this may also lead to
tion [69], shark smell optimization [70], spotted hyena
pre-mature convergence.
optimizer [84], emperor penguins colony [88].
From physics based algorithms- galaxy based search Besides, the limitation of the metaheuristic algorithms,
algorithm [91], curved space optimization [97], ray opti- the modified and enhanced version of the algorithms were
mization [98], lightning search algorithm [105], thermal developed which are successfully applied to the feature selec-
exchange optimization [111], find fix finish exploit ana- tion problems. Also, a categorization is presented based on
lyze [112]. the behaviour of algorithms; evolution-based, swarm-based,
From human related algorithm- league championship physics-related and human behaviour related algorithms.
algorithm [38], human inspired algorithm [113], social This paper benefits in such a way that a list of metaheuristic
emotional optimization [114]. algorithms is presented based on their classification. It also
(3) The algorithms mentioned above have not been applied benefits for the application point of view as it consists of
to solve the feature selection problem. Therefore, binary a case study. The case study presents the eight benchmark
variants of these algorithms can be implemented in clas- datasets and the optimal feature subsets are found by imple-
sification problems. menting different metaheuristic algorithms.
(4) In addition to developing binary variants of meta- It can be observed that very few algorithms are proposed
heuristic algorithms, many researchers have used well in the evolution and human-related category, but there are
known S and V-shaped transfer functions. New S and several algorithms have been designed in the swarm and
V-shaped transfer functions can be formulated and physics-related algorithms. It implies that there is a scope
used for designing the binary version of metaheuristic to develop or propose new metaheuristic algorithms in these
algorithms. categories. This paper mainly focuses on solving the fea-
(5) There is some less explored area(s) such as stock market ture selection problem using binary variants of metaheuristic
prediction, short term load forecasting, weather predic- algorithms. Hence, extensive literature is presented in every
tion, spam detection, Parkinson disease. These prob- class of metaheuristic algorithms. All binary variants of all
lems can be investigated further using metaheuristic reviewed algorithms regarding feature selection problems
algorithms. are pointed. In swarm-based category, all binary variants of
(6) In the literature mostly two objectives i.e. maximizing Cuckoo search, Bat algorithm, Firefly algorithm, flower pol-
the accuracy and minimizing the number of selected fea- lination algorithm, Krill herd algorithm, Grey wolf optimizer,
tures are considered. Besides these objectives, interested Ant lion optimizer, Dragonfly algorithm, Whale optimiza-
researchers can consider computational time, complex- tion algorithm, Grasshopper optimization algorithm, Salp
ity, stability and scalability in multi-objective in feature swarm algorithm are reviewed with the key factor of solving
selection. feature selection problem. Moreover, hybrid approaches are

VOLUME 9, 2021 26783

 P. Agrawal et al.: Metaheuristic Algorithms on Feature Selection: A Survey of One Decade of Research (2009-2019)

also reviewed in the process of solving the feature selection [9] S. Ahmed, M. Zhang, and L. Peng, ‘‘Enhanced feature selection for
problem. biomarker discovery in LC-MS data using GP,’’ in Proc. IEEE Congr.
Evol. Comput., Jun. 2013, pp. 584–591.
It can be concluded that there is some area(s) which are [10] M. H. Aghdam, N. Ghasem-Aghaee, and M. E. Basiri, ‘‘Text feature
less explored, such as spam detection, theft detection and selection using ant colony optimization,’’ Expert Syst. Appl., vol. 36, no. 3,
weather prediction. However, lots of research has been done pp. 6843–6853, Apr. 2009.
[11] A. Ghosh, A. Datta, and S. Ghosh, ‘‘Self-adaptive differential evolution
on the well-known datasets of UCI repository and in medical
for feature selection in hyperspectral image data,’’ Appl. Soft Comput.,
diagnosis (cancer classification), intrusion detection systems, vol. 13, no. 4, pp. 1969–1977, Apr. 2013.
text classification, multimedia etc. Hence, researchers should [12] M. Dash and H. Liu, ‘‘Feature selection for classification,’’ Intell. Data
pay great attention to explore this area with metaheuristic Anal., vol. 1, nos. 1–4, pp. 131–156, 1997.
[13] I. Guyon and A. Elisseeff, ‘‘An introduction to variable and feature
algorithms. Moreover, there are some algorithms in the liter- selection,’’ J. Mach. Learn. Res., vol. 3, pp. 1157–1182, Mar. 2003.
ature for which binary variants are not developed yet such as [14] H. Liu, H. Motoda, R. Setiono, and Z. Zhao, ‘‘Feature selection: An ever
PFA, CGS, TCO, ES, HSO, WSA, BMO, OptBees, TGSR, evolving frontier in data mining,’’ in Feature Selection in Data Mining.
Hyderabad, India, 2010, pp. 4–13.
EVOA, VCS, EPC, GbSA, CSO, WEO, LCA, EMA, VPL.
[15] N. Hoque, D. K. Bhattacharyya, and J. K. Kalita, ‘‘MIFS-ND: A mutual
These algorithms benefit classification after developing their information-based feature selection method,’’ Expert Syst. Appl., vol. 41,
binary version. From the literature, it can be observed that no. 14, pp. 6371–6385, Oct. 2014.
the researcher has to face many challenges to obtain the [16] Z. Xu, I. King, M. R.-T. Lyu, and R. Jin, ‘‘Discriminative semi-supervised
feature selection via manifold regularization,’’ IEEE Trans. Neural Netw.,
best feature subset of the considered classification problem. vol. 21, no. 7, pp. 1033–1047, Jul. 2010.
A good choice of classifier has a significant impact of the [17] J. Tang, S. Alelyani, and H. Liu, ‘‘Feature selection for classification:
quality of obtained solution such KNN classifier is the most A review,’’ in Data Classification: Algorithms and Applications. 2014,
p. 37.
used classifier in getting the best subset with well-known
[18] A. Jović, K. Brkić, and N. Bogunović, ‘‘A review of feature selection
datasets of UCI repository. After that, SVM classifier used to methods with applications,’’ in Proc. 38th Int. Conv. Inf. Commun. Tech-
classify in different applications such as medical diagnosis, nol., Electron. Microelectron. (MIPRO), 2015, pp. 1200–1205.
pattern recognition, image analysis etc. There are some other [19] Z. Sun, G. Bebis, and R. Miller, ‘‘Object detection using feature subset
selection,’’ Pattern Recognit., vol. 37, no. 11, pp. 2165–2176, Nov. 2004.
classifiers which are less used in terms of classification.
[20] A. K. Jain, R. P. W. Duin, and J. Mao, ‘‘Statistical pattern recognition:
Hence, this another gap to use different classifiers in classifi- A review,’’ IEEE Trans. Pattern Anal. Mach. Intell., vol. 22, no. 1,
cation problem and compared with most used ones. Finally, pp. 4–37, Jan. 2000.
researchers will get the benefit of this study as they could find [21] H. Liu and H. Motoda, Feature Extraction, Construction Selection:
A Data Mining Perspective, vol. 453. USA: Springer, 1998.
all the key factors in solving the feature selection problem [22] S. Mirjalili, S. M. Mirjalili, and A. Lewis, ‘‘Grey wolf optimizer,’’ Adv.
using metaheuristic algorithms under one roof. Eng. Softw., vol. 69, pp. 46–61, Mar. 2014.
[23] O. Olorunda and A. P. Engelbrecht, ‘‘Measuring exploration/exploitation
in particle swarms using swarm diversity,’’ in Proc. IEEE Congr. Evol.
ACKNOWLEDGMENT Comput. (IEEE World Congr. Comput. Intell.), Jun. 2008, pp. 1128–1134.
The authors would like to acknowledge the Editors and [24] L. Lin and M. Gen, ‘‘Auto-tuning strategy for evolutionary algorithms:
anonymous reviewers for providing their valuable comments Balancing between exploration and exploitation,’’ Soft Comput., vol. 13,
no. 2, pp. 157–168, Jan. 2009.
and suggestions. [25] A. W. Mohamed, A. A. Hadi, and A. K. Mohamed, ‘‘Gaining-sharing
knowledge based algorithm for solving optimization problems: A novel
REFERENCES nature-inspired algorithm,’’ Int. J. Mach. Learn. Cybern., vol. 11,
pp. 1501–1529, 2020.
[1] B. Xue, M. Zhang, W. N. Browne, and X. Yao, ‘‘A survey on evolutionary [26] J. H. Holland, ‘‘Genetic algorithms,’’ Sci. Amer., vol. 267, no. 1,
computation approaches to feature selection,’’ IEEE Trans. Evol. Com- pp. 66–73, 1992.
put., vol. 20, no. 4, pp. 606–626, Aug. 2016. [27] I. Rechenberg, ‘‘Evolutionsstrategien,’’ in Simulationsmethoden der
[2] P. Y. Lee, W. P. Loh, and J. F. Chin, ‘‘Feature selection in multimedia: Medizin und Biologie. Berlin, Germany: Springer, 1978, pp. 83–114.
The state-of-the-art review,’’ Image Vis. Comput., vol. 67, pp. 29–42, [28] J. H. Holland, Adaptation in Natural and Artificial Systems: An Intro-
Nov. 2017. ductory Analysis With Applications to Biology, Control, and Artificial
[3] B. Remeseiro and V. Bolon-Canedo, ‘‘A review of feature selection meth- Intelligence. Cambridge, MA, USA: MIT Press, 1992.
ods in medical applications,’’ Comput. Biol. Med., vol. 112, Sep. 2019, [29] F. Glover, ‘‘Future paths for integer programming and links to arti-
Art. no. 103375. ficial intelligence,’’ Comput. Oper. Res., vol. 13, no. 5, pp. 533–549,
[4] M. Sharma and P. Kaur, ‘‘A comprehensive analysis of nature-inspired Jan. 1986.
meta-heuristic techniques for feature selection problem,’’ Arch. Comput. [30] R. Storn and K. Price, ‘‘Differential evolution—A simple and efficient
Methods Eng., pp. 1–25, Feb. 2020, doi: 10.1007/s11831-020-09412-6. heuristic for global optimization over continuous spaces,’’ J. Global
[5] M. Z. Asghar, A. Khan, S. Ahmad, and F. M. Kundi, ‘‘A review of feature Optim., vol. 11, no. 4, pp. 341–359, 1997.
extraction in sentiment analysis,’’ J. Basic Appl. Sci. Res., vol. 4, no. 3, [31] J. Kennedy and R. Eberhart, ‘‘Particle swarm optimization,’’ in Proc.
pp. 181–186, 2014. IEEE Int. Conf. Neural Netw. (ICNN), vol. 4, Nov./Dec. 1995,
[6] Y. Saeys, I. Inza, and P. Larrañaga, ‘‘A review of feature selec- pp. 1942–1948.
tion techniques in bioinformatics,’’ Bioinformatics, vol. 23, no. 19, [32] M. Dorigo, V. Maniezzo, and A. Colorni, ‘‘Ant system: Optimization by
pp. 2507–2517, Oct. 2007. a colony of cooperating agents,’’ IEEE Trans. Syst., Man, Cybern. B,
[7] V. Bolón-Canedo and A. Alonso-Betanzos, ‘‘Ensembles for feature Cybern., vol. 26, no. 1, pp. 29–41, Feb. 1996.
selection: A review and future trends,’’ Inf. Fusion, vol. 52, pp. 1–12, [33] D. Karaboga, ‘‘An idea based on honey bee swarm for numerical
Dec. 2019. optimization,’’ Dept. Comput. Eng., Erciyes Univ., Kayseri, Turkey,
[8] H. Liu and L. Yu, ‘‘Toward integrating feature selection algorithms for Tech. Rep. 06, 2005.
classification and clustering,’’ IEEE Trans. Knowl. Data Eng., vol. 17, [34] R.-Q. Zhao and W.-S. Tang, ‘‘Monkey algorithm for global numerical
no. 4, pp. 491–502, Apr. 2005. optimization,’’ J. Uncertain Syst., vol. 2, no. 3, pp. 165–176, 2008.

26784 VOLUME 9, 2021

P. Agrawal et al.: Metaheuristic Algorithms on Feature Selection: A Survey of One Decade of Research (2009-2019)

[35] S. Kirkpatrick, C. D. Gelatt, and M. P. Vecchi, ‘‘Optimization by simu- [58] A. Askarzadeh and A. Rezazadeh, ‘‘A new heuristic optimization algo-
lated annealing,’’ Science, vol. 220, no. 4598, pp. 671–680, 1983. rithm for modeling of proton exchange membrane fuel cell: Bird mat-
[36] Z. Woo Geem, J. Hoon Kim, and G. V. Loganathan, ‘‘A new heuristic ing optimizer,’’ Int. J. Energy Res., vol. 37, no. 10, pp. 1196–1204,
optimization algorithm: Harmony search,’’ Simulation, vol. 76, no. 2, Aug. 2013.
pp. 60–68, Feb. 2001. [59] P. Civicioglu, ‘‘Artificial cooperative search algorithm for numerical
[37] R. V. Rao, V. J. Savsani, and D. P. Vakharia, ‘‘Teaching–learning-based optimization problems,’’ Inf. Sci., vol. 229, pp. 58–76, Apr. 2013.
optimization: An optimization method for continuous non-linear large [60] A. H. Gandomi and A. H. Alavi, ‘‘Krill herd: A new bio-inspired opti-
scale problems,’’ Inf. Sci., vol. 183, no. 1, pp. 1–15, Jan. 2012. mization algorithm,’’ Commun. Nonlinear Sci. Numer. Simul., vol. 17,
[38] A. H. Kashan, ‘‘League championship algorithm: A new algorithm for no. 12, pp. 4831–4845, Dec. 2012.
numerical function optimization,’’ in Proc. Int. Conf. Soft Comput. Pat- [61] H. Hernández and C. Blum, ‘‘Distributed graph coloring: An approach
tern Recognit., 2009, pp. 43–48. based on the calling behavior of Japanese tree frogs,’’ Swarm Intell.,
[39] P. Civicioglu, ‘‘Transforming geocentric Cartesian coordinates to geode- vol. 6, no. 2, pp. 117–150, Jun. 2012.
tic coordinates by using differential search algorithm,’’ Comput. Geosci., [62] R. D. Maia, L. N. de Castro, and W. M. Caminhas, ‘‘Bee colonies as
vol. 46, pp. 229–247, Sep. 2012. model for multimodal continuous optimization: The OptBees algorithm,’’
[40] P. Civicioglu, ‘‘Backtracking search optimization algorithm for numer- in Proc. IEEE Congr. Evol. Comput., Jun. 2012, pp. 1–8.
ical optimization problems,’’ Appl. Math. Comput., vol. 219, no. 15, [63] R. Tang, S. Fong, X.-S. Yang, and S. Deb, ‘‘Wolf search algorithm with
pp. 8121–8144, Apr. 2013. ephemeral memory,’’ in Proc. 7th Int. Conf. Digit. Inf. Manage. (ICDIM),
[41] H. Salimi, ‘‘Stochastic fractal search: A powerful metaheuristic algo- Aug. 2012, pp. 165–172.
rithm,’’ Knowl.-Based Syst., vol. 75, pp. 1–18, Feb. 2015. [64] A. Mozaffari, A. Fathi, and S. Behzadipour, ‘‘The great salmon run:
[42] T. T. Dhivyaprabha, P. Subashini, and M. Krishnaveni, ‘‘Synergistic A novel bio-inspired algorithm for artificial system design and opti-
fibroblast optimization: A novel nature-inspired computing algorithm,’’ misation,’’ Int. J. Bio-Inspired Comput., vol. 4, no. 5, pp. 286–301,
Frontiers Inf. Technol. Electron. Eng., vol. 19, no. 7, pp. 815–833, 2012.
Jul. 2018. [65] A. Kaveh and N. Farhoudi, ‘‘A new optimization method: Dolphin echolo-
[43] F. Comellas and J. Martinez-Navarro, ‘‘Bumblebees: A multiagent com- cation,’’ Adv. Eng. Softw., vol. 59, pp. 53–70, May 2013.
binatorial optimization algorithm inspired by social insect behaviour,’’ [66] M. Neshat, G. Sepidnam, and M. Sargolzaei, ‘‘Swallow swarm optimiza-
in Proc. 1st ACM/SIGEVO Summit Genet. Evol. Comput., 2009, tion algorithm: A new method to optimization,’’ Neural Comput. Appl.,
pp. 811–814. vol. 23, no. 2, pp. 429–454, Aug. 2013.
[44] U. Premaratne, J. Samarabandu, and T. Sidhu, ‘‘A new biologically [67] C. Sur, S. Sharma, and A. Shukla, ‘‘Egyptian vulture optimization
inspired optimization algorithm,’’ in Proc. Int. Conf. Ind. Inf. Syst. (ICIIS), algorithm—A new nature inspired meta-heuristics for knapsack prob-
Dec. 2009, pp. 279–284. lem,’’ in Proc. 9th Int. Conf. Comput. Inf. Technol. (IC2IT). Berlin,
[45] X.-S. Yang and S. Deb, ‘‘Cuckoo search via Lévy flights,’’ in Germany: Springer, 2013, pp. 227–237.
Proc. World Congr. Nature Biol. Inspired Comput. (NaBIC), 2009, [68] X. Meng, Y. Liu, X. Gao, and H. Zhang, ‘‘A new bio-inspired algorithm:
pp. 210–214. Chicken swarm optimization,’’ in Proc. Int. Conf. Swarm Intell. Cham,
[46] S. He, Q. H. Wu, and J. R. Saunders, ‘‘Group search optimizer: An opti- Switzerland: Springer, 2014, pp. 86–94.
mization algorithm inspired by animal searching behavior,’’ IEEE Trans. [69] X. Li, J. Zhang, and M. Yin, ‘‘Animal migration optimization: An opti-
Evol. Comput., vol. 13, no. 5, pp. 973–990, Oct. 2009. mization algorithm inspired by animal migration behavior,’’ Neural Com-
[47] S. Iordache, ‘‘Consultant-guided search: A new Metaheuristic for combi- put. Appl., vol. 24, nos. 7–8, pp. 1867–1877, Jun. 2014.
natorial optimization problems,’’ in Proc. 12th Annu. Conf. Genetic Evol. [70] O. Abedinia, N. Amjady, and A. Ghasemi, ‘‘A new metaheuristic algo-
Comput., 2010, pp. 225–232. rithm based on shark smell optimization,’’ Complexity, vol. 21, no. 5,
[48] X.-S. Yang, ‘‘A new metaheuristic bat-inspired algorithm,’’ in Nature pp. 97–116, May 2016.
Inspired Cooperative Strategies for Optimization. Berlin, Germany: [71] S. Mirjalili, ‘‘The ant lion optimizer,’’ Adv. Eng. Softw., vol. 83, pp. 80–98,
Springer, 2010, pp. 65–74. May 2015.
[49] R. Hedayatzadeh, F. Akhavan Salmassi, M. Keshtgari, R. Akbari, and [72] X.-B. Meng, X. Z. Gao, L. Lu, Y. Liu, and H. Zhang, ‘‘A new bio-inspired
K. Ziarati, ‘‘Termite colony optimization: A novel approach for opti- optimisation algorithm: Bird swarm algorithm,’’ J. Experim. Theor. Artif.
mizing continuous problems,’’ in Proc. 18th Iranian Conf. Electr. Eng., Intell., vol. 28, no. 4, pp. 673–687, Jul. 2016.
May 2010, pp. 553–558. [73] M. D. Li, H. Zhao, X. W. Weng, and T. Han, ‘‘A novel nature-inspired
[50] R. Oftadeh, M. J. Mahjoob, and M. Shariatpanahi, ‘‘A novel meta- algorithm for optimization: Virus colony search,’’ Adv. Eng. Softw.,
heuristic optimization algorithm inspired by group hunting of animals: vol. 92, pp. 65–88, Feb. 2016.
Hunting search,’’ Comput. Math. Appl., vol. 60, no. 7, pp. 2087–2098, [74] S. A. Uymaz, G. Tezel, and E. Yel, ‘‘Artificial algae algorithm (AAA) for
Oct. 2010. nonlinear global optimization,’’ Appl. Soft Comput., vol. 31, pp. 153–171,
[51] X.-S. Yang and S. Deb, ‘‘Eagle strategy using lévy walk and fire- Jun. 2015.
fly algorithms for stochastic optimization,’’ in Nature Inspired Coop- [75] S. Mirjalili, ‘‘Dragonfly algorithm: A new meta-heuristic optimization
erative Strategies for Optimization. Berlin, Germany: Springer, 2010, technique for solving single-objective, discrete, and multi-objective prob-
pp. 101–111. lems,’’ Neural Comput. Appl., vol. 27, no. 4, pp. 1053–1073, May 2016.
[52] H. Chen, Y. Zhu, K. Hu, and X. He, ‘‘Hierarchical swarm model: [76] W. Yong, W. Tao, Z. Cheng-Zhi, and H. Hua-Juan, ‘‘A new stochastic
A new approach to optimization,’’ Discrete Dyn. Nature Soc., vol. 2010, optimization approach—Dolphin swarm optimization algorithm,’’ Int.
pp. 1–30, Jan. 2010. J. Comput. Intell. Appl., vol. 15, no. 2, 2016, Art. no. 1650011.
[53] X-S. Yang, ‘‘Firefly algorithm, stochastic test functions and design [77] A. Askarzadeh, ‘‘A novel Metaheuristic method for solving constrained
optimisation,’’ Int. J. Bio-Inspired Comput., vol. 2, no. 2, pp. 78–84, engineering optimization problems: Crow search algorithm,’’ Comput.
2010. Struct., vol. 169, pp. 1–12, Jun. 2016.
[54] W.-T. Pan, ‘‘A new fruit fly optimization algorithm: Taking the financial [78] S. Mirjalili and A. Lewis, ‘‘The whale optimization algorithm,’’ Adv. Eng.
distress model as an example,’’ Knowl.-Based Syst., vol. 26, pp. 69–74, Softw., vol. 95, pp. 51–67, May 2016.
Feb. 2012. [79] E. Jahani and M. Chizari, ‘‘Tackling global optimization problems with
[55] R. S. Parpinelli and H. S. Lopes, ‘‘An eco-inspired evolutionary algorithm a novel algorithm–mouth brooding fish algorithm,’’ Appl. Soft Comput.,
applied to numerical optimization,’’ in Proc. 3rd World Congr. Nature vol. 62, pp. 987–1002, Jan. 2018.
Biol. Inspired Comput., Oct. 2011, pp. 466–471. [80] X. Qi, Y. Zhu, and H. Zhang, ‘‘A new meta-heuristic butterfly-inspired
[56] T. Ting, K. L. Man, S.-U. Guan, M. Nayel, and K. Wan, ‘‘Weightless algorithm,’’ J. Comput. Sci., vol. 23, pp. 226–239, Nov. 2017.
swarm algorithm (WSA) for dynamic optimization problems,’’ in Proc. [81] F. Fausto, E. Cuevas, A. Valdivia, and A. González, ‘‘A global optimiza-
IFIP Int. Conf. Netw. Parallel Comput. Berlin, Germany: Springer, 2012, tion algorithm inspired in the behavior of selfish herds,’’ Biosystems,
pp. 508–515. vol. 160, pp. 39–55, Oct. 2017.
[57] X.-S. Yang, ‘‘Flower pollination algorithm for global optimization,’’ in [82] S. Saremi, S. Mirjalili, and A. Lewis, ‘‘Grasshopper optimisation algo-
Proc. Int. Conf. Unconv. Comput. Natural Comput. Berlin, Germany: rithm: Theory and application,’’ Adv. Eng. Softw., vol. 105, pp. 30–47,
Springer, 2012, pp. 240–249. Mar. 2017.

VOLUME 9, 2021 26785

 P. Agrawal et al.: Metaheuristic Algorithms on Feature Selection: A Survey of One Decade of Research (2009-2019)

[83] S. Mirjalili, A. H. Gandomi, S. Z. Mirjalili, S. Saremi, H. Faris, and [107] A. Kaveh and T. Bakhshpoori, ‘‘Water evaporation optimization: A novel
S. M. Mirjalili, ‘‘Salp swarm algorithm: A bio-inspired optimizer for physically inspired optimization algorithm,’’ Comput. Struct., vol. 167,
engineering design problems,’’ Adv. Eng. Softw., vol. 114, pp. 163–191, pp. 69–85, Apr. 2016.
Dec. 2017. [108] S. Mirjalili, S. M. Mirjalili, and A. Hatamlou, ‘‘Multi-verse optimizer:
[84] G. Dhiman and V. Kumar, ‘‘Spotted hyena optimizer: A novel bio- A nature-inspired algorithm for global optimization,’’ Neural Comput.
inspired based Metaheuristic technique for engineering applications,’’ Appl., vol. 27, no. 2, pp. 495–513, Feb. 2016.
Adv. Eng. Softw., vol. 114, pp. 48–70, Dec. 2017. [109] A. F. Nematollahi, A. Rahiminejad, and B. Vahidi, ‘‘A novel physical
[85] G. Dhiman and V. Kumar, ‘‘Emperor penguin optimizer: A bio-inspired based meta-heuristic optimization method known as lightning attach-
algorithm for engineering problems,’’ Knowl.-Based Syst., vol. 159, ment procedure optimization,’’ Appl. Soft Comput., vol. 59, pp. 596–621,
pp. 20–50, Nov. 2018. Oct. 2017.
[86] M. Jain, V. Singh, and A. Rani, ‘‘A novel nature-inspired algorithm for [110] A. Tabari and A. Ahmad, ‘‘A new optimization method: Electro-search
optimization: Squirrel search algorithm,’’ Swarm Evol. Comput., vol. 44, algorithm,’’ Comput. Chem. Eng., vol. 103, pp. 1–11, Aug. 2017.
pp. 148–175, Feb. 2019. [111] A. Kaveh and A. Dadras, ‘‘A novel meta-heuristic optimization algorithm:
[87] S. Arora and S. Singh, ‘‘Butterfly optimization algorithm: A novel Thermal exchange optimization,’’ Adv. Eng. Softw., vol. 110, pp. 69–84,
approach for global optimization,’’ Soft Comput., vol. 23, no. 3, Aug. 2017.
pp. 715–734, Feb. 2019. [112] A. Husseinzadeh Kashan, R. Tavakkoli-Moghaddam, and M. Gen, ‘‘Find-
[88] S. Harifi, M. Khalilian, J. Mohammadzadeh, and S. Ebrahimnejad, fix-finish-exploit-analyze (F3EA) meta-heuristic algorithm: An effective
‘‘Emperor penguins colony: A new Metaheuristic algorithm for optimiza- algorithm with new evolutionary operators for global optimization,’’
tion,’’ Evol. Intell., vol. 12, no. 2, pp. 211–226, Jun. 2019. Comput. Ind. Eng., vol. 128, pp. 192–218, Feb. 2019.
[89] E. Rashedi, H. Nezamabadi-pour, and S. Saryazdi, ‘‘GSA: A gravitational [113] L. M. Zhang, C. Dahlmann, and Y. Zhang, ‘‘Human-inspired algorithms
search algorithm,’’ Inf. Sci., vol. 179, no. 13, pp. 2232–2248, Jun. 2009. for continuous function optimization,’’ in Proc. IEEE Int. Conf. Intell.
[90] A. Kaveh and S. Talatahari, ‘‘A novel heuristic optimization method: Comput. Intell. Syst., Nov. 2009, pp. 318–321.
Charged system search,’’ Acta Mechanica, vol. 213, nos. 3–4, [114] Y. Xu, Z. Cui, and J. Zeng, ‘‘Social emotional optimization algo-
pp. 267–289, Sep. 2010. rithm for nonlinear constrained optimization problems,’’ in Proc. Int.
[91] H. Shah-Osseini, ‘‘Principal components analysis by the galaxy-based Conf. Swarm, Evol., Memetic Comput. Berlin, Germany: Springer, 2010,
search algorithm: A novel metaheuristic for continuous optimisation,’’ pp. 583–590.
Int. J. Comput. Sci. Eng., vol. 6, nos. 1–2, pp. 132–140, 2011. [115] Y. Shi, ‘‘Brain storm optimization algorithm,’’ in Proc. Int. Conf. Swarm
[92] E. Cuevas, D. Oliva, D. Zaldivar, M. Pérez-Cisneros, and H. Sossa, ‘‘Cir- Intell. Berlin, Germany: Springer, 2011, pp. 303–309.
cle detection using electro-magnetism optimization,’’ Inf. Sci., vol. 182, [116] A. Ahmadi-Javid, ‘‘Anarchic society optimization: A human-inspired
no. 1, pp. 40–55, Jan. 2012. method,’’ in Proc. IEEE Congr. Evol. Comput. (CEC), Jun. 2011,
[93] B. Alatas, ‘‘ACROA: Artificial chemical reaction optimization algo- pp. 2586–2592.
rithm for global optimization,’’ Expert Syst. Appl., vol. 38, no. 10, [117] N. Moosavian and B. Kasaee Roodsari, ‘‘Soccer league competition
pp. 13170–13180, Sep. 2011. algorithm: A novel meta-heuristic algorithm for optimal design of
[94] K. Tamura and K. Yasuda, ‘‘Spiral optimization—A new multipoint water distribution networks,’’ Swarm Evol. Comput., vol. 17, pp. 14–24,
search method,’’ in Proc. IEEE Int. Conf. Syst., Man, Cybern., Oct. 2011, Aug. 2014.
pp. 1759–1764. [118] F. Ramezani and S. Lotfi, ‘‘Social-based algorithm (SBA),’’ Appl. Soft
[95] A. Hatamlou, ‘‘Black hole: A new heuristic optimization approach for Comput., vol. 13, no. 5, pp. 2837–2856, May 2013.
data clustering,’’ Inf. Sci., vol. 222, pp. 175–184, Feb. 2013. [119] N. Ghorbani and E. Babaei, ‘‘Exchange market algorithm,’’ Appl. Soft
[96] H. Eskandar, A. Sadollah, A. Bahreininejad, and M. Hamdi, ‘‘Water Comput., vol. 19, pp. 177–187, Jun. 2014.
cycle algorithm—A novel Metaheuristic optimization method for solv- [120] M. A. Eita and M. M. Fahmy, ‘‘Group counseling optimization,’’ Appl.
ing constrained engineering optimization problems,’’ Comput. Struct., Soft Comput., vol. 22, pp. 585–604, Sep. 2014.
vols. 110–111, pp. 151–166, Nov. 2012. [121] R. V. Rao, ‘‘Jaya: A simple and new optimization algorithm for solving
[97] F. Farrahi Moghaddam, R. Farrahi Moghaddam, and M. Cheriet, ‘‘Curved constrained and unconstrained optimization problems,’’ Int. J. Ind. Eng.
space optimization: A random search based on general relativity theory,’’ Computations, vol. 7, no. 1, pp. 19–34, 2016.
2012, arXiv:1208.2214. [Online]. Available: http://arxiv.org/abs/1208. [122] R. Moghdani and K. Salimifard, ‘‘Volleyball premier league algorithm,’’
2214 Appl. Soft Comput., vol. 64, pp. 161–185, Mar. 2018.
[98] A. Kaveh and M. Khayatazad, ‘‘A new meta-heuristic method: [123] L. Gan and H. Duan, ‘‘Biological image processing via chaotic differen-
Ray optimization,’’ Comput. Struct., vols. 112–113, pp. 283–294, tial search and lateral inhibition,’’ Optik, vol. 125, no. 9, pp. 2070–2075,
Dec. 2012. May 2014.
[99] A. Sadollah, A. Bahreininejad, H. Eskandar, and M. Hamdi, ‘‘Mine blast [124] M. Negahbani, S. Joulazadeh, H. R. Marateb, and M. Mansourian, ‘‘Coro-
algorithm: A new population based algorithm for solving constrained nary artery disease diagnosis using supervised fuzzy C-means with dif-
engineering optimization problems,’’ Appl. Soft Comput., vol. 13, no. 5, ferential search algorithm-based generalized Minkowski metrics,’’ Arch.
pp. 2592–2612, May 2013. Biomed. Sci. Eng., vol. 1, pp. 6–14, Jun. 2015.
[100] M. Abdechiri, M. R. Meybodi, and H. Bahrami, ‘‘Gases Brownian motion [125] C. Zhang, J. Zhou, C. Li, W. Fu, and T. Peng, ‘‘A compound structure of
optimization: An algorithm for optimization (GBMO),’’ Appl. Soft Com- ELM based on feature selection and parameter optimization using hybrid
put., vol. 13, no. 5, pp. 2932–2946, May 2013. backtracking search algorithm for wind speed forecasting,’’ Energy Con-
[101] G.-W. Yan and Z.-J. Hao, ‘‘A novel optimization algorithm based on vers. Manage., vol. 143, pp. 360–376, Jul. 2017.
atmosphere clouds model,’’ Int. J. Comput. Intell. Appl., vol. 12, no. 1, [126] K. G. Dhal, J. Gálvez, S. Ray, A. Das, and S. Das, ‘‘Acute lymphoblastic
Mar. 2013, Art. no. 1350002. leukemia image segmentation driven by stochastic fractal search,’’ Mul-
[102] S. Moein and R. Logeswaran, ‘‘KGMO: A swarm optimization algo- timedia Tools Appl., vol. 79, pp. 12227–12255, 2020.
rithm based on the kinetic energy of gas molecules,’’ Inf. Sci., vol. 275, [127] K. M. Hosny, M. A. Elaziz, I. M. Selim, and M. M. Darwish, ‘‘Classifica-
pp. 127–144, Aug. 2014. tion of galaxy color images using quaternion polar complex exponential
[103] A. Kaveh and V. R. Mahdavi, ‘‘Colliding bodies optimization: A novel transform and binary stochastic fractal search,’’ Astron. Comput., vol. 31,
meta-heuristic method,’’ Comput. Struct., vol. 139, pp. 18–27, Jul. 2014. Apr. 2020, Art. no. 100383.
[104] A. Baykasoğlu and Ş. Akpinar, ‘‘Weighted superposition attraction [128] V. Tiwari, ‘‘Face recognition based on cuckoo search algorithm,’’ Image,
(WSA): A swarm intelligence algorithm for optimization problems— vol. 7, no. 8, p. 9, 2012.
Part 1: Unconstrained optimization,’’ Appl. Soft Comput., vol. 56, [129] D. Rodrigues, L. A. M. Pereira, T. N. S. Almeida, J. P. Papa, A. N. Souza,
pp. 520–540, Jul. 2017. C. C. O. Ramos, and X.-S. Yang, ‘‘BCS: A binary cuckoo search algo-
[105] H. Shareef, A. A. Ibrahim, and A. H. Mutlag, ‘‘Lightning search algo- rithm for feature selection,’’ in Proc. IEEE Int. Symp. Circuits Syst.
rithm,’’ Appl. Soft Comput., vol. 36, pp. 315–333, Nov. 2015. (ISCAS), May 2013, pp. 465–468.
[106] S. Mirjalili, ‘‘SCA: A sine cosine algorithm for solving [130] C. Gunavathi and K. Premalatha, ‘‘Cuckoo search optimisation for feature
optimization problems,’’ Knowl.-Based Syst., vol. 96, pp. 120–133, selection in cancer classification: A new approach,’’ Int. J. Data Mining
Mar. 2016. Bioinf., vol. 13, no. 3, pp. 248–265, 2015.

26786 VOLUME 9, 2021

P. Agrawal et al.: Metaheuristic Algorithms on Feature Selection: A Survey of One Decade of Research (2009-2019)

[131] M. N. Sudha and S. Selvarajan, ‘‘Feature selection based on enhanced [152] Y. Zhang, X.-F. Song, and D.-W. Gong, ‘‘A return-cost-based binary fire-
cuckoo search for breast cancer classification in mammogram image,’’ fly algorithm for feature selection,’’ Inf. Sci., vols. 418–419, pp. 561–574,
Circuits Syst., vol. 07, no. 04, pp. 327–338, 2016. Dec. 2017.
[132] S. Salesi and G. Cosma, ‘‘A novel extended binary cuckoo search algo- [153] L. Zhang, L. Shan, and J. Wang, ‘‘Optimal feature selection using
rithm for feature selection,’’ in Proc. 2nd Int. Conf. Knowl. Eng. Appl. distance-based discrete firefly algorithm with mutual information
(ICKEA), Oct. 2017, pp. 6–12. criterion,’’ Neural Comput. Appl., vol. 28, no. 9, pp. 2795–2808,
[133] A. C. Pandey, D. S. Rajpoot, and M. Saraswat, ‘‘Feature selection Sep. 2017.
method based on hybrid data transformation and binary binomial cuckoo [154] H. Xu, S. Yu, J. Chen, and X. Zuo, ‘‘An improved firefly algorithm for
search,’’ J. Ambient Intell. Hum. Comput., vol. 11, no. 2, pp. 719–738, feature selection in classification,’’ Wireless Pers. Commun., vol. 102,
Feb. 2020. no. 4, pp. 2823–2834, Oct. 2018.
[134] A. Frank. (2010). UCI Machine Learning Repository. [Online]. Available: [155] B. Selvakumar and K. Muneeswaran, ‘‘Firefly algorithm based feature
http://archive.ics.uci.edu/ml selection for network intrusion detection,’’ Comput. Secur., vol. 81,
[135] S. Sarvari, N. F. Mohd Sani, Z. Mohd Hanapi, and M. T. Abdullah, pp. 148–155, Mar. 2019.
‘‘An efficient anomaly intrusion detection method with feature selection [156] A. Kazem, E. Sharifi, F. K. Hussain, M. Saberi, and O. K. Hussain, ‘‘Sup-
and evolutionary neural network,’’ IEEE Access, vol. 8, pp. 70651–70663, port vector regression with chaos-based firefly algorithm for stock market
2020. price forecasting,’’ Appl. Soft Comput., vol. 13, no. 2, pp. 947–958,
[136] R. P. Shetty, A. Sathyabhama, and P. S. Pai, ‘‘An efficient online Feb. 2013.
sequential extreme learning machine model based on feature selection [157] L. Zhang, K. Mistry, C. P. Lim, and S. C. Neoh, ‘‘Feature selection using
and parameter optimization using cuckoo search algorithm for multi- firefly optimization for classification and regression models,’’ Decis.
step wind speed forecasting,’’ Soft Comput., vol. 25, pp. 1277–1295, Support Syst., vol. 106, pp. 64–85, Feb. 2018.
2021. [158] R. R. Chhikara, P. Sharma, and L. Singh, ‘‘An improved dynamic discrete
[137] S. Marso and M. El Merouani, ‘‘Predicting financial distress using hybrid firefly algorithm for blind image steganalysis,’’ Int. J. Mach. Learn.
feedforward neural network with cuckoo search algorithm,’’ Procedia Cybern., vol. 9, no. 5, pp. 821–835, May 2018.
Comput. Sci., vol. 170, pp. 1134–1140, 2020. [159] S. Dash, R. Thulasiram, and P. Thulasiraman, ‘‘Modified firefly algorithm
[138] R. Y. M. Nakamura, L. A. M. Pereira, K. A. Costa, D. Rodrigues, with chaos theory for feature selection: A predictive model for medical
J. P. Papa, and X.-S. Yang, ‘‘BBA: A binary bat algorithm for feature data,’’ Int. J. Swarm Intell. Res., vol. 10, no. 2, pp. 1–20, Apr. 2019.
selection,’’ in Proc. 25th SIBGRAPI Conf. Graph., Patterns Images, [160] D. Rodrigues, X.-S. Yang, A. N. De Souza, and J. P. Papa, ‘‘Binary flower
Aug. 2012, pp. 291–297. pollination algorithm and its application to feature selection,’’ in Recent
[139] M. A. Laamari and N. Kamel, ‘‘A hybrid bat based feature selection Advances in Swarm Intelligence and Evolutionary Computation. Cham,
approach for intrusion detection,’’ in Bio-Inspired Computing-Theories Switzerland: Springer, 2015, pp. 85–100.
and Applications. Berlin, Germany: Springer, 2014, pp. 230–238. [161] H. M. Zawbaa and E. Emary, ‘‘Applications of flower pollination algo-
[140] D. Rodrigues, L. A. M. Pereira, R. Y. M. Nakamura, K. A. P. Costa, rithm in feature selection and knapsack problems,’’ in Nature-Inspired
X.-S. Yang, A. N. Souza, and J. P. Papa, ‘‘A wrapper approach for feature Algorithms and Applied Optimization. Cham, Switzerland: Springer,
selection based on bat algorithm and optimum-path forest,’’ Expert Syst. 2018, pp. 217–243.
Appl., vol. 41, no. 5, pp. 2250–2258, Apr. 2014. [162] S. P. Rajamohana, K. Umamaheswari, and B. Abirami, ‘‘Adaptive binary
[141] A.-C. Enache and V. Sgârciu, ‘‘A feature selection approach implemented flower pollination algorithm for feature selection in review spam detec-
with the binary bat algorithm applied for intrusion detection,’’ in Proc. tion,’’ in Proc. Int. Conf. Innov. Green Energy Healthcare Technol.
38th Int. Conf. Telecommun. Signal Process. (TSP), 2015, pp. 11–15. (IGEHT), 2017, pp. 1–4.
[142] B. Yang, Y. Lu, K. Zhu, G. Yang, J. Liu, and H. Yin, ‘‘Feature selection [163] H. Majidpour and F. Soleimanian Gharehchopogh, ‘‘An improved flower
based on modified bat algorithm,’’ IEICE Trans. Inf. Syst., vol. E100.D, pollination algorithm with AdaBoost algorithm for feature selection in
no. 8, pp. 1860–1869, 2017. text documents classification,’’ J. Adv. Comput. Res., vol. 9, no. 1,
[143] T. Kaur, B. S. Saini, and S. Gupta, ‘‘A novel feature selection method for pp. 29–40, 2018.
brain tumor MR image classification based on the Fisher criterion and [164] S. A.-F. Sayed, E. Nabil, and A. Badr, ‘‘A binary clonal flower polli-
parameter-free bat optimization,’’ Neural Comput. Appl., vol. 29, no. 8, nation algorithm for feature selection,’’ Pattern Recognit. Lett., vol. 77,
pp. 193–206, Apr. 2018. pp. 21–27, Jul. 2016.
[144] A. K. Naik, V. Kuppili, and D. R. Edla, ‘‘Efficient feature selection [165] C. Yan, J. Ma, H. Luo, G. Zhang, and J. Luo, ‘‘A novel feature selection
using one-pass generalized classifier neural network and binary bat method for high-dimensional biomedical data based on an improved
algorithm with a novel fitness function,’’ Soft Comput., vol. 24, no. 6, binary clonal flower pollination algorithm,’’ Hum. Heredity, vol. 84, no. 1,
pp. 4575–4587, Mar. 2020. pp. 1–13, 2019.
[145] B. Alsalibi, I. Venkat, and M. A. Al-Betar, ‘‘A membrane-inspired bat [166] D. Rodrigues, L. A. M. Pereira, J. P. Papa, and S. A. T. Weber, ‘‘A binary
algorithm to recognize faces in unconstrained scenarios,’’ Eng. Appl. krill herd approach for feature selection,’’ in Proc. 22nd Int. Conf. Pattern
Artif. Intell., vol. 64, pp. 242–260, Sep. 2017. Recognit., Aug. 2014, pp. 1407–1412.
[146] S.-W. Fei, ‘‘Fault diagnosis of bearing based on relevance vector machine [167] A. Mohammadi, M. S. Abadeh, and H. Keshavarz, ‘‘Breast can-
classifier with improved binary bat algorithm for feature selection cer detection using a multi-objective binary krill herd algorithm,’’
and parameter optimization,’’ Adv. Mech. Eng., vol. 9, no. 1, 2017, in Proc. 21th Iranian Conf. Biomed. Eng. (ICBME), Nov. 2014,
Art. no. 1687814016685294. pp. 128–133.
[147] S. Jeyasingh and M. Veluchamy, ‘‘Modified bat algorithm for fea- [168] R. R. Rani and D. Ramyachitra, ‘‘Krill herd optimization algorithm for
ture selection with the Wisconsin diagnosis breast cancer (WDBC) cancer feature selection and random forest technique for classification,’’
dataset,’’ Asian Pacific J. Cancer Prevention, vol. 18, no. 5, p. 1257, in Proc. 8th IEEE Int. Conf. Softw. Eng. Service Sci. (ICSESS), Nov. 2017,
2017. pp. 109–113.
[148] T. Niu, J. Wang, K. Zhang, and P. Du, ‘‘Multi-step-ahead wind speed [169] G. Zhang, J. Hou, J. Wang, C. Yan, and J. Luo, ‘‘Feature selection
forecasting based on optimal feature selection and a modified bat algo- for microarray data classification using hybrid information gain and a
rithm with the cognition strategy,’’ Renew. Energy, vol. 118, pp. 213–229, modified binary krill herd algorithm,’’ Interdiscipl. Sci., Comput. Life
Apr. 2018. Sci., vol. 12, no. 3, pp. 288–301, Sep. 2020.
[149] E. Emary, H. M. Zawbaa, K. K. A. Ghany, A. E. Hassanien, and [170] E. Emary, H. M. Zawbaa, and A. E. Hassanien, ‘‘Binary grey wolf opti-
B. Parv, ‘‘Firefly optimization algorithm for feature selection,’’ in Proc. mization approaches for feature selection,’’ Neurocomputing, vol. 172,
7th Balkan Conf. Informat. Conf., Sep. 2015, pp. 1–7. pp. 371–381, Jan. 2016.
[150] T. Kanimozhi and K. Latha, ‘‘An integrated approach to region based [171] P. Sharma, S. Sundaram, M. Sharma, A. Sharma, and D. Gupta, ‘‘Diag-
image retrieval using firefly algorithm and support vector machine,’’ nosis of Parkinson’s disease using modified grey wolf optimization,’’
Neurocomputing, vol. 151, pp. 1099–1111, Mar. 2015. Cognit. Syst. Res., vol. 54, pp. 100–115, May 2019.
[151] V. Subha and D. Murugan, ‘‘Opposition based firefly algorithm opti- [172] Y. Pathak, K. V. Arya, and S. Tiwari, ‘‘Feature selection for image
mized feature subset selection approach for fetal risk anticipation,’’ Mach. steganalysis using levy flight-based grey wolf optimization,’’ Multimedia
Learn. Appl., Int. J., vol. 3, no. 2, pp. 55–64, Jun. 2016. Tools Appl., vol. 78, no. 2, pp. 1473–1494, Jan. 2019.

VOLUME 9, 2021 26787

 P. Agrawal et al.: Metaheuristic Algorithms on Feature Selection: A Survey of One Decade of Research (2009-2019)

[173] K. Devanathan, N. Ganapathy, and R. Swaminathan, ‘‘Binary grey wolf [194] A. Naik, V. Kuppili, and D. Reddy Edla, ‘‘Binary dragonfly algorithm
optimizer based feature selection for nucleolar and centromere staining and Fisher score based hybrid feature selection adopting a novel fitness
pattern classification in indirect immunofluorescence images,’’ in Proc. function applied to microarray data,’’ in Proc. Int. Conf. Appl. Mach.
41st Annu. Int. Conf. IEEE Eng. Med. Biol. Soc. (EMBC), Jul. 2019, Learn. (ICAML), May 2019, pp. 40–43.
pp. 7040–7043. [195] M. Yasen, N. Al-Madi, and N. Obeid, ‘‘Optimizing neural networks
[174] Q. Al-Tashi, H. Rais, and S. Jadid, ‘‘Feature selection method based using dragonfly algorithm for medical prediction,’’ in Proc. 8th Int. Conf.
on grey wolf optimization for coronary artery disease classification,’’ Comput. Sci. Inf. Technol. (CSIT), Jul. 2018, pp. 71–76.
in Proc. Int. Conf. Reliable Inf. Commun. Technol. Cham, Switzerland: [196] R. Sawhney and R. Jain, ‘‘Modified binary dragonfly algorithm for fea-
Springer, 2018, pp. 257–266. ture selection in human papillomavirus-mediated disease treatment,’’ in
[175] Q. Al-Tashi, S. J. Abdulkadir, H. M. Rais, S. Mirjalili, H. Alhussian, Proc. Int. Conf. Commun., Comput. Internet Things (IC3IoT), Feb. 2018,
M. G. Ragab, and A. Alqushaibi, ‘‘Binary multi-objective grey wolf pp. 91–95.
optimizer for feature selection in classification,’’ IEEE Access, vol. 8, [197] J. Li, H. Kang, G. Sun, T. Feng, W. Li, W. Zhang, and B. Ji, ‘‘IBDA:
pp. 106247–106263, 2020. Improved binary dragonfly algorithm with evolutionary population
[176] P. Hu, J.-S. Pan, and S.-C. Chu, ‘‘Improved binary grey wolf optimizer dynamics and adaptive crossover for feature selection,’’ IEEE Access,
and its application for feature selection,’’ Knowl.-Based Syst., vol. 195, vol. 8, pp. 108032–108051, 2020.
May 2020, Art. no. 105746. [198] X. Cui, Y. Li, J. Fan, T. Wang, and Y. Zheng, ‘‘A hybrid improved
[177] Q. Li, H. Chen, H. Huang, X. Zhao, Z. Cai, C. Tong, W. Liu, and X. Tian, dragonfly algorithm for feature selection,’’ IEEE Access, vol. 8,
‘‘An enhanced grey wolf optimization based feature selection wrapped pp. 155619–155629, 2020.
kernel extreme learning machine for medical diagnosis,’’ Comput. Math. [199] A. G. Hussien, A. E. Hassanien, E. H. Houssein, S. Bhattacharyya, and
Methods Med., vol. 2017, pp. 1–15, Jan. 2017. M. Amin, ‘‘S-shaped binary whale optimization algorithm for feature
[178] A. Sahoo and S. Chandra, ‘‘Multi-objective grey wolf optimizer for selection,’’ in Recent Trends in Signal and Image Processing. Singapore:
improved cervix lesion classification,’’ Appl. Soft Comput., vol. 52, Springer, 2019, pp. 79–87.
pp. 64–80, Mar. 2017. [200] A. G. Hussien, E. H. Houssein, and A. E. Hassanien, ‘‘A binary whale
[179] J. Too, A. Abdullah, N. Mohd Saad, N. Mohd Ali, and W. Tee, ‘‘A new optimization algorithm with hyperbolic tangent fitness function for fea-
competitive binary grey wolf optimizer to solve the feature selection ture selection,’’ in Proc. 8th Int. Conf. Intell. Comput. Inf. Syst. (ICICIS),
problem in EMG signals classification,’’ Computers, vol. 7, no. 4, p. 58, Dec. 2017, pp. 166–172.
Nov. 2018. [201] G. I. Sayed, A. Darwish, and A. E. Hassanien, ‘‘A new chaotic whale
[180] N. P. Nirmala Sreedharan, B. Ganesan, R. Raveendran, P. Sarala, optimization algorithm for features selection,’’ J. Classification, vol. 35,
B. Dennis, and R. Boothalingam, ‘‘Grey wolf optimisation-based feature no. 2, pp. 300–344, Jul. 2018.
selection and classification for facial emotion recognition,’’ IET Biomet- [202] M. Tubishat, M. A. M. Abushariah, N. Idris, and I. Aljarah, ‘‘Improved
rics, vol. 7, no. 5, pp. 490–499, Sep. 2018. whale optimization algorithm for feature selection in Arabic senti-
[181] A. K. Abasi, A. T. Khader, M. A. Al-Betar, S. Naim, S. N. Makhadmeh, ment analysis,’’ Int. J. Speech Technol., vol. 49, no. 5, pp. 1688–1707,
and Z. A. A. Alyasseri, ‘‘An improved text feature selection for clustering May 2019.
using binary grey wolf optimizer,’’ in Proc. 11th Nat. Tech. Seminar [203] M. Mafarja and S. Mirjalili, ‘‘Whale optimization approaches for wrapper
Unmanned Syst. Technol. Singapore: Springer, 2021, pp. 503–516. feature selection,’’ Appl. Soft Comput., vol. 62, pp. 441–453, Jan. 2018.
[182] H. M. Zawbaa, E. Emary, and B. Parv, ‘‘Feature selection based on [204] R. K. Agrawal, B. Kaur, and S. Sharma, ‘‘Quantum based whale opti-
antlion optimization algorithm,’’ in Proc. 3rd World Conf. Complex Syst. mization algorithm for wrapper feature selection,’’ Appl. Soft Comput.,
(WCCS), Nov. 2015, pp. 1–7. vol. 89, Apr. 2020, Art. no. 106092.
[183] E. Emary, H. M. Zawbaa, and A. E. Hassanien, ‘‘Binary ant lion [205] Y. Zheng, Y. Li, G. Wang, Y. Chen, Q. Xu, J. Fan, and X. Cui, ‘‘A novel
approaches for feature selection,’’ Neurocomputing, vol. 213, pp. 54–65, hybrid algorithm for feature selection based on whale optimization algo-
Nov. 2016. rithm,’’ IEEE Access, vol. 7, pp. 14908–14923, 2019.
[184] M. Mafarja, D. Eleyan, S. Abdullah, and S. Mirjalili, ‘‘S-shaped vs. [206] Q.-T. Bui, M. V. Pham, Q.-H. Nguyen, L. X. Nguyen, and H. M. Pham,
V-shaped transfer functions for ant lion optimization algorithm in fea- ‘‘Whale optimization algorithm and adaptive neuro-fuzzy inference sys-
ture selection problem,’’ in Proc. Int. Conf. Future Netw. Distrib. Syst., tem: A hybrid method for feature selection and land pattern clas-
Jul. 2017, pp. 1–7. sification,’’ Int. J. Remote Sens., vol. 40, no. 13, pp. 5078–5093,
[185] H. M. Zawbaa, E. Emary, and C. Grosan, ‘‘Feature selection via Jul. 2019.
chaotic antlion optimization,’’ PLoS ONE, vol. 11, no. 3, Mar. 2016, [207] M. A. Tawhid and A. M. Ibrahim, ‘‘Feature selection based on rough set
Art. no. e0150652. approach, wrapper approach, and binary whale optimization algorithm,’’
[186] E. Emary and H. M. Zawbaa, ‘‘Feature selection via Lèvy antlion opti- Int. J. Mach. Learn. Cybern., vol. 11, no. 3, pp. 573–602, Mar. 2020.
mization,’’ Pattern Anal. Appl., vol. 22, no. 3, pp. 857–876, Aug. 2019. [208] H. T. Ibrahim, W. J. Mazher, O. N. Ucan, and O. Bayat, ‘‘A grasshopper
[187] M. M. Mafarja and S. Mirjalili, ‘‘Hybrid binary ant lion optimizer with optimizer approach for feature selection and optimizing SVM parameters
rough set and approximate entropy reducts for feature selection,’’ Soft utilizing real biomedical data sets,’’ Neural Comput. Appl., vol. 31, no. 10,
Comput., vol. 23, no. 15, pp. 6249–6265, Aug. 2019. pp. 5965–5974, 2019.
[188] M. Wang, C. Wu, L. Wang, D. Xiang, and X. Huang, ‘‘A feature selection [209] M. Mafarja, I. Aljarah, A. A. Heidari, A. I. Hammouri, H. Faris,
approach for hyperspectral image based on modified ant lion optimizer,’’ A. M. Al-Zoubi, and S. Mirjalili, ‘‘Evolutionary population dynamics and
Knowl.-Based Syst., vol. 168, pp. 39–48, Mar. 2019. grasshopper optimization approaches for feature selection problems,’’
[189] A. T. Azar, N. Banu, and A. Koubaa, ‘‘Rough set based ant-lion optimizer Knowl.-Based Syst., vol. 145, pp. 25–45, Apr. 2018.
for feature selection,’’ in Proc. 6th Conf. Data Sci. Mach. Learn. Appl. [210] H. Hichem, M. Elkamel, M. Rafik, M. T. Mesaaoud, and
(CDMA), Mar. 2020, pp. 81–86. C. Ouahiba, ‘‘A new binary grasshopper optimization algorithm for
[190] S. A. Medjahed, T. A. Saadi, A. Benyettou, and M. Ouali, ‘‘Kernel-based feature selection problem,’’ J. King Saud Univ.-Comput. Inf. Sci.,
learning and feature selection analysis for cancer diagnosis,’’ Appl. Soft Nov. 2019, doi: 10.1016/j.jksuci.2019.11.007.
Comput., vol. 51, pp. 39–48, Feb. 2017. [211] M. Mafarja, I. Aljarah, H. Faris, A. I. Hammouri, A. M. Al-Zoubi, and
[191] M. M. Mafarja, D. Eleyan, I. Jaber, A. Hammouri, and S. Mirjalili, S. Mirjalili, ‘‘Binary grasshopper optimisation algorithm approaches for
‘‘Binary dragonfly algorithm for feature selection,’’ in Proc. Int. Conf. feature selection problems,’’ Expert Syst. Appl., vol. 117, pp. 267–286,
New Trends Comput. Sci. (ICTCS), Oct. 2017, pp. 12–17. Mar. 2019.
[192] M. Mafarja, I. Aljarah, A. A. Heidari, H. Faris, P. Fournier-Viger, X. Li, [212] H. T. Ibrahim, W. J. Mazher, O. N. Ucan, and O. Bayat, ‘‘Feature selec-
and S. Mirjalili, ‘‘Binary dragonfly optimization for feature selection tion using salp swarm algorithm for real biomedical datasets,’’ IJCSNS,
using time-varying transfer functions,’’ Knowl.-Based Syst., vol. 161, vol. 17, no. 12, pp. 13–20, 2017.
pp. 185–204, Dec. 2018. [213] G. I. Sayed, G. Khoriba, and M. H. Haggag, ‘‘A novel chaotic
[193] A. A. Karizaki and M. Tavassoli, ‘‘A novel hybrid feature selection based salp swarm algorithm for global optimization and feature selec-
on ReliefF and binary dragonfly for high dimensional datasets,’’ in Proc. tion,’’ Int. J. Speech Technol., vol. 48, no. 10, pp. 3462–3481,
9th Int. Conf. Comput. Knowl. Eng. (ICCKE), Oct. 2019, pp. 300–304. Oct. 2018.

26788 VOLUME 9, 2021

P. Agrawal et al.: Metaheuristic Algorithms on Feature Selection: A Survey of One Decade of Research (2009-2019)

[214] H. Faris, M. M. Mafarja, A. A. Heidari, I. Aljarah, A. M. Al-Zoubi, [235] P. Agrawal, T. Ganesh, and A. W. Mohamed, ‘‘A novel binary
S. Mirjalili, and H. Fujita, ‘‘An efficient binary salp swarm algorithm with gaining–sharing knowledge-based optimization algorithm for
crossover scheme for feature selection problems,’’ Knowl.-Based Syst., feature selection,’’ Neural Comput. Appl., pp. 1–20, Oct. 2020,
vol. 154, pp. 43–67, Aug. 2018. doi: 10.1007/s00521-020-05375-8.
[215] M. Tubishat, S. Ja’afar, M. Alswaitti, S. Mirjalili, N. Idris, M. A. Ismail, [236] A. I. Hafez, A. E. Hassanien, H. M. Zawbaa, and E. Emary, ‘‘Hybrid
and M. S. Omar, ‘‘Dynamic salp swarm algorithm for feature selection,’’ monkey algorithm with krill herd algorithm optimization for feature
Expert Syst. Appl., vol. 164, Feb. 2021, Art. no. 113873. selection,’’ in Proc. 11th Int. Comput. Eng. Conf. (ICENCO), Dec. 2015,
[216] A. E. Hegazy, M. A. Makhlouf, and G. S. El-Tawel, ‘‘Improved salp pp. 273–277.
swarm algorithm for feature selection,’’ J. King Saud Univ.-Comput. Inf. [237] M. M. Mafarja and S. Mirjalili, ‘‘Hybrid whale optimization algo-
Sci., vol. 32, no. 3, pp. 335–344, Mar. 2020. rithm with simulated annealing for feature selection,’’ Neurocomputing,
[217] A. Ibrahim, A. Ahmed, S. Hussein, and A. E. Hassanien, ‘‘Fish image vol. 260, pp. 302–312, Oct. 2017.
segmentation using salp swarm algorithm,’’ in Proc. Int. Conf. Adv. Mach. [238] S. Arora, H. Singh, M. Sharma, S. Sharma, and P. Anand, ‘‘A new hybrid
Learn. Technol. Appl. Cham, Switzerland: Springer, 2018, pp. 42–51. algorithm based on grey wolf optimization and crow search algorithm for
[218] M. Tubishat, N. Idris, L. Shuib, M. A. M. Abushariah, and S. Mirjalili, unconstrained function optimization and feature selection,’’ IEEE Access,
‘‘Improved salp swarm algorithm based on opposition based learning and vol. 7, pp. 26343–26361, 2019.
novel local search algorithm for feature selection,’’ Expert Syst. Appl., [239] M. E. Abd Elaziz, A. A. Ewees, D. Oliva, P. Duan, and S. Xiong,
vol. 145, May 2020, Art. no. 113122. ‘‘A hybrid method of sine cosine algorithm and differential evolution
[219] S. K. Baliarsingh, S. Vipsita, K. Muhammad, and S. Bakshi, ‘‘Analysis of for feature selection,’’ in Proc. Int. Conf. Neural Inf. Process. Cham,
high-dimensional biomedical data using an evolutionary multi-objective Switzerland: Springer, 2017, pp. 145–155.
emperor penguin optimizer,’’ Swarm Evol. Comput., vol. 48, pp. 262–273, [240] M. A. Tawhid and K. B. Dsouza, ‘‘Hybrid binary bat enhanced particle
Aug. 2019. swarm optimization algorithm for solving feature selection problems,’’
[220] S. K. Baliarsingh and S. Vipsita, ‘‘Chaotic emperor penguin optimised Appl. Comput. Informat., vol. 16, nos. 1–2, pp. 117–136, Apr. 2018.
extreme learning machine for microarray cancer classification,’’ IET Syst. [241] E. Pashaei, E. Pashaei, and N. Aydin, ‘‘Gene selection using hybrid binary
Biol., vol. 14, no. 2, pp. 85–95, Apr. 2020. black hole algorithm and modified binary particle swarm optimization,’’
[221] G. Dhiman, D. Oliva, A. Kaur, K. K. Singh, S. Vimal, A. Sharma, Genomics, vol. 111, no. 4, pp. 669–686, Jul. 2019.
and K. Cengiz, ‘‘BEPO: A novel binary emperor penguin optimizer for [242] N. Neggaz, A. A. Ewees, M. A. Elaziz, and M. Mafarja, ‘‘Boosting salp
automatic feature selection,’’ Knowl.-Based Syst., vol. 211, Jan. 2021, swarm algorithm by sine cosine algorithm and disrupt operator for feature
Art. no. 106560. selection,’’ Expert Syst. Appl., vol. 145, May 2020, Art. no. 113103.
[222] A. K. Abasi, A. T. Khader, M. A. Al-Betar, S. Naim, S. N. Makhadmeh, [243] S. K. Baliarsingh, W. Ding, S. Vipsita, and S. Bakshi, ‘‘A memetic
and Z. A. A. Alyasseri, ‘‘A text feature selection technique based on algorithm using emperor penguin and social engineering optimization
binary multi-verse optimizer for text clustering,’’ in Proc. IEEE Jordan for medical data classification,’’ Appl. Soft Comput., vol. 85, Dec. 2019,
Int. Joint Conf. Electr. Eng. Inf. Technol. (JEEIT), Apr. 2019, pp. 1–6. Art. no. 105773.
[223] J. P. Papa, G. H. Rosa, A. N. de Souza, and L. C. S. Afonso, ‘‘Feature [244] J. Yang and H. Gao, ‘‘Cultural emperor penguin optimizer and its appli-
selection through binary brain storm optimization,’’ Comput. Electr. Eng., cation for face recognition,’’ Math. Problems Eng., vol. 2020, pp. 1–16,
vol. 72, pp. 468–481, Nov. 2018. Nov. 2020.
[224] F. Pourpanah, C. P. Lim, X. Wang, C. J. Tan, M. Seera, and Y. Shi, [245] A. K. Shukla, P. Singh, and M. Vardhan, ‘‘A new hybrid wrapper TLBO
‘‘A hybrid model of fuzzy min–max and brain storm optimization for and SA with SVM approach for gene expression data,’’ Inf. Sci., vol. 503,
feature selection and data classification,’’ Neurocomputing, vol. 333, pp. 238–254, Nov. 2019.
pp. 440–451, Mar. 2019. [246] S. K. Baliarsingh, S. Vipsita, and B. Dash, ‘‘A new optimal gene selec-
[225] F. Pourpanah, Y. Shi, C. P. Lim, Q. Hao, and C. J. Tan, ‘‘Feature selection tion approach for cancer classification using enhanced jaya-based for-
based on brain storm optimization for data classification,’’ Appl. Soft est optimization algorithm,’’ Neural Comput. Appl., vol. 32, no. 12,
Comput., vol. 80, pp. 761–775, Jul. 2019. pp. 8599–8616, Jun. 2020.
[226] E. Tuba, I. Strumberger, T. Bezdan, N. Bacanin, and M. Tuba, ‘‘Classifi- [247] R. Purushothaman, S. P. Rajagopalan, and G. Dhandapani, ‘‘Hybridizing
cation and feature selection method for medical datasets by brain storm gray wolf optimization (GWO) with grasshopper optimization algorithm
optimization algorithm and support vector machine,’’ Procedia Comput. (GOA) for text feature selection and clustering,’’ Appl. Soft Comput.,
Sci., vol. 162, pp. 307–315, Jan. 2019. vol. 96, Nov. 2020, Art. no. 106651.
[227] D. Oliva and M. A. Elaziz, ‘‘An improved brainstorm optimization [248] E. O. Reséndiz-Flores, J. A. Navarro-Acosta, and
using chaotic opposite-based learning with disruption operator for global A. Hernández-Martínez, ‘‘Optimal feature selection in industrial foam
optimization and feature selection,’’ Soft Comput., vol. 24, no. 18, injection processes using hybrid binary particle swarm optimization and
pp. 14051–14072, 2020. gravitational search algorithm in the Mahalanobis–Taguchi system,’’ Soft
[228] M. S. Krishna and S. Vishwakarma, ‘‘Improve the performance of fin- Comput., vol. 24, no. 1, pp. 341–349, Jan. 2020.
gerprint recognition using feature selection and teacher learning based [249] E.-S.-M. El-Kenawy, M. M. Eid, M. Saber, and A. Ibrahim, ‘‘MbGWO-
optimization (TLBO),’’ Int. J. Master Eng. Res. Technol., vol. 2, no. 10, SFS: Modified binary grey wolf optimizer based on stochastic fractal
pp. 167–172, 2015. search for feature selection,’’ IEEE Access, vol. 8, pp. 107635–107649,
[229] K. Jain and S. S. Bhadauria, ‘‘Enhanced content based image retrieval 2020.
using feature selection using teacher learning based optimization,’’ Int. [250] P. Bansal, S. Kumar, S. Pasrija, and S. Singh, ‘‘A hybrid grasshop-
J. Comput. Sci. Inf. Secur., vol. 14, no. 11, pp. 1052–1057, 2016. per and new cat swarm optimization algorithm for feature selection
[230] H. E. Kiziloz, A. Deniz, T. Dokeroglu, and A. Cosar, ‘‘Novel multi- and optimization of multi-layer perceptron,’’ Soft Comput., pp. 1–27,
objective TLBO algorithms for the feature subset selection problem,’’ Mar. 2020.
Neurocomputing, vol. 306, pp. 94–107, Sep. 2018. [251] V. Bolón-Canedo, D. Rego-Fernández, D. Peteiro-Barral, A. Alonso-
[231] M. Allam and M. Nandhini, ‘‘Optimal feature selection using binary Betanzos, B. Guijarro-Berdiñas, and N. Sánchez-Maroño, ‘‘On the scal-
teaching learning based optimization algorithm,’’ J. King Saud Univ.- ability of feature selection methods on high-dimensional data,’’ Knowl.
Comput. Inf. Sci., Dec. 2018, doi: 10.1016/j.jksuci.2018.12.001. Inf. Syst., vol. 56, no. 2, pp. 395–442, 2018.
[232] M. Manonmani and S. Balakrishnan, ‘‘Feature selection using improved [252] R. Wald, T. M. Khoshgoftaar, and A. Napolitano, ‘‘Stability of filter- and
teaching learning based algorithm on chronic kidney disease dataset,’’ wrapper-based feature subset selection,’’ in Proc. IEEE 25th Int. Conf.
Procedia Comput. Sci., vol. 171, pp. 1660–1669, 2020. Tools with Artif. Intell., Nov. 2013, pp. 374–380.
[233] H. Das, B. Naik, and H. S. Behera, ‘‘A Jaya algorithm based [253] U. M. Khaire and R. Dhanalakshmi, ‘‘Stability of feature selection
wrapper method for optimal feature selection in supervised algorithm: A review,’’ J. King Saud Univ.-Comput. Inf. Sci., Jun. 2019,
classification,’’ J. King Saud Univ.-Comput. Inf. Sci., May 2020, doi: 10.1016/j.jksuci.2019.06.012.
doi: 10.1016/j.jksuci.2020.05.002. [254] H. Bostani and M. Sheikhan, ‘‘Hybrid of binary gravitational search
[234] M. A. Awadallah, M. A. Al-Betar, A. I. Hammouri, and O. A. Alomari, algorithm and mutual information for feature selection in intrusion
‘‘Binary JAYA algorithm with adaptive mutation for feature selection,’’ detection systems,’’ Soft Comput., vol. 21, no. 9, pp. 2307–2324,
Arabian J. Sci. Eng., vol. 45, no. 12, pp. 10875–10890, Dec. 2020. May 2017.

VOLUME 9, 2021 26789

 P. Agrawal et al.: Metaheuristic Algorithms on Feature Selection: A Survey of One Decade of Research (2009-2019)

[255] G. Liu, B. Zhang, X. Ma, and J. Wang, ‘‘Network intrusion detection [276] D. Rodrigues, L. A. Pereira, J. P. Papa, C. C. Ramos, A. N. Souza,
based on chaotic multi-verse optimizer,’’ in Proc. 11th EAI Int. Conf. and L. P. Papa, ‘‘Optimizing feature selection through binary charged
Mobile Multimedia Commun. Ghent, Belgium: European Alliance for system search,’’ in Proc. Int. Conf. Comput. Anal. Images Patterns. Berlin,
Innovation, 2018, p. 218. Germany: Springer, 2013, pp. 377–384.
[256] F. Barani, M. Mirhosseini, and H. Nezamabadi-pour, ‘‘Application of [277] O. S. Qasim, N. A. Al-Thanoon, and Z. Y. Algamal, ‘‘Feature selection
binary quantum-inspired gravitational search algorithm in feature sub- based on chaotic binary black hole algorithm for data classification,’’
set selection,’’ Int. J. Speech Technol., vol. 47, no. 2, pp. 304–318, Chemometric Intell. Lab. Syst., vol. 204, Sep. 2020, Art. no. 104104.
Sep. 2017. [278] N. Dif and Z. Elberrichi, ‘‘Microarray data feature selection and clas-
[257] A. A. Ewees, M. A. El Aziz, and A. E. Hassanien, ‘‘Chaotic multi-verse sification using an enhanced multi-verse optimizer and support vector
optimizer-based feature selection,’’ Neural Comput. Appl., vol. 31, no. 4, machine,’’ in Proc. 3rd Int. Conf. Netw. Adv. Syst., 2017.
pp. 991–1006, Apr. 2019. [279] A. I. Hafez, H. M. Zawbaa, E. Emary, and A. E. Hassanien, ‘‘Sine cosine
[258] M. Belazzoug, M. Touahria, F. Nouioua, and M. Brahimi, ‘‘An improved optimization algorithm for feature selection,’’ in Proc. Int. Symp. Innov.
sine cosine algorithm to select features for text categorization,’’ J. King Intell. Syst. Appl. (INISTA), Aug. 2016, pp. 1–5.
Saud Univ.-Comput. Inf. Sci., vol. 32, no. 4, pp. 454–464, May 2020.
[259] X. Han, X. Chang, L. Quan, X. Xiong, J. Li, Z. Zhang, and Y. Liu, ‘‘Fea-
ture subset selection by gravitational search algorithm optimization,’’ Inf.
Sci., vol. 281, pp. 128–146, Oct. 2014.
[260] E. Pashaei, M. Ozen, and N. Aydin, ‘‘An application of black hole
PRACHI AGRAWAL received the B.Sc. degree
algorithm and decision tree for medical problem,’’ in Proc. IEEE 15th in mathematics from the Ramjas College, Delhi
Int. Conf. Bioinf. Bioeng. (BIBE), Nov. 2015, pp. 1–6. University, Delhi, India, in 2012, and the M.Sc.
[261] E. Pashaei and N. Aydin, ‘‘Binary black hole algorithm for feature selec- degree in mathematics from the Indian Institute of
tion and classification on biological data,’’ Appl. Soft Comput., vol. 56, Technology Delhi, India, in 2014. She is currently
pp. 94–106, Jul. 2017. pursuing the Ph.D. degree with the Department of
[262] B. Mohseni Bababdani and M. Mousavi, ‘‘Gravitational search algorithm: Mathematics and Scientific Computing, National
A new feature selection method for QSAR study of anticancer potency Institute of Technology Hamirpur, India. She has
of imidazo[4,5-b]pyridine derivatives,’’ Chemometric Intell. Lab. Syst., published more than ten research articles in peer
vol. 122, pp. 1–11, Mar. 2013. reviewed journals such as neural computing and
[263] A. Adam, N. Mokhtar, M. Mubin, Z. Ibrahim, M. Z. M. Tumari, and applications, OPSERCH, International journal of swarm and intelligent
M. I. Shapiai, ‘‘Feature selection and classifier parameter estimation for research, Complexity and presented various papers in international confer-
EEG signal peak detection using gravitational search algorithm,’’ in ences. Her main research interests are in the areas of stochastic optimization
Proc. 4th Int. Conf. Artif. Intell. with Appl. Eng. Technol., Dec. 2014, using computational intelligence techniques. Her research areas include
pp. 103–108.
optimization, metaheuristic algorithms, and evolutionary programming.
[264] M. Taradeh, M. Mafarja, A. A. Heidari, H. Faris, I. Aljarah, S. Mirjalili,
and H. Fujita, ‘‘An evolutionary gravitational search-based feature selec-
tion,’’ Inf. Sci., vol. 497, pp. 219–239, Sep. 2019.
[265] R. Guha, M. Ghosh, A. Chakrabarti, R. Sarkar, and S. Mirjalili, ‘‘Intro-
ducing clustering based population in binary gravitational search algo-
rithm for feature selection,’’ Appl. Soft Comput., vol. 93, Aug. 2020, HATTAN F. ABUTARBOUSH received the B.Sc
Art. no. 106341. (Hons.) degree in electrical communications and
[266] C. C. O. Ramos, D. Rodrigues, A. N. de Souza, and J. P. Papa, ‘‘On the electronics engineering from Greenwich Univer-
study of commercial losses in Brazil: A binary black hole algorithm sity, London, U.K., in 2005, the M.Sc. degree
for theft characterization,’’ IEEE Trans. Smart Grid, vol. 9, no. 2, in mobile personal and satellite communications
pp. 676–683, Mar. 2018.
from the Department of Electrical and Electron-
[267] R. Hans and H. Kaur, ‘‘Binary multi-verse optimization (BMVO)
ics Engineering, Westminster University, London,
approaches for feature selection,’’ Int. J. Interact. Multimedia Artif. Intell.,
U.K., in 2007, and the Ph.D. degree in antennas
vol. 6, no. 1, p. 91, 2020.
[268] R. Sindhu, R. Ngadiran, Y. M. Yacob, N. A. H. Zahri, and M. Hariharan, and propagation from the Department of Electron-
‘‘Sine–cosine algorithm for feature selection with elitism strategy and ics and Computer Engineering, Brunel University,
new updating mechanism,’’ Neural Comput. Appl., vol. 28, no. 10, London, U.K., in July 2011. He was a Research Visitor to the Hong Kong
pp. 2947–2958, Oct. 2017. University and National Physical Laboratory (NPL), Teddington, U.K.,
[269] S. Taghian and M. H. Nadimi-Shahraki, ‘‘Binary sine cosine algo- in 2010. He worked as a Research Associate for the American University in
rithms for feature selection from medical data,’’ 2019, arXiv:1911.07805. Cairo (AUC), in 2011, where he worked on the packaging of novel mm-wave
[Online]. Available: http://arxiv.org/abs/1911.07805 antennas. He was a Research Fellow at the King Abdullah University of
[270] L.-Y. Chuang, H.-W. Chang, C.-J. Tu, and C.-H. Yang, ‘‘Improved binary Science and Technology (KAUST), working on mm-wave and inkjet-printed
PSO for feature selection using gene expression data,’’ Comput. Biol. antennas, from 2011 to 2013. He also worked as a Research Assistant in
Chem., vol. 32, no. 1, pp. 29–38, Feb. 2008. microwave medical imaging at Bristol University, Bristol, U.K., from 2013 to
[271] X. He, Q. Zhang, N. Sun, and Y. Dong, ‘‘Feature selection with discrete 2015. In 2015, he joined the Electrical Engineering Department, Taibah
binary differential evolution,’’ in Proc. Int. Conf. Artif. Intell. Comput. University, where he is currently an Assistant Professor. He received the
Intell., vol. 4, 2009, pp. 327–330. Vice-Chancellor’s Prize Award for the Best Doctoral Research for 2011 from
[272] N. Al-Madi, H. Faris, and S. Mirjalili, ‘‘Binary multi-verse optimization Brunel University, London, U.K., and the outstanding achievement award
algorithm for global optimization and discrete problems,’’ Int. J. Mach.
from the Saudi Arabian Cultural Bureau in London, for the outstanding
Learn. Cybern., vol. 10, no. 12, pp. 3445–3465, Dec. 2019.
overall achievement in the Ph.D. degree. Also, he was selected for the top
[273] J. P. Papa, A. Pagnin, S. A. Schellini, A. Spadotto, R. C. Guido, M. Ponti,
10 Reviewers Award by the Editorial Board for his exceptional performance
G. Chiachia, and A. X. Falcao, ‘‘Feature selection through gravitational
search algorithm,’’ in Proc. IEEE Int. Conf. Acoust., Speech Signal Pro- in the IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, in 2014 and 2015.
cess. (ICASSP), May 2011, pp. 2052–2055. He has authored/coauthored over 50 technical articles in international jour-
[274] M. Bardamova, A. Konev, I. Hodashinsky, and A. Shelupanov, ‘‘A fuzzy nals and conferences. He also has served as a Reviewer and an Associate
classifier with feature selection based on the gravitational search algo- Editor for different international journals and conferences in the areas of
rithm,’’ Symmetry, vol. 10, no. 11, p. 609, Nov. 2018. antennas and propagation. His current research interests lie in the design of
[275] X. Han, D. Li, P. Liu, and L. Wang, ‘‘Feature selection by recursive binary reconfigurable antennas, antennas for mobile phones, RF/microwave circuit
gravitational search algorithm optimization for cancer classification,’’ design, mm-wave antennas, biosensors, and microwave medical imaging for
Soft Comput., vol. 24, no. 6, pp. 4407–4425, Mar. 2020. breast and brain cancer detection.

26790 VOLUME 9, 2021

P. Agrawal et al.: Metaheuristic Algorithms on Feature Selection: A Survey of One Decade of Research (2009-2019)

TALARI GANESH received the M.Sc. and Ph.D. ALI WAGDY MOHAMED received the B.Sc.,
degrees in statistics from Sri Venkateswara Uni- M.Sc., and Ph.D. degrees from Cairo University,
versity, Tirupati, Andhra Pradesh. He is currently in 2000, 2004 and 2010, respectively. He is an
an Assistant Professor with the Department of Associate Professor with the Operations Research
Mathematics and Scientific Computing, National Department, Faculty of Graduate Studies for Sta-
Institute of Technology, Hamirpur, Himachal tistical Research, Cairo University, Egypt. He is
Pradesh. He had worked as an Assistant Professor currently an Associate Professor of statistics with
with the Seshadripuram Institute of Management the Wireless Intelligent Networks Center (WINC),
Studies, Bangalore, and also joined as a Guest Fac- Faculty of Engineering and Applied Sciences, Nile
ulty at the Department of Statistics, Pondicherry University. Recently, he has recognized among the
University, Puducherry, from September 2014 to August 2015. He was a top 2% scientists according to Stanford University report 2019. He serves as
Lecturer at the Dr. Jyothirmayi Degree College, from 2009 to 2010. His a Reviewer of more than 60 international accredited top-tier journals and
main research interests are in the areas of multi-criterion decision aid, oper- has been awarded the Publons Peer Review Awards 2018, for placing in
ations research, statistics and big data science. He has worked on stochastic the top 1% of reviewers worldwide in assorted field. He is editor in more
modeling, time series analysis, and forecasting by applying to the gold than ten journals of information sciences, applied mathematics, engineering,
mineralization and rainfall data using SPSS and R tool respectively. He also system science, and operations research. He has presented and participated in
worked in analytical hierarchy process and goal programming in dealing with more than five international conferences. He participated as a member of the
multi criterion decision aid. Also, he has published various papers in the field reviewer committee for 35 different conferences sponsored by Springer and
of distribution theory, convolution and mixture distribution. He is currently IEEE. He has Obtained Rank 3 in CEC’17 competition on single objective
working on classification techniques, data mining and its applications to the bound constrained real-parameter numerical optimization in Proc of IEEE
wide varieties of high-volume data which comes under big data analytics Congress on Evolutionary Computation, IEEE-CEC 2017, San Sebastián,
and data science. His current research is focused on multi-choice stochastic Spain. Besides, Obtained Rank 3 and Rank 2 in CEC’18 competitions on
transportation problem involving fuzzy programming and evolutionary algo- single objective bound constrained real-parameter numerical optimization
rithms. He has published more than 12 articles in reputed international and and Competition on Large scale global optimization, in Proc of IEEE
national journals and presented more than ten papers in various international Congress on Evolutionary Computation, IEEE-CEC 2017, Sao Paolo, Brazil.
and national conferences. He published more than 55 papers in reputed and high impact journals like
Information Sciences, Swarm and Evolutionary Computation, Computers
& Industrial Engineering, Intelligent Manufacturing, Soft Computing and
International Journal of Machine Learning and Cybernetics. He is interested
in mathematical and statistical modeling, stochastic and deterministic opti-
mization, swarm intelligence and evolutionary computation. Additionally,
he is also interested in real world problems such as industrial, transportation,
manufacturing, education and capital investment problems. Two Ph.D. and
one master have been completed under his supervision.

VOLUME 9, 2021 26791