Improvements of bat algorithm for optimal feature selection: A systematic literature review

Abstract

Bat Algorithm (BA) has been extensively applied as an optimal Feature Selection (FS) technique for solving a wide variety of optimization problems due to its impressive characteristics compared to other swarm intelligence methods. Nevertheless, BA still suffers from several problems such as poor exploration search, falling into local optima, and has many parameters that need to be controlled appropriately. Consequently, many researchers have proposed different techniques to handle such problems. However, there is a lack of systematic review on BA which could shed light on its variants. In the literature, several review papers have been reported, however, such studies were neither systematic nor comprehensive enough. Most studies did not report specifically which components of BA was modified. The range of improvements made to the BA varies, which often difficult for any enhancement to be accomplished if not properly addressed. Given such limitations, this study aims to review and analyse the recent variants of latest improvements in BA for optimal feature selection. The study has employed a standard systematic literature review method on four scientific databases namely, IEEE Xplore, ACM, Springer, and Science Direct. As a result, 147 research publications over the last ten years have been collected, investigated, and summarized. Several critical and significant findings based on the literature reviewed were reported in this paper which can be used as a guideline for the scientists in the future to do further research.

Keywords

Bat algorithm feature selection bat improvements systematic review

1. Introduction

Recently, many fields such as Social Network sites (SNs), scientific research, and business have experienced a huge increase in data volume [1]. This increasing represented by the term “Big Data”, which is described by five V’s characteristics: volume, variety, velocity, veracity, and value [2]. Volume is a major problem for machine learning algorithms, especially in analysing large amount of data [1]. This is because of the “curse of dimensionality”, which affects the performance of the traditional machine learning algorithms. Notably, high dimension feature space contains features of different types e.g., relevant, irrelevant, and redundant features [3]. Relevant features provide valuable information, irrelevant features mislead the machine learning algorithms, and redundant features give the similar or the same information. To overcome the problem of high dimensionality, two common methods have been used; Feature Extraction (FE) and Feature Selection (FS). FE generates a new feature space with low dimensionality, meanwhile, FS selects an optimal subset of relevant features and at the same time removes redundant and irrelevant features [4]. FS methods have been widely used to enhance many underline machine learning algorithms for different real-life applications [5, 6, 7, 8, 9, 10]. The main advantage of FS over FE is that it does not change the original structure of the data features as well as it keeps their true interpretation [3]. In addition, FS increases predictive accuracy, improves learning process, minimizes memory requirement, reduces the training and computational time, and improves the generalization by reducing the overfitting problem [11, 12, 13, 14]. FS methods are generally classified into filter, wrapper, embedded, and hybrid methods [15, 16]. Regardless of the FS advantages, it introduce an extra level of complexity that need to be address thoughtfully. As a matter of fact, FS is an NP-hard problem that makes exhaustive search techniques impractical to implement [4]. Hence, it becomes necessary to find a search technique that can find an optimal feature subset effectively and efficiently. In the last two decades, a family of population-based optimization algorithms called Swarm Intelligence (SI) which were inspired by the behaviour of biological systems have emerged as powerful global searching techniques [1, 4, 17]. Such algorithms have been widely applied to solve FS problems in different applications in order to achieve various objectives such as high performance, benefits, and profits, or low cost, and time [18, 19].

Among the well-known representatives of SI for FS is Bat Algorithm (BA). BA was developed by Xin-She Yang, based on the echolocation behaviour of bats [20]. BA has been utilized as an optimal FS technique, that selects either optimal features [21], values [22] or factors [23] from a set of possible options in order to increase the performance of a model and/or system in different applications. Many studies have demonstrated the merit and effectiveness of BA in achieving better results compared to various existing benchmark [20, 21, 22, 23, 24, 25, 26]. Recently, BA has been extensively employed to select the optimal features in different data mining applications, particularly, detection tasks e.g., intrusion detection [31], spam detection [21], community detection [32], gene detection [33], cancer cell detection [34], e-fraud detection [35], anomaly detection [36], and diabetes disease detection [37]. However, there is a lack of studies that use BA to solve FS problems in other important text mining task such as Event detection (ED) or Topic Detection (TD) [24, 38]. These tasks have appeared to be an active and hot research areas in text mining [39, 44]. In addition, text data is characterized by the high dimensionality of feature space on which BA has not properly tested [30, 45, 46, 47, 48, 49]. In fact, in order to deal with these applications, many modifications have been introduced to the BA’s structure in order to overcome its weaknesses and obtain better results, which in turn led to the creation of many variants [50, 51, 52, 53, 54, 55, 56, 57]. Despite the success of BA algorithm and its variants in the area of FS, no research has been found in the literature that were focused on analysing BA’s studies for FS. In contrast, few review papers have been found in the literature that only interested in describing and summarizing studies which have used BA or one of its variant [24, 50, 51, 52, 53, 54, 55, 56, 57]. However, such surveys have some shortcomings as illustrated in Table 1.

Table 1
Focus and limitations of existing surveys

BA survey papers	Year	Focus of survey	Limitations of existing surveys
Bangyal et al. [56], Jayabarathi et al. [24]	2018	Reviewed just few variants of BA briefly and pointed out their	No survey has done a systematic review, whereby existing surveys have selected the studies randomly without
Yadav and Phogat [55]	2017	diverse applications	following any systematic methodology
Sharma et al. [54]	2016		No survey has done on BA for FS
Chawla and Duhan [52]	2015		No survey reported specifically which components of BA
Jr et al. [51], Induja and Eswaramurthy [53]	2014		exactly have been improved e.g., population initialization, frequency, velocity, new generation, local search, emission
Yang and He [50]	2013		rate and loudness update equationsNo survey reported the type of techniques used to improve
Bangyal et al. [56]	2018	Describe and analysis briefly the	the different components of BA or its variants
Yadav and Phogat [55]	2017	recent improvements that have been done on BA and its variants	No survey categorized BA’s studies based on the size of data i.e., small, medium and large
Kongkaew [57]	2017	Focus only on reviewing the recent applications of BA for discrete optimization problems

Given such limitations, this paper focuses primarily on providing a systematic literature review on BA and its recent developments as an optimal feature selection technique for solving different optimization problems. In particular, the study reviews, analyses and discusses the related works in terms of BA’s encoding schemes, objective function, and improvements on the components of BA e.g., initialize population, frequency, velocity, new solution generation, local search, emission rate, and loudness equations. In addition, techniques that have been used to improve such components of BA are also reported. This paper also seeks to identify if there is any work that applies BA for detection tasks, particularly in, ED and TD tasks from textual data. The authors of this paper are confident that this systematic review can offer researchers a better understanding about BA from different perspectives and encourage them to do more investigations on it in order to improve BA and its variants further.

2. Paper selection methodology

In this section, a systematic methodology was developed for searching the relevant literature for this survey as shown in Fig. 1.

Figure 1.

Standard systematic literature review method used to select papers for this study.

The major steps for conducting the literature search are adopted from the methodological guidelines introduced by Kitchenham [58]. In the following sub-sections, the major steps are explained as follows.

2.1 Constructing search terms

Search criteria keywords were selected by identifying the attributes and alternative synonyms terms for several concepts such as Bat, feature selection, event $\backslash$ topic detection, bat improvements, mining tasks, source of data, and review paper. The concepts and their associated terms are listed in Table 2. These terms were used to build the search criteria utilizing Boolean operators like AND and OR according to Eq. (1).

$\displaystyle\textit{Search string}=BA\cap(FS\cup ED\cup BI\cup MT\cup SD\cup RP)$ (1)

Table 2

Search criteria keywords

Symbol	Concept	Key words
BA	BA	BA, BBA, Bat, “Binary Bat”
FS	Feature Selection	Feature, Selection, Reduction, Dimension, “Feature selection”, “Feature Reduction”,
		“Dimension reduction”, Filter, Wrapper, Embedded
ED	Event Detection	Events, “Event mining”, “Event detection”, “Event Extraction”, “Event identification”,
		“Emerging event”, “Event exploration”, ED, RED, NED, “Retrospective event detection”,
		“New event detection”, “Offline event detection”, “Online event detection”, “Event
		discovery”, “Event evolution”, “Event Tracking”, “Rare events”, “Event categorization”,
		Topics, “Topic Mining”, “Topic detection”, “Topic extraction”, “Topic identification”,
		“Emerging Topic”, “Topic exploration”, “Topic Categorization”, TDT, “Topic detection
		and tracking”, “Topic discovery”, “Topic evolution”
BI	Bat Improvements	Hybrid, Discrete, Adaptive, Self-adaptive, Dynamic, Improved, Modified, Chaotic, Novel,
		New, “Parameter control”
MT	Mining Task	Clustering, Classification, Categorization, Mining
SD	Source of Data	Facebook, Twitter, LinkedIn, Blogs, News, “Social Networks”, “Social Media”, SN, SNS
RP	Review Papers	Survey, Reviews, Systematic, Study, “State of art”, “Systematic literature review”,
		“Systematic review”

2.2 Articles search strategy

The search was limited to four scientific databases that mostly cover computer science-related works: IEEE Xplore, ACM, Springer, and Science Direct. Such databases were used to search and filter the relevant papers. The four databases with their corresponding search results are given in Table 3.

Table 3
Scientific databases and results for literature search

No	Data source	Total	Exclusion 1	Exclusion 2	Inclusion	Final selection
1	IEEE	6,093	5,974	3,482	47	147 Articles
2	ACM	115,493	31,493	132	37
3	ScienceDirect	103	99	99	22
4	Springer	39,773	27,313	3,295	41
	Total	161,462	64,879	7,008	147

2.3 Exclusion and inclusion criteria

This study implemented some exclusion and inclusion criteria as follows.

2.3.1 Exclusion criteria

Two exclusion criteria were used to exclude literature that was not relevant to this study. First criteria focus on removing all articles that were not written in the English language as well as removing magazines, books, early access articles and other literature sources apart from journals and conferences. Second criteria remove all papers that were not published within the period from Jan 2010 to Feb 2020, and this specific period has been chosen because BA was introduced for the first time in 2010.

2.3.2 Inclusion criteria

The inclusion criteria were used to determine the relevant works through reading the paper’s title and abstract to perceive how it relates to this survey. In this process authors considered the following inclusion criteria:

•
Studies that have reviewed BA or its variants in the context of optimal feature selection technique.
•
Studies that are relevant to the eight major research questions (RQ1-RQ8) in which this study attempts to answer. Such questions are basically constructed to overcome the limitations identified from the existing surveys of BA. The research questions are as follows:

RQ1: What are the recent variants of BA in the context of optimal FS technique for solving various optimization problems?

RQ2: What is the most frequently used encoding scheme in BA?

RQ3: What is the most common objective function used in BA’s applications?

RQ4: Which components of BA algorithms have been improved?

RQ5: What are the techniques that have been used to improve the different components of BA?

RQ6: What are the prominent applications of BA as an optimal feature selection technique?

RQ7: What is the size of the data that BA and its variants have been applied on?

RQ8: Is there any application of BA and its versions on event $\backslash$ topic detection from textual data?

Considering both exclusion and inclusion criteria, a total of 147 relevant studies are chosen as final selection for this systematic review. The distribution of selected works over the past 10 years is visible in Fig. 2.

Figure 2.
Distribution of selected papers over the years (2010–2020).

3. Bat algorithm

BA is a SI algorithm which was developed based on the echolocation property of microbats [20]. Echoes are analysed and evaluated by bats to use them in communication, identifying various types of insects, detecting the direction and distance of their prey as well as avoiding to bump with other close objects while moving in complete darkness [59]. To emphasize, basic BA consists of four main stages as follows [55]:

1.
Initialize Bat Population: The population of bats is initialized using randomly selected value from a collection of real numbers (i.e., from lower to higher values) to find a high-quality optimal solution for the given problem in d-dimensional search area. Subsequently, the solutions which are found by population are evaluated using Eq. (2):

$\displaystyle x_{ij}=x_{\min}+\varphi({x_{\max}-x_{\min}})$ (2)

where $x_{\min}$ and $x_{\max}$ are lower and higher borders for the dimension space $j=1,2\ldots D$ , while $i=1,2,3,\ldots N$ , is the population number of BA. $\varphi$ is a randomly generated value from [0, 1].
2.
Updating Frequency, Velocity, and New Solution: Within this stage, the position $x_{i}$ and velocity $v_{i}$ of every bat in a d-dimensional search space are updated throughout the iterations. Thus, the new solutions $x_{i}^{t}$ and velocities $v_{i}^{t}$ at time step $t$ can be calculated by:

$\displaystyle f_{i}=f_{\min}+(f_{\max}-f_{\min})\beta,$ (3) $\displaystyle v_{i}^{t}=v_{i}^{t-1}+(x_{i}^{t-1}-x_{\ast})f_{i}$ (4) $\displaystyle x_{i}^{t}=x_{i}^{t-1}+v_{i}^{t},$ (5)

where $\beta$ is a random vector which is selected uniformly from [0, 1]. $x_{\ast}$ is the currently best global solution (i.e., location) found so far after comparing all existing solutions among all $N$ bats. To emphasize, the result of the following product ( $\lambda_{i}f_{i}$ ) is called velocity increment and to modify the velocity either $f_{i}$ or $\lambda_{i}$ can be used i.e. while fixing one of the factors $f_{i}$ or $\lambda_{i}$ and taking into account the type of the problem in hand. Additionally, every bat is given a random frequency value that is chosen uniformly from [ $f_{\min},f_{\max}$ ]. The range varies depending on the domain size for the problem of interest. Surprisingly, steps of updating both velocities and positions of bats are similar to the steps of the basic Particle Swarm Optimization (PSO) algorithm where, $f_{i}$ basically controls the range and speed movement of the swarm particles. Therefore, BA can be considered sometimes as a balanced integration of basic PSO and an exhaustive local search ruled by the loudness and emission pulse rate. On the other hand, in the context of local search, a solution was proposed called a random walk in which each bat walks around the existing best solutions and hence, a new solution for every single bat is regenerated locally using Eq. (6):

$\displaystyle x_{\textit{new}}=x_{\textit{old}}+\varepsilon At,$ (6)

where $\varepsilon$ is a random value selected from [ $-$ 1, 1], $At=<A_{i}^{t}>$ represents the average loudness of all bats at $t$ time step and $x_{\textit{old}}$ is the best solution recognized by a specific mechanism. In fact, above updating equation is a very special state of BA, that is identical to the key equation of Simulated Anneling (SA) algorithm.
3.
Updating emission rate ( $r$ ) and loudness ( $A$ ): In practice, when a bat finds its prey, the $A$ normally decreases, while $r$ increases. Indeed, both $A$ and $r$ are updated using Eqs (7) and (8), respectively, as the iterations advance. For instance, $A$ can be selected as any value of convenience, e.g., use $A_{0}=$ 100 and $A_{\min}=$ 1. Other values also can be used e.g., $A_{0}=$ 1 and $A_{\min}=$ 0, with an assumption that $A_{\min}=$ 0, denotes that a bat has just found its prey and for the time being stop producing any sound. Fister et al. [55] stated that some experiments are needed to be done before selecting the better values for the parameters.

$\displaystyle A_{i}^{t+1}=\alpha A_{i}^{t},r_{i}^{t+1}=r_{i}^{0}[{1-\exp({-% \gamma t})}]$ (7)

where $\gamma$ and $\alpha$ are constants; $\alpha$ is the cooling aspect of a cooling schedule in SA algorithm. For any 0 $<\alpha$ , $\gamma<$ 1, then have:

$\displaystyle A_{i}^{t}\to 0,r_{i}^{t}\to r_{i}^{0},\textit{ as }t\to\infty$ (8)

Every bat has various initial values for ( $A$ ) and ( $r$ ); this is done through using the random mechanism. For example, $r$ can be zero or any random value is drawn from [0, 1]. Meanwhile, ( $A$ ) can be any value in [1, 2]. To emphasize, ( $A$ ) and ( $r$ ) are updated just if the new solutions are enhanced, which indicates that bats are leading on the way to achieve the optimal solutions.
4.
Evaluation, Saving and Ranking Best Solutions: After updating both ( $A$ ) and ( $r$ ), an evaluation process is carried out to evaluate newly generated solutions for all bats. If the obtained solutions satisfy the given condition, then they will be archived conditionally as best solutions. Finally, a ranking process will be performed on all bats to find the current best solution ( $x_{\ast}$ ). The fundamental steps of a standard BA can be summarized in Algorithm (1) which is given by Yang (2010). Essentially, each distinct bat $x_{i}$ is recognized by five characteristics: a present position in search area $x_{i}^{t}$ , a current velocity $v_{i}^{t}$ , a loudness $A_{i}^{t}$ a, emission rate $r_{i}^{t}$ and a frequency $f_{i}$ [see Algorithm 1]:

Algorithm 1: Basic Bat Algorithm (BA)

1. Objective function: $f(x)$ , $x=(x_{1},\ldots,x_{d})^{T}$

2. Initialize the bat population: $x_{i}(i=1,2,\ldots,n)$ and $v_{i}$

3. Define pulse frequency: $f_{i}$ at $x_{i}$

4. Initialize pulse rates $r$ and the loudness $A$

5. while (t $<$ Max number of iterations)
a.
Generate new solutions by adjusting frequency, and updating velocities and locations/solutions [Eqs (2) to (4)] b.
if (rand $>r_{i}$ )

i.
Select a solution among the best solutions
ii.
Generate a local solution around the selected best solution

c.
end if
d.
Generate a new solution by flying randomly
e.
If (rand $<A_{i}$ & $f(x_{i})<f(x_{\ast})$ )

i.
Accept the new solutions
ii.
Update $r_{i}$ and reduce $A_{i}$

f.
end if
g.
Rank the bats and find the current best solution ( $x_{\ast}$ )

6. end while

7. Postprocess results and visualization

3.1 Search strategies of bat algorithm

Algorithm 1: Basic Bat Algorithm (BA)
1. Objective function: $f(x)$ , $x=(x_{1},\ldots,x_{d})^{T}$
2. Initialize the bat population: $x_{i}(i=1,2,\ldots,n)$ and $v_{i}$
3. Define pulse frequency: $f_{i}$ at $x_{i}$
4. Initialize pulse rates $r$ and the loudness $A$
5. while (t $<$ Max number of iterations) a. Generate new solutions by adjusting frequency, and updating velocities and locations/solutions [Eqs (2) to (4)] b. if (rand $>r_{i}$ ) i. Select a solution among the best solutions ii. Generate a local solution around the selected best solution c. end if d. Generate a new solution by flying randomly e. If (rand $<A_{i}$ & $f(x_{i})<f(x_{\ast})$ ) i. Accept the new solutions ii. Update $r_{i}$ and reduce $A_{i}$ f. end if g. Rank the bats and find the current best solution ( $x_{\ast}$ ) 6. end while
7. Postprocess results and visualization

Practically, BA is based on three search strategies namely, global, local and random searches, which are executed in sequence. Initially, global and random strategies are implemented to allow BA to explore diverse search space areas effectively. These two strategies are employed through modifying frequency and velocity parts of BA that were represented in Eqs (3) and (4), respectively. The updating process of velocity helps in exploration search, which mainly means to prevent BA from falling into local optimum solutions. Subsequently, BA applies local search which is done using Eq. (6). Local strategy assists BA to identify the best solution in the present region which was recognized from previous global and random strategies [60]. However, A and r parameters of BA have a significant impact on the performance of these strategies [26, 72]. Parameter A is responsible for controlling the performance and continuity of the global and random search whereas parameter r controls the implementation of the local search [25, 26, 28, 59, 73]. In fact, A and r are primarily responsible for shifting between the exploration and exploitation processes in order to achieve the right balance between them [61].

3.2 Advantages and disadvantages of bat algorithm

Table 4 briefly summarizes the main advantages and disadvantages of BA. Starting with the main advantages, BA is a popular SI algorithm that integrates key features of several optimization algorithms like PSO, Harmony Search (HS), and SA. In fact, PSO and HS are considered as special cases of BA under some conditions, and the parameter alpha in Eq. (7) imitates cooling schedule factor in SA. In addition, BA possesses more diversity of solutions due to the population and frequency variations characteristics. Moreover, all parameters of BA can be varied as the iteration proceeds while the values of the parameters in basic Genetic Algorithm (GA), PSO, and Differential Evolution (DE) versions are fixed. Furthermore, BA has the capability of auto zooming into a promising solution region that holds local intensive exploitation; other SI algorithms however do not have this ability. Finally, BA has a fast convergence rate especially, at the early stages of the iterations. Therefore, BA is very useful to obtain good solutions to some complicated and complex problems in a short time.

Regardless the advantages BA, it is suffering from several issues as reported in several literatures [24, 38, 50]. BA has poor exploration skill due to its fast convergence rate which makes BA unable to explore the whole search space in order to identify the global optimal solution for the given problem. In addition, BA comprises of several parameters like $r$ , $A$ , $\beta$ , $\gamma$ , $\varepsilon$ , and $\alpha$ whose interaction can affect the quality of the solution [27, 56, 60, 66]. Among such parameters, r and A have considered as the most influencing factors on the performance of BA as they are responsible to achieve the right balance between exploration and exploitation processes of BA [61, 62]. To illustrate, if there is too much exploitation and a little exploration, BA converges very fast at early stages and consequently, fall into local optimum solutions [38, 55]. In contrast, if there are too much exploration and little exploitation, then it is become difficult to converge and thus, slow down the overall performance of BA. As a result, determining the best settings for such parameters has become a very challenging task [38].

Table 4
Advantages and disadvantages of bat algorithm

Advantages	Disadvantages
• Simple, easy to use, flexible, scalable • Combine the advantages of popular existing optimization algorithms • Strong exploitation capability • Preserve diversity of solutions in population • Has very fast convergence rate • Has an automatic zooming mechanism • Its parameters can be updated as iteration proceeds • Ability to solve multi-model and multi-objective problems effectively • Ability to hybridize with other SI algorithms	• Has poor exploration ability • May not find the global optimal solution due to its fast convergence rate • Has many parameters that required to be tuned in advance or controlled during the execution • The exploration and exploitation need to be balanced in right manner

Advantages

Disadvantages

•

Simple, easy to use, flexible, scalable

•

Combine the advantages of popular existing optimization algorithms

•

Strong exploitation capability

•

Preserve diversity of solutions in population

•

Has very fast convergence rate

•

Has an automatic zooming mechanism

•

Its parameters can be updated as iteration proceeds

•

Ability to solve multi-model and multi-objective problems effectively

•

Ability to hybridize with other SI algorithms

•

Has poor exploration ability

•

May not find the global optimal solution due to its fast convergence rate

•

Has many parameters that required to be tuned in advance or controlled during the execution

•

The exploration and exploitation need to be balanced in right manner

3.3 Variants of bat algorithm

To overcome the shortcomings of BA and improve its performance, many methods have been introduced by different researchers [4, 13, 38, 48, 67]. This, in turn, led to the creation of many advanced and enhanced variants of BA. Therefore, this study provides a systematic literature review on the latest variants of BA that have been used as optimal feature selection technique for solving different real-world optimization problems. The related studies are precisely summarized and categorized based on three main concepts (see Tables 5 and 6).

Table 5
Definitions and notations of summary’s concepts

Concepts	Definitions and notations
Encoding scheme	Co: Continues, B: Binary, D: Discrete
Objective function	Single: Single-Objective Function, Multi: Multi-Objective Function
Modified components of BA	IP: Initialize Population, FE: Frequency Equation, VE: Velocity Equation, NSG: New
	Solution Generation Equation, LS: Local Search Equation, A: Loudness Update Equation,
	r: Emission Rate Update Equation, Ft: Fitness Function, Hy: Hybrid, NN: None (use the
	standard BA)

Table 6

Summary of recent variants of bat for optimal feature selection

Ref

Encoding scheme

Objective function

Modified components of BA

Single

Multi

NSG

Yuan et al. [23]

\surd

\surd

\surd

\surd

Fister Jr et al. [68]

\surd

\surd

\surd

\surd

\surd

\surd

\surd

\surd

Dhal and Das [63]

\surd

\surd

\surd

\surd

\surd

\surd

\surd

Gupta et al. [61]

\surd

\surd

\surd

Lyu et al. [66]

\surd

\surd

\surd

\surd

\surd

\surd

\surd

\surd

Jordehi [69]

\surd

\surd

\surd

\surd

Abdel-Raouf et al. [70]

\surd

\surd

\surd

\surd

Afrabandpey et al. [71]

\surd

\surd

\surd

\surd

\surd

Cai et al. [72]

\surd

\surd

\surd

\surd

\surd

Jensi and Jiji [73]

\surd

\surd

\surd

Baziar et al. [74]

\surd

\surd

\surd

\surd

Liu and Qi [75]

\surd

\surd

\surd

\surd

\surd

Perez et al. [76]

\surd

\surd

\surd

\surd

Pérez et al. [77]

\surd

\surd

\surd

\surd

\surd

\surd

Barbosa and Vasconcelos [78]

\surd

\surd

\surd

Enache and Science [31]

\surd

\surd

\surd

\surd

Sharma and Annappa [32]

\surd

\surd

\surd

\surd

Alomari et al. [33]

\surd

\surd

\surd

Akinyelu and Adewumi [35]

\surd

\surd

\surd

Cheruku et al. [37]

\surd

\surd

\surd

\surd

Jr et al. [64]

\surd

\surd

\surd

Chakri et al. [45]

\surd

\surd

\surd

\surd

\surd

\surd

\surd

Yılmaz and Küçüksille [59]

\surd

\surd

\surd

\surd

\surd

Sabba and Chikhi [60]

\surd

\surd

\surd

\surd

Amir H Gandomi and Yang [62]

\surd

\surd

\surd

\surd

\surd

\surd

Taha et al. [27]

\surd

\surd

\surd

\surd

\surd

Niu et al. [79]

\surd

\surd

\surd

\surd

\surd

M. Mohamed and Moftah [80]

\surd

\surd

\surd

\surd

Ramli et al. [81]

\surd

\surd

\surd

\surd

Mahdad [82]

\surd

\surd

\surd

Heraguemi et al. [83]

\surd

\surd

\surd

Shan et al. [84]

\surd

\surd

\surd

\surd

Li and Zhou [85]

\surd

\surd

\surd

\surd

\surd

Saji and Riffi [86]

\surd

\surd

\surd

\surd

\surd

Cai et al. [87]

\surd

\surd

\surd

Osaba et al. [88]

\surd

\surd

\surd

\surd

Alihodzic and Tuba [89]

\surd

\surd

\surd

\surd

\surd

Cai et al. [90]

\surd

\surd

\surd

\surd

Perez et al. [91]

\surd

\surd

\surd

\surd

Li, Z. Zhao, R. Li [92]

\surd

\surd

\surd

\surd

Ahmadi and Nikravesh [93]

\surd

\surd

\surd

\surd

Cui et al. [94]

\surd

\surd

\surd

Perez et al. [95]

\surd

\surd

\surd

\surd

\surd

Chowdhury et al. [96]

\surd

\surd

\surd

\surd

Naik et al. [97]

\surd

\surd

\surd

Zhou et al. [98]

\surd

\surd

\surd

\surd

Mirjalili et al. [99]

\surd

\surd

\surd

\surd

Wang et al. [100]

\surd

\surd

\surd

\surd

\surd

Gan and Lai [101]

\surd

\surd

\surd

\surd

Zade and Aliahmadipour [102]

\surd

\surd

\surd

\surd

Jaddi et al.

\surd

\surd

\surd

\surd

\surd

Table 6, continued
No	Ref	Encoding scheme			Objective function		Modified components of BA
		Co	B	D	Single	Multi	IP	FE	VE	NSG	LS	A	r	Ft	Hy	NN
52	Enache and Sgârciu [104]		$\surd$		$\surd$						$\surd$			$\surd$
53	Huang et al. [105]		$\surd$		$\surd$				$\surd$	$\surd$
54	Taha et al. [106]	$\surd$			$\surd$										$\surd$
55	Emary et al. [26]	$\surd$			$\surd$		$\surd$				$\surd$			$\surd$
56	Yang et al. [107]	$\surd$			$\surd$				$\surd$	$\surd$				$\surd$
57	Zhao and He [28]		$\surd$		$\surd$			$\surd$						$\surd$
58	Lin et al. [108]	$\surd$			$\surd$			$\surd$			$\surd$
59	Rizk-Allah and Hassanien [109]		$\surd$		$\surd$										$\surd$
60	Messaoudi and Kamel [110]	$\surd$				$\surd$	$\surd$			$\surd$					$\surd$
61	Enache and Sgârciu [111]	$\surd$			$\surd$						$\surd$
62	Srividhya and Mallika [112]	$\surd$			$\surd$		$\surd$		$\surd$						$\surd$
63	Murugan et al. [113]	$\surd$			$\surd$							$\surd$	$\surd$		$\surd$
64–65	Pan et al. [114], Rozlini et al. [115]	$\surd$			$\surd$										$\surd$
66–67	Enache et al. [116], Enache and Sgârciu [117]		$\surd$		$\surd$						$\surd$
68–69	Hasan et al. [118], Shan and Cheng [119]	$\surd$			$\surd$					$\surd$	$\surd$
70–71	Chen et al. [120], Topal and Altun [121]	$\surd$			$\surd$			$\surd$	$\surd$	$\surd$	$\surd$
72–73	Zhou et al. [22], Wang et al. [122]	$\surd$			$\surd$				$\surd$	$\surd$
74–75	Heraguemi et al. [123], Seyman [124]	$\surd$				$\surd$								$\surd$
76–77	Chakri et al. [125], Reddy et al. [126]	$\surd$			$\surd$				$\surd$	$\surd$	$\surd$	$\surd$	$\surd$
78–80	Bavafa et al. [127], Zhao et al. [128], Rahimi et al. [129]	$\surd$			$\surd$					$\surd$
81–83	Zhang et al. [49], Chaudhary and Banati [130], Fister et al. [131]	$\surd$			$\surd$										$\surd$
84–86	Bansal and Sahoo [132], Reddy et al. [133], Xie et al. [134]		$\surd$		$\surd$									$\surd$
87–94	Rajalaxmi and Ramesh [21], Nakamura et al. [25], Rani and Rajalaxmi [29], Taghian et al. [135], Rodrigues et al. [136], Deshpande et al. [137], Liu et al. [138], Kang et al. [139]		$\surd$		$\surd$					$\surd$
95–103	Deng and Duan [140], Raj et al. [141], Priyadharshini et al. [142], Tharwat, Zawbaa, Gaber, Hassanien [143], Hamidzadeh et al. [144], Al-Betar et al. [145], Satapathy et al. [146], Kotteeswaran and Sivakumar [147], Ghanem and Jantan [148]	$\surd$			$\surd$						$\surd$
104–119	Meng et al. [149], Ibrahim et al. [150], Hafezi et al. [151], Pravesjit [152], Bento et al. [153], Wang et al. [128], Al-Betar et al. [154], Yaseen et al. [155], Paiva et al. [156], Banati and Chaudhary [130], Salgotra and Singh [157], Kaur et al. [158], Arunarani and Manjula [159], Kiran and Reddy [160], Shehab et al. [161], Saha et al. [162]	$\surd$			$\surd$										$\surd$
120–125	Alomari et al , Chen et al. , Li and Le , Parashar et al. , Tawhid and Dsouza , Tufail et al.		$\surd$		$\surd$										$\surd$

Table 6, continued
No	Ref	Encoding scheme			Objective function		Modified components of BA
		Co	B	D	Single	Multi	IP	FE	VE	NSG	LS	A	r	Ft	Hy	NN
126–147	Yusof and Wahidi , Rattan et al. , Enache and Sgarciu , Kavitha and Christopher , Ye et al. , Imane and Nadjet , Cheng and Bao , Qiao et al. , Khennak and Drias , Srivastava et al. , Oshaba et al. , Sambariya and Prasad , Das et al. , Jalal et al. , Amir Hossein Gandomi et al. , Sahlol et al. , Mohamed et al. , Ye et al. , Sathananthavathi and Indumathi , Aasha , Tang et al. , Agarwal and Mehta	$\surd$			$\surd$											$\surd$

3.3.1 Variants of bat based on encoding scheme

Considering the encoding schemes, four variants of BA are defined by this study, which is called; continues, binary, discrete, and mix. Figure 3 depicts the percentage of each scheme type that occurred in the total collected articles. Obviously, the continuous encoding scheme is the most common one, appearing in 114 works (77%), followed by binary encoding with 29 papers (20%) as well as discrete and mix encoding with 2% and 1% papers, respectively. A brief description of these schemes is provided in the following sub-sections.

Figure 3.

Percentages of publications based on encoding scheme of bat.

3.3.1.1 Continuous encoding scheme

BA basically was developed to deal with and solve various continuous optimization problems. For this kind of version, the dimension search space is represented using continuous values which are derived randomly from (0, 1) [57]. The first set of applications of BA for continuous optimization problems was in the context of engineering design optimization which has been extensively studied and proved that BA can achieve optimal solutions accurately for highly nonlinear optimization problems [155, 157, 160, 162, 167, 175, 176, 177, 178, 187]. Later, its applications have been extended to include estimating the optimal values of the parameters for various domains, such as integer programming [70], feature selection [168], clustering [73], image processing [169], community detection [110], energy management problem [74], truck and trailer routing problem [100], intrusion detection [92], wireless sensor network [87], etc. The wide range of applications used for the continuous version of BA since its appearance in 2010, justifies the high percentage of studies used this scheme compared to other encoding schemes.

3.3.1.2 Binary encoding scheme

Binary BA (BBA) was introduced in 2012, by [25] to solve FS problems. The encoding scheme for BBA is represented as a single-dimensional vector whose length depends on the number of features and/or values in the given dataset, and each feature within the vector may have only two values; {1} indicates that the corresponding feature is selected whereas {0} depicts that the feature is not selected. Practically, a pseudo-code of BBA is similar to the basic BA with a slight difference in the update bat’s position equation, where it is replaced with binary vectors through applying one of the transfer functions like sigmoid function as follow:

$\displaystyle S({v_{i}^{j}})=\frac{1}{1+e^{-v_{i}^{j}}}$ (9)

Thus, Eq. (5) of generating a new BA’s location is modified as follows:

$\displaystyle x_{i}^{j}=\left\{{{\begin{array}[]{ll}1&\text{if }S({v_{i}^{j}})% >\sigma\\ 0&\text{if otherwise}\\ \end{array}}}\right.$ (10)

where (1) means the feature is selected and (0) the feature is not selected in which $\sigma\sim U({0,1})$ . BBA has recently gained a lot of popularity and this is clear from Fig. 3, as this version represents approximately 20% of the total studies collected for this study. This probably happened because of the promising results of BBA in comparison with the continuous BA [4, 48] as well as other binary SI algorithms in various applications such as image steganalysis [138], classification [135], clustering [137], intrusion detection [116], gene selection [163], speaker verification [133], digital watermarking for relational database [167], E-fraud detection [35], knapsack problem [109], analog test point selection [28], spam detection [21], etc.

3.3.1.3 Discrete encoding scheme

For a discrete encoding scheme, there are some researchers that encode every bat in the population into a version of integer numbers. Each BA is represented as $x_{i}=(5,2,4,3,1)$ . Similar to BBA, to get discrete representation some modification is required to be done on the formulations of BA’s movement and local search part [86, 88]. Based on the collected papers, a limited number of works (3) are identified that utilized discrete BA version. These studies mainly focus on applications like traveling salesman problems [86, 88], community detection [83, 98].

3.3.1.4 Mix encoding scheme

In this type, researchers have incorporated different types of encoding schemes into a single algorithm. Only one study was found out of all studies collected for this review that used both binary and discrete encoding schemes in order to solve the multidimensional knapsack problem [60].

3.3.2 Variants of bat based on single and multi-objective functions

Regarding the application of BA for single or multi-objective problems, it is cleared from Fig. 4, that most of the works belong to single-objective optimization problems with 140 papers (95%). Meanwhile, only 5% of studies have concerned about solving multi-objective optimization problems.

Figure 4.

Percentages of publications based on objective function type (single vs multi-objective).

Most works have focused on achieving only a single objective, according to the requirements of the model. For instance, the goal could be obtaining a high-performance accuracy of a model [135, 165], or achieving low mean square error [102, 103]. In addition, other objectives could be reducing the lost cost [82, 92], or execution time of the given model [59, 186], and so forth. The objective function is also known as a fitness function has been used to measure and evaluate the performance of the BA in various applications [29, 136]. From reading and investigating the related works of this study, the researchers have found that some scientists have used fitness functions introduced by their own [75, 107, 171], or adopt functions introduced by other authors [163]. The success of BA as a single-objective optimizer technique (i.e., especially when dealing with continuous search spaces) has encouraged researchers to extend the use of this algorithm in various domains, and this explains a large number of works used this particular variant.

On the other hand, several studies have interested in achieving multiple objectives simultaneously, as optimization problems with more than one objective function are quite popular in many areas. For example, researchers of [98] concerned about obtaining high modularity value for community detection and at the same time focused on decreasing the error rate. In addition, [23] aimed to minimize fuel cost and emission for the power flow problem. In contrast, authors in [80] were concerned with decreasing the number of selected features as well as reduce the error rate for their model. Similarly, the high quality of association rules and a minimum execution time were obtained in [123]. In the same context, several authors have used two objective functions constructed on several measures [83, 110, 124].

3.3.3 Variants based on modified components of bat algorithm

Taking into consideration the issues experienced by BA, which were previously explained (see Table 4), and with the aim to improve BA’s performance, many researchers have focused either on controlling and tuning BA’s parameters, modifying different components of BA, or hybridizing BA with existing SI algorithms or other methods. Table 7 provides a summary of the techniques that have been proposed by different scientists to improve BA as an optimal feature selection technique for solving various optimization problems.

Table 7
Techniques used for improving components of bat algorithm

Part of BA	Techniques used for improvements
Initialize Population (IP)	Normal, large, small, and mixed initialization techniques [26], Mean shift algorithm [110], K-means [92], [101], [112], Experimental plan [68], Half number of features [29]
Frequency Equation	Deterministic: {(Chaotic map) [28], [62], [70], [71], [96], [108] (Levy flight) [84]} Adaptive: {(Feedback simple rule) [54, 115], Fuzzy logic [67, 68, 86]}, Self-adaptive [75], Experimental plan [59, 95], Different scale for each BA [72], Modified [28, 36, 57, 116]
Velocity Equation	Adapt inertia weight [23], [63], [79], [81], [94], [105], [187], Adapt velocity max value [27], Random BA [107], Modified [22], [66], [72], [85], [86], [88], [90], [102], [120]–[122], [125], Inertia and coefficient factors [59], Euclidean distance [112], Fuzzy logic [95]
New Generation Solution Equation	Guidance of neighbor BA [105], [128], Mutation operator [73], [79], [107], [110], [129], [140], Modified [22], [45], [66], [85], [86], [88], [93], [98], [118]–[122], [125], [187], Different transfer functions [92], [99], [135], Self-adaptive [127]
Local Search	Deterministic: {(Chaotic $+$ Levy flight) [108], [146], (Chaotic map) [62]–[64], [140]–[142], [144], (Gaussian walk) [64], [72], (Levy flight) [31], [64], [73], [111], [116], [117], [147], (RSITFC) [64], (Uniform)[64]} Adaptive: {(Feedback simple rule) [119], [143], Fuzzy logic [77], [118]} Self-adaptive [26], [74] Modified [32], [35], [66], [89], [93], [120], [121], [125], [148], [187], Five different neighbourhood structures [100], Euclidean distance [96, 111], Foraging strategy [90], Six adapt new selection schemes [145], Acceleration factor [79], Condition for accept solution [45]
Loudness equation (A) and Emission rate equation (r)	Auto zooming: {[13], [21], [25], [26], [31]–[37], [59], [61], [72], [73], [78], [80], [81], [83]–[85], [88], [90], [92]–[94], [97], [99], [101], [102], [104]–[106], [108]–[112], [114]–[123], [127], [129], [131], [134], [137], [139], [141], [142], [146], [147], [150], [153]–[156], [159], [160], [162]–[165], [167]–[179], [181]–[184]} Fixed value: {[23], [28]–[30], [35], [49], [64], [79], [86], [98], [103], [124], [128], [132], [133], [135], [136], [138], [140], [143], [144], [148], [151], [152], [157], [161], [166], [173], [180], [186], [188]]} Deterministic (chaotic map): {[62], [69], [71], [113]} Adaptive: {(Feedback simple rule) [45], [63], [125], [187], Fuzzy logic: [77], [91]} Experimental plan: [68] Self-adaptive: [66], [75], [100], [149]}
A	Deterministic (chaotic map) [82], Self-adaptive [74], Adaptive: {(Feedback simple rule) [27], [96], Fuzzy logic [95]} Fixed value: [76] Auto-zooming [70], [87], [89], [96]
r	Fixed value [27], [96] Deterministic (Chaotic map) [70] Auto-zooming [74], [82], [95] Adaptive {(Feedback simple rule) [87], [89], Fuzzy logic [76]}
Hybrid	Mutual Information (MI) $+$ BA [106], BA $+$ Invasive weed seed algorithm [59], BBA $+$ Rough set scheme [109], BA $+$ PSO [78], [102], [155], BA $+$ Mean shift algorithm [110], BA $+$ Dempster theory [115], BA $+$ Artificial Bee Colony (ABC) [49], [114], [159], BA $+$ K-means $+$ Invasive Weed Optimization (IWO) [112], BBA $+$ MRMR [33], [167], IBA $+$ Shuffled complex evolution algorithm [35], [189], BA $+$ Differential Evolution (DE) strategies [89], [131], [188], Cuckoo Search Algorithm (CSA)-BA-ABC [113], BBA $+$ Spark framework [164], BBA $+$ Cross-entropy method [165], BA $+$ K-means [92], [101], BA $+$ Correlation-based FS [150], BA $+$ Cauchy mutation operator and Elite opposition- based learning [156], BA $+$ Shuffling and memeplex formulation [130], BA $+$ Flower pollination algorithm [157], BA $+$ Fisher criterion [158], rMRMR $+$ BBA with B hill climbing local search [163], BA $+$ Social insect termites [160], CSA $+$ BA [161], BA $+$ Opposition based [84], [162], BA $+$ Genetic Algorithm (GA) and IWO [152], BBA $+$ BPSO [30], Rough set theory $+$ BA [37], TRIZ inventive $+$ BA [154], BA $+$ Doppler effect in echoes [149]
Fitness Function	Based on rough set theory [26], Based on accuracy and number of features [31], [104], [107], Elitist store strategy [28], Built on different classifiers [61], Calculated in a hierarchical manner [132], Based highest accuracy among classifiers [133], Proposed fitness function based on the problem [37], [85], [86], [97], [98], [134], Multi-objective function [23], [80], [123], [124]
None	[13], [25], [29], [34]–[36], [60], [83], [103], [136]–[139], [151], [153], [166], [168]–[186]

BA involves several algorithm-specific parameters, which their interaction affects the performance of BA. Parameters such as, $\beta$ , $\varepsilon$ , ( $\alpha$ and $A$ ), ( $\gamma$ and $r$ ), which are placed within different components of BA like frequency, local search, loudness update, emission rate update equations, respectively. Through analysis of the previous table, it is clear that most of the proposed techniques, which aim to improve BA are concerned with tuning or controlling these parameters. In detail, different studies have used a fixed predefined single or range of values for various parameters of BA, which are determined either by preliminary experiments or taken from previous studies that have proven the effectiveness of such values in achieving better results. However, tuning techniques are quite time-consuming due to the large number of values that each parameter should try out until the optimal value is found [96, 186]. Consequently, researchers have exploited various controlling techniques. Recently, deterministic chaotic maps were used extensively to control almost all parameters of BA due to their distinctive characters i.e., irregularity, ergodicity, and pseudo-randomness [108]. Alternatively, other deterministic techniques like Gaussian walk, Levy flight, RSITFC, and Uniform were used and limited to control only ( $\varepsilon$ ) parameter in the local search equation. Such deterministic randomness techniques use a predefined function that modifies dynamically the parameter value over time [191]. Not only this but also they were reported to have the ability to improve convergence rate and hence, avoid BA to fall into local optimum solution [64].

In addition, adaptive and self-adaptive techniques were also exploited by some studies to control different parameters of BA as mentioned in Table 7. These techniques can autonomously determine parameter settings for optimal performance. The main idea behind adaptation techniques is to minimize the number of controllable parameters, hence decreasing the complexity of implementation as well as reducing the experimentation time significantly [100, 192, 193]. Important to realize that, among adaptive techniques feedback simple rule and fuzzy logic techniques were the most used techniques. This is due to their simplicity and low computational time in comparison to other adaptation strategies [191, 194]. On contrary, Self-adapted technique the parameter is directly coded in the solution vector as extra dimensions and hence, adjusting itself during the optimization process. In fact, self-adaptive basically follows the selection idea in Evolutionary Computation Algorithms (ECA). Given such features, it goes without saying that adaptation techniques become recently preferable strategies for a large number of researchers in the field of improving BA [100, 192, 193]. However, such techniques consume relatively higher computation time compared to the original BA, where the parameter’s values are fixed during the optimization process. This is because the modifications and strategies for setting up the parameters must be performed during the execution time of the algorithm.

Considering the most important parameters of BA named A and r [61, 62], a large number of works have been satisfied with utilizing the auto-zooming mechanism (see Table 7). This distinct inbuilt capability in BA is called automatic zooming and it distinguishes BA from other SI algorithms. Whereby, auto-zooming assists BA to transit into a region where there is a possible optimal solution as well as it helps BA to automatically switching from exploration move to local exploitation move. To make this auto-zooming work, different values are given to $\alpha$ and $\gamma$ parameters, so the values of A and r are dynamically updated throughout the iterations. However, there are still many challenges related to how to vary the values of such parameters as well as the answer to what are the most fitting values for different parameters of BA. It is still unclear and difficult to determine the most fitting values for different parameters of BA [24, 78]. Besides developing techniques for tuning and controlling BA’s parameters, other modifications and methods to different components of BA are introduced in order to improve its performance. Methods such as modifying the original formula of the components or incorporating some operators of other existing methods as it is concluded from Table 7. Furthermore, with the aim to make BA more powerful, it was combined with some SI algorithms and traditional techniques. This variant is called hybridized BA, various hybrid variants of BA were reported in Table 7. However, hybrid BA is still in its infancy stage as there is a wide range of recent SI that can be combined with BA to improve its performance [52].

Another critical component of BA is the fitness function. Generally, different available traditional evaluation metrics in the literature have been used as fitness function by many studies to assess the performance of BA as well as the model in which BA was used. In contrast, several researchers have improved some existing fitness functions or introduced new ones to suit the objective of their studies. Such functions were constructed using various criteria or built based on different strategies.

4. Applications of bat

Since the original BA has been introduced in 2010, it has been applied to solve many real-world optimization problems due to its efficiency and flexibility. The range area applications of BA and its variants have been increasing day by day, and the quality of results obtained using BA outperformed or matched those achieved using other popular SI algorithms. Because of the impressive advantages of BA, great research applications from different critical research domains have been attempted. Based on that, the authors of this review have defined five main groups of application domains and their sub-domains (see Table 8).

Table 8
Recent applications of bat for optimal feature selection

Domain	Sub-domain	Studies
Machine learning	Classification	[13], [25]–[27], [30], [37], [102], [103], [106], [107], [112],
		[115], [134]–[136], [164], [165], [181]
	Clustering	[29]
	Information retrieval	[137], [173]
	Intrusion detection	[31], [36], [92], [104], [111], [116], [117]
	Diabetes detection	[37]
	E-Fraud detection	[35]
	Spam and fraud detection	[21], [35], [154]
	Pattern recognition	[180], [184]
	Estimate parameters values for	[73], [144], [166], [171], [172], [182], [185], [186]
	machine learning techniques
	Training neural networks	[79], [151], [153]
	Mining association rules	[83], [123]
	Community detection	[32], [98], [110], [170]
Electric and electronic	Zero one knapsack problems	[22], [60], [75], [105], [109]
engineering	Design and tuning controllers	[23], [28], [59], [74], [82], [88], [108], [113], [124], [125],
		[127], [129], [147], [149], [155], [157], [160], [162], [167],
		[175]–[178], [187]
Computer science		[80], [150], [159], [174]
engineering
Bio-medical	Protein-protein interaction network	[96]
engineering	Gene detection	[33], [132], [154], [163]
Image processing		[34], [61], [63], [101], [138], [140]–[142], [146], [158], [169],
		[183]
Other real-life		[70], [78], [86], [100], [133], [139], [143], [155], [168], [188]
applications

Based on the survey, machine learning applications are the most used field for testing and applying BA and its variants. In particular, FS for classification applications. Electric and electronic engineering is one popular field too, where continuous variants of BA have been exhaustively and extensively applied. Microarray gene selection is another field in biomedical engineering which is becoming more famous because it plays a vital part in disease classifications. In the same way, computer science and image processing are getting more attention from researchers in recent years due to their direct influence on the quality of life of human beings.

5. Datasets

The authors of this study have classified the datasets which have been used to test the different variants of BA into four groups small, medium, large, and mixed as shown in Table 9. Small, large, and mixed categories are adapted from [4], who categorized the datasets into the small datasets (up to 150 features), large dataset (greater than 2000 features), and mixed dataset i.e., which includes a combination of datasets from the previous two categories. However, the authors of this survey introduce a new category which is called medium dataset that has features between (150 and 2000 features), meanwhile, the mixed dataset includes a collection of datasets from the three categories i.e., small, medium, and large. It is obvious from the Table 9, that the majority of works have done on small-size datasets. In contrast, only 14 studies have applied BA on high dimensional datasets and these studies involves gene detection [33, 132, 163], protein recognition [96], mining association rules [83, 123], community detection [110], classification benchmark datasets from UCI [112, 135] training neural network [79], retrieve medical articles [173], signal beats processing [168], E-fraud and spam detection [35, 154]. Furthermore, BA has employed for medium size datasets in applications like text categorization [164], speech processing voice signals [133], handwritten Arabic characters recognition [180], synthetic & real networks detection [32, 98, 170], knapsack problems [22, 60] travelling salesman problem [86], and UCI Classification [107, 134, 136]. Through analysis of collected papers, it was discovered that many works have tested and evaluated their improved BA on different benchmark test functions such as uni-model, multi-model, constrained, and unconstrained [93, 94, 95, 96, 97, 118, 119, 120, 121, 122]. Such studies have been categorized under a different category called artificially generated benchmarks. Apart from all previous categories, some researchers have applied their enhanced variants of BA first on benchmark functions in order to prove their effectiveness. Subsequently, they have implemented these variants to solve real-world problems using their corresponding datasets. These studies were categorized as others in this research.

Table 9
Dataset size used in evaluating BA performance for feature selection

Datasets	Studies
Small scale	[13], [21], [23], [25]–[31], [34], [36], [37], [59], [61], [63], [70], [73]–[75], [78], [80], [82], [87], [88], [92], [100]–[103], [105], [106], [108], [109], [111], [113], [115]–[117], [124], [125], [127], [129], [137]–[144], [146], [147], [149]–[151], [153], [155], [157]–[160], [165]–[167], [169], [171], [172], [174]–[178], [181]–[183], [185]–[188]
Medium scale	[22], [32], [60], [86], [98], [107], [133], [134], [136], [164], [170], [180]
Large scale	[33], [35], [79], [83], [96], [110], [112], [123], [132], [135], [154], [163], [168], [173]
Mixed	[27], [37], [59], [78], [105], [149], [166]
Artificially generated benchmarks	[45], [49], [59], [62], [64], [66], [68], [69], [71], [72], [76], [77], [81], [84], [85], [89]–[91], [93]–[97], [99], [105], [113], [114], [118]–[122], [128], [130], [131], [145], [148], [149], [152], [156]–[158], [161], [179], [189]
Others	[96], [113], [157], [158]

6. Discussion and future directions

In this section, the answers to the questions that were raised at the beginning of this work (Section 2.3.2) were presented and discussed as well as some future works related to each question were recommended.

RQ1: What are the recent variants of BA in the context of optimal FS technique for solving various optimization problems? Generally, this study has collected 147 papers that either used the standard BA or its variants for optimal FS. From these 147 papers, 22 studies have used the standard BA without any modification, which are specified by NN column in the Table 6 (No. 126-147). On the other hand, 125 studies were recognized as variants for BA as the researchers for such studies have introduced several modifications on the original structure of BA. These variants were divided into two categories. The first category contains 85 studies that have modified different components of BA, for example on the IP, FE, VE, NSG, LS, A, r, and/or Ft as illustrated in Table 6 (No. 1-14, 16-17, 19, 21-22, 24-31, 33-36, 38-48, 51-53, 55-58, 61, 66-80, 84-103). The second category includes 40 studies which have hybrid other SI algorithm/existing techniques into the original components of BA either to improve one or more components of BA, for instance Yılmaz and Küçüksille [59] has use the hybrid of invasive weed seed algorithm to modify VE and NSG, or as an added value to BA as a FS techniques. Alomari et al. [33] have hybrid BA with Minimum Redundancy Maximum Relevancy (MRMR) method for gene selection. These studies are indicated by Hy column in Table 6 (No. 15, 18, 20, 23, 32, 37, 49-50, 54, 59-60, 62-65, 81-83, 104-125).

RQ2: What is the most frequently used encoding scheme in BA? According to Table 6 and Fig. 3, it can be easily observed that a continuous encoding scheme is the most used scheme, meanwhile binary and discrete schemes are quite limited. This is because BA has originally introduced to deal with continuous optimization problems. However, continuous BA represents the search space as N-dimensional continuous values i.e., real values within [0, 1], in which the search process only ends when the stopping criteria is met. Hence, it does not offer the ability to explore the whole feature space. In contrast, each BA in a binary scheme has a single-dimensional vector whose length depends on the number of features in the given dataset. Several works have used BBA to select the optimal features and their experiments have shown its effectiveness in achieving better results in comparison with the continuous BA and against other popular benchmark binary SI algorithms [102, 109, 137, 138, 166]. Despite the success of BBA, studies on it remain relatively small with an average of 20% compared to the overall papers collected for the purpose of this review. Therefore, more investigations are needed on this version.

RQ3: What is the most common objective function used in BA’s applications? By looking at Table 6 and Fig. 4, it is observed that a wide range of studies has utilized a single objective function. Many researchers were interested in achieving only a single objective function i.e., either maximizing the model’s performance or minimizing cost and $\backslash$ or time. In contrast, applications of multi-objective BA are relatively less, and this include applications like engineering applications [19], community detection [32, 98], and mining association rules [83, 123]. Scientists in such applications are most concerned about reducing the cost of building a model and at the same time trying to increase the performance of the model.

RQ4: Which components of BA algorithms have been improved? And RQ5: What are the techniques that have been used to improve the different components of BA? Table 7 shows the different parts of BA that have been improved over the last ten years. It is clear that a large number of works have focused on tuning and controlling different parameters of BA e.g., $\beta$ , $\varepsilon$ , $\alpha$ , A, $\gamma$ , and $r$ . The majority of existing studies have tuned such parameters by trial and error which is time-consuming. Hence, there is a necessity to develop controlling techniques that have the capability of self-tuning its parameters automatically. In addition, many researchers are satisfied with the inbuilt feature of BA named auto-zooming technique that is used to control $A$ and $r$ parameters. This technique helps BA to automatically switch from exploration search to local exploitation search. Such search strategies are very important processes of BA in order to find optimal solutions. Despite the importance of $A$ and $r$ parameters on controlling the exploration and exploitation processes, yet almost all studies that applied BA on high-dimensional datasets have utilized either fixed values [96, 79, 132, 135, 173] or auto-zooming mechanisms [33, 35, 83, 112, 123, 154, 163, 168]. For small and medium datasets, researchers have proposed various controlling techniques like deterministic, adaptive, and self-adaptive techniques as depicted in Tables 7 and 9. Among such techniques, deterministic strategies are extensively used to control almost all parameters of BA. In contrast, few numbers of studies have utilized adaptive and self-adaptive techniques to control parameters of BA. Despite their advantages in minimizing the number of controllable parameters as well as reducing the experimentation time significantly, it is still not known if there is a technique that could automatically control parameters of BA to get its optimal performance [24, 56, 78]. Thus, more controlling techniques are required to be introduced and detailed parametric experiments should be carried out for solving real-world problems. Parameters of BA are not the only matter that gained the attention of scientists, but also various components of BA (e.g., frequency, velocity, new solution generation, and local search equations) have attracted a fair amount of attention (refer Table 7). On one hand, some scholars modified the original formula of the components while many others have incorporated various conventional techniques or hybrid BA with different SI algorithms. Examples of traditional techniques includes K-means, mean shift, MI, MRMR, rough set theory, and so forth. Most SI algorithms that were hybridized with BA, are PSO, ABC, GA, DE, and CS. However, given the numerous modern SI algorithms and many other traditional techniques that exist at present, more possible hybrid BA versions can be developed improve its performance.

RQ6: What are the prominent applications of BA as an optimal feature selection technique? And RQ7: What is the size of data that BA and its variants have been applied on? Table 9 shows the various applications of BA. Machine learning applications represent the largest part of the applications. More specifically, classification applications. Reason for this is because such applications were applied to benchmark datasets which are labelled and can easily be obtained from the literature. In contrast, clustering applications for BA are still in its infancy stage and hence, further development and research are needed to be done in this field. To highlight, BA and its variants have been applied for solving feature selection problems in different detection applications such as intrusion detection [31, 36, 92, 104, 111, 116, 117], spam detection [17], community detection [28], gene detection [29], e-fraud detection [35], and diabetes disease [37], etc. However, many of such studies have been implemented over small size datasets as it is noted by investigating Tables 8 and 9. Thus, there is a massive scale for further research on BA and its variants to be tested on medium or large-size datasets.

RQ8: Is there any application of BA and its versions on event $\backslash$ topic detection from textual data? The authors have identified that there is no work found so far which has applied BA or any of its variants on event $\backslash$ topic detection applications from textual data. On this basis, the authors of this study aim to fill this gap in their future work by employing one variant of BA to solve the FS problem in event detection application from textual datasets.

7. Conclusion

This work presents the first systematic literature review of BA as an optimal selection technique considering qualitative and quantitative aspects. The study was conducted by collecting 147 papers published in the period from 2010 up to the beginning of 2020, using four well-known scientific databases: ACM, IEEE Xplore, Springer, and ScienceDirect. The collected papers were analysed and categorized based on the encoding schemes, objective function, various components of BA that have been improved using different techniques as well as the size of data used, and the main real-life applications of BA which were thoroughly reported. The main focus of this paper is to analyse, discuss, and criticized the latest variants of BA for FS. Based on the investigation, it is observed that the majority of articles have utilized a continuous encoding scheme, while few works have applied binary encoding scheme for FS. However, such variant was able to achieve significant results compared to the continuous version and other binary SI algorithms. Searching for the most used objective functions, it is noted that many papers have attempted to accomplish single-objective function, and just seven articles were interested to achieve multi-objective function. Despite the small number of studies that were found in this study and which were interested in the use of binary BA and multi-objective BA, such variants have gained more focus in recent times due to their promising performances in various fields. The performance of the swarm in BA is affected by several parameters, so most of the modifications have been done by introducing many techniques to control such parameters. More precisely, the study reveals that a large portion of techniques have been proposed to control $A$ and $r$ parameters. Regarding BA’s applications, most of studies have applied BA or its variants to solve FS problems in different machine learning applications, mainly, classification applications. As a consequence, the majority of studies have used mostly small or medium benchmark datasets to evaluate their performances. Finally, the study has confirmed that there is no work is done so far which focuses on using BA or any of its variants to solve FS problems for event $\backslash$ topic detection applications from textual data. Finally, this survey has drawn a future map for the researchers and practitioners who are currently working or will work on enhancing the performance of standard BA. In addition, it may provide guidelines for the researchers who are active in the area of BA as well as encourage scientists from various disciplines to better understand the awareness about BA and its variants, to enhance them, or to develop a new one.

Footnotes

Acknowledgments

Authors would like to thank editors and anonymous reviewers for their valuable comments.

References

Nguyen

B.H.

Xue

and Zhang

, A survey on swarm intelligence approaches to feature selection in data mining, Swarm Evol. Comput 54 (2020), 100663.

Anuradha

, A brief introduction on Big Data 5Vs characteristics and Hadoop technology, Procedia Comput. Sci 48 (2015), 319–324.

Cherrington

Airehrour

Thabtah

and Madanian

, Particle Swarm Optimization for Feature Selection: A Review of Filter-based Classification to Identify Challenges and Opportunities, in: 2019 IEEE 10th Annual Information Technology, Electronics and Mobile Communication Conference (IEMCON), 2019, pp. 523–529.

Brezočnik

Fister

and Podgorelec

, Swarm intelligence algorithms for feature selection: A review, Appl. Sci 8(9) (2018), 1521.

Forman

, An extensive empirical study of feature selection metrics for text classification, J. Mach. Learn. Res 3 (2003), 1289–1305.

Gomez

J.C.

Boiy

and Moens

M.-F.

, Highly discriminative statistical features for email classification, Knowl. Inf. Syst 31(1) (2012), 23–53.

Kumar

and Minz

, Poem classification using machine learning approach, in: Proceedings of the Second International Conference on Soft Computing for Problem Solving (SocProS 2012), December 28–30, 2012, 2014, pp. 675–682.

Egozi

Gabrilovich

and Markovitch

, Concept-Based Feature Generation and Selection for Information Retrieval, in: AAAI, Vol. 8, 2008, pp. 1132–1137.

J.G.

Brodley

C.E.

Kak

Broderick

L.S.

and Aisen

A.M.

, Unsupervised feature selection applied to content-based retrieval of lung images, IEEE Trans. Pattern Anal. Mach. Intell 25(3) (2003), 373–378.

10.

Lee

Stolfo

S.J.

and Mok

K.W.

, Adaptive intrusion detection: A data mining approach, Artif. Intell. Rev 14(6) (2000), 533–567.

11.

Punlumjeak

and Rachburee

, A comparative study of feature selection techniques for classify student performance, in: Information Technology and Electrical Engineering (ICITEE), 2015 7th International Conference on, 2015, pp. 425–429.

12.

Bharti

K.K.

and Singh

P.K.

, A three-stage unsupervised dimension reduction method for text clustering, J. Comput. Sci 5(2) (2014), 156–169.

13.

Yusof

M.M.

and Wahidi

, A Comparative Study of Feature Selection Techniques for Bat Algorithm in Various Applications, Vol. 06006, 2018, 1–6.

14.

Ahmad

F.K.

Md Norwawi

deris

and Othman

, A review of feature selection techniques via gene expression profiles, 2008.

15.

Kashef

and Nezamabadi-pour

, An Advanced ACO Algorithm for Feature subset Selection, Neurocomputing, 2014. doi: 10.1016/j.neucom.2014.06.067.

16.

Jeyaraj

, Comparison of Feature Selection Strategies for Classification using Rapid Miner, no. July 2016, 2018. doi: 10.15680/IJIRCCE.2016.

17.

Review

, applied sciences Swarm Intelligence Algorithms for Feature Selection, 2018. doi: 10.3390/app8091521.

18.

Ali

A.F.

, Accelerated bat algorithm for solving integer programming problems, Egypt. Comput. Sci. J 39(1) (2015), 507–518.

19.

Yang

X.-S.

, Bat algorithm for multi-objective optimisation, arXiv Prepr. arXiv1203.6571, 2012.

20.

Yang

X.-S.

, A new metaheuristic bat-inspired algorithm, Nat. inspired Coop. Strateg. Optim. (NICSO 2010), 2010, 65–74.

21.

Rajalaxmi

R.R.

and Ramesh

, Binary bat approach for effective spam classification in online social networks, Aust. J. Basic Appl. Sci 18 (2014), 383–388.

22.

Zhou

and Ma

, A complex-valued encoding bat algorithm for solving 0–1 knapsack problem, Neural Process. Lett 44(2) (Oct. 2016), 407–430. doi: 10.1007/s11063-015-9465-y.

23.

Yuan

Wang

and Yuan

, Application of improved bat algorithm in optimal power flow problem, Appl. Intell 48(8) (Aug. 2018), 2304–2314. doi: 10.1007/s10489-017-1081-2.

24.

Jayabarathi

Raghunathan

and Gandomi

A.H.

, The Bat Algorithm, Variants and Some Practical Engineering Applications: A Review, in: Nature-Inspired Algorithms and Applied Optimization, Springer, 2018, pp. 313–330.

25.

Nakamura

R.Y.M.

Pereira

L.A.M.

Costa

K.A.

Rodrigues

Papa

J.P.

and Yang

X.-S.

, BBA: a binary bat algorithm for feature selection, in: Graphics, Patterns and Images (SIBGRAPI), 2012 25th SIBGRAPI Conference on, 2012, pp. 291–297.

26.

Emary

Yamany

and Hassanien

A.E.

, New approach for feature selection based on rough set and bat algorithm, in: Computer Engineering & Systems (ICCES), 2014 9th International Conference on, 2014, pp. 346–353.

27.

Taha

A.M.

Mustapha

and Chen

S.-D.

, Naive bayes-guided bat algorithm for feature selection, Sci. World J 2013 (2013).

28.

Zhao

and He

, Chaotic binary bat algorithm for analog test point selection, Analog Integr. Circuits Signal Process 84(2) (2015), 201–214.

29.

Rani

A.S.S.

and Rajalaxmi

R.R.

, Unsupervised feature selection using binary bat algorithm, in: Electronics and Communication Systems (ICECS), 2015 2nd International Conference on, 2015, pp. 451–456.

30.

Tawhid

M.A.

and Dsouza

K.B.

, Hybrid Binary Bat Enhanced Particle Swarm Optimization Algorithm for solving feature selection problems, Appl. Comput. Informatics, 2018.

31.

Enache

and Science

, Intelligent Feature Selection Method rooted in Binary Bat Algorithm for Intrusion Detection, 2015, 517–521.

32.

Sharma

and Annappa

, Community detection using meta-heuristic approach: Bat algorithm variants, in: 2016 Ninth International Conference on Contemporary Computing (IC3), Aug. 2016, pp. 1–7. doi: 10.1109/IC3.2016.7880209.

33.

Alomari

O.A.

Khader

A.T.

Al-Betar

M.A.

and Abualigah

L.M.

, Gene selection for cancer classification by combining minimum redundancy maximum relevancy and bat-inspired algorithm, Int. J. Data Min. Bioinform 19(1) (2017), 32–51.

34.

Rattan

Kaur

Kansal

and Kaur

, An optimized lung cancer classification system for computed tomography images, in: 2017 Fourth International Conference on Image Information Processing (ICIIP), 2017, pp. 1–6.

35.

Akinyelu

A.A.

and Adewumi

A.O.

, On the performance of cuckoo search and bat algorithms based instance selection techniques for SVM speed optimization with application to e-fraud detection, KSII Trans. Internet Inf. Syst 12(3) (2018).

36.

Enache

A.-C.

and Sgarciu

, Anomaly intrusions detection based on support vector machines with bat algorithm, in: 2014 18th International Conference on System Theory, Control and Computing (ICSTCC), 2014, pp. 856–861.

37.

Cheruku

Edla

D.R.

Kuppili

and Dharavath

, RST-BatMiner: A fuzzy rule miner integrating rough set feature selection and bat optimization for detection of diabetes disease, Appl. Soft Comput 67 (Jun. 2018), 764–780. doi: 10.1016/J.ASOC.2017.06.032.

38.

Fister

Yang

X.-S.

Fong

and Zhuang

, Bat algorithm: recent advances, in: Computational Intelligence and Informatics (CINTI), 2014 IEEE 15th International Symposium on, 2014, pp. 163–167.

39.

Nurwidyantoro

and Winarko

, Event detection in social media: A survey, in: ICT for Smart Society (ICISS), 2013 International Conference on, 2013, pp. 1–5.

40.

Goswami

and Kumar

, A survey of event detection techniques in online social networks, Soc. Netw. Anal. Min 6(1) (2016), 107.

41.

Hassanian-esfahani

and Kargar

, A survey on web news retrieval and mining, in: Web Research (ICWR), 2016 Second International Conference on, 2016, pp. 90–101.

42.

Panagiotou

Katakis

and Gunopulos

, Detecting events in online social networks: Definitions, trends and challenges, in: Lecture Notes in Computer Science (Including Subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), Springer, Vol. 9580, 2016, pp. 42–84.

43.

Ramadan

Q.H.

and Mohd

, A review of retrospective news event detection, in: Semantic Technology and Information Retrieval (STAIR), 2011 International Conference on, 2011, pp. 209–214.

44.

Dai

and Sun

, A Two-layer text clustering approach for retrospective news event detection, in: Artificial Intelligence and Computational Intelligence (AICI), 2010 International Conference on, Vol. 1, 2010, pp. 364–368.

45.

Chakri

Khelif

Benouaret

and Yang

X.-S.

, New directional bat algorithm for continuous optimization problems, Expert Syst. Appl 69 (2017), 159–175.

46.

Bharti

K.K.

and Singh

P.K.

, Opposition chaotic fitness mutation based adaptive inertia weight BPSO for feature selection in text clustering, Appl. Soft Comput 43 (2016), 20–34.

47.

Mafarja

M.M.

and Mirjalili

, Hybrid Whale Optimization Algorithm with simulated annealing for feature selection, Neurocomputing 260 (2017), 302–312.

48.

Arora

and Anand

, Binary butterfly optimization approaches for feature selection, Expert Syst. Appl 116 (2019), 147–160.

49.

Zhang

Yuan

Sun

Shi

and Zhou

, A two-stage framework for bat algorithm, Neural Comput. Appl 28(9) (2017), 2605–2619.

50.

Yang

X.-S.

and He

, Bat algorithm: Literature review and applications, Int. J. Bio-Inspired Comput 5(3) (2013), 141–149.

51.

I.F.

Fister

and Yang

, Bat algorithm: Recent advances, 2014, 163–167.

52.

Chawla

and Duhan

, Bat algorithm: A survey of the state-of-the-art, Appl. Artif. Intell 29(6) (2015), 617–634.

53.

Induja

and Eswaramurthy

V.P.

, Bat algorithm: An overview and its applications, Int. J. Adv. Res. Comput. Commun. Eng 5(1) (2016).

54.

Sharma

Luhach

A.K.

and Jyoti

, Research & analysis of advancements in BAT algorithm, in: 2016 3rd International Conference on Computing for Sustainable Global Development (INDIACom), 2016, pp. 2391–2396.

55.

Yadav

S.L.

and Phogat

, A review on bat algorithm, Int. J. Comput. Sci. Eng 5(7) (2017).

56.

Bangyal

W.H.

Ahmad

Rauf

H.T.

and Pervaiz

, An overview of mutation strategies in bat algorithm, Int. J. Adv. Comput. Sci. Appl 9(8) (2018), 523–534.

57.

Kongkaew

, Bat algorithm in discrete optimization: A review of recent applications 39(5) (2017), 641–650.

58.

Kitchenham

, Procedures for performing systematic reviews, Keele, UK, Keele Univ 33(2004) (2004), 1–26.

59.

Yılmaz

and Küçüksille

E.U.

, A new modification approach on bat algorithm for solving optimization problems, Appl. Soft Comput 28 (2015), 259–275.

60.

Sabba

and Chikhi

, A discrete binary version of bat algorithm for multidimensional knapsack problem, Int. J. bio-inspired Comput 6(2) (2014), 140–152.

61.

Gupta

Arora

Agrawal

Khanna

and de Albuquerque

V.H.C.

, Optimized Binary Bat Algorithm for classification of White Blood Cells, Measurement, 2019.

62.

Gandomi

A.H.

and Yang

X.-S.

, Chaotic bat algorithm, J. Comput. Sci 5(2) (2014), 224–232.

63.

Dhal

K.G.

and Das

, A dynamically adapted and weighted Bat algorithm in image enhancement domain, Evol. Syst., 2018, 1–19.

64.

I.F.

Yang

Brest

Fister

and Fister

, Analysis of randomisation methods in swarm intelligence, Int. J. bio-inspired Comput 7(1) (2015), 36–49.

65.

Parvin

AsgharNadri

and Rad

, Optimal Feature Selection for Data Classification and Clustering: Techniques and Guidelines, 2016.

66.

Lyu

Huang

Wang

and Hu

, Improved self-adaptive bat algorithm with step-control and mutation mechanisms, J. Comput. Sci 30 (2019), 65–78.

67.

Agarwal

and Ranjan

, Dimensionality R eduction Methods Clas-Reduction sical and Recent Trends: A Survey Surv Recent 9(10) (2016), 4801–4808.

68.

Fister

Jr Fister

and Yang

X.-S.

, Towards the development of a parameter-free bat algorithm, in: StuCoSReC: Proceedings of the 2015 2nd Student Computer Science Research Conference, 2015, pp. 31–34.

69.

Jordehi

A.R.

, Chaotic bat swarm optimisation (CBSO), Appl. Soft Comput 26 (2015), 523–530.

70.

Abdel-Raouf

Abdel-Baset

and El-Henawy

, An improved chaotic bat algorithm for solving integer programming problems, Int. J. Mod. Educ. Comput. Sci 6(8) (2014), 18.

71.

Afrabandpey

Ghaffari

Mirzaei

and Safayani

, A novel bat algorithm based on chaos for optimization tasks, in: Intelligent Systems (ICIS), 2014 Iranian Conference on, 2014, pp. 1–6.

72.

Cai

Wang

Kang

and Wu

, Bat algorithm with Gaussian walk, Int. J. Bio-Inspired Comput 6(3) (2014), 166–174.

73.

Jensi

and Jiji

G.W.

, MBA-LF: A new data clustering method using modified bat algorithm and levy flight, ICTACT J. Soft Comput 6(1) (2015).

74.

Baziar

Kavoosi-Fard

and Zare

, A novel self adaptive modification approach based on bat algorithm for optimal management of renewable MG, J. Intell. Learn. Syst. Appl 5(1) (2013), 11.

75.

Liu

and Qi

, A Self-adaptive Bat Algorithm for Camera Calibration, in: 2nd International Conference on Advances in Mechanical Engineering and Industrial Informatics (AMEII 2016), 2016.

76.

Perez

Valdez

Castillo

and Roeva

, Bat algorithm with parameter adaptation using interval type-2 fuzzy logic for benchmark mathematical functions, in: 2016 IEEE 8th International Conference on Intelligent Systems (IS), 2016, pp. 120–127.

77.

Pérez

Valdez

and Castillo

, Modification of the bat algorithm using fuzzy logic for dynamical parameter adaptation, in: 2015 IEEE Congress on Evolutionary Computation (CEC), 2015, pp. 464–471.

78.

Barbosa

C.E.M.

and Vasconcelos

G.C.

, Eight Bio-inspired Algorithms Evaluated for Solving Optimization Problems, in: International Conference on Artificial Intelligence and Soft Computing, 2018, pp. 290–301.

79.

Niu

Wang

Zhang

and Du

, Multi-step-ahead wind speed forecasting based on optimal feature selection and a modified bat algorithm with the cognition strategy, Renew. Energy 118 (Apr. 2018), 213–229. doi: 10.1016/J.RENENE.2017.10.075.

80.

Mohamed

T.M.

and Moftah

H.M.

, Simultaneous ranking and selection of keystroke dynamics features through a novel multi-objective binary bat algorithm, Futur. Comput. Informatics J 3(1) (Jun. 2018), 29–40. doi: 10.1016/J.FCIJ.2017.11.005.

81.

Ramli

M.R.

Abas

Z.A.

Desa

M.I.

Abidin

Z.Z.

and Alazzam

M.B.

, Enhanced convergence of bat algorithm based on dimensional and inertia weight factor, J. King Saud Univ. – Comput. Inf. Sci 31(4) (Oct. 2019), 452–458. doi: 10.1016/J.JKSUCI.2018.03.010.

82.

Mahdad

, Solution of non-smooth economic dispatch using interactive grouped adaptive bat algorithm, Int. J. Energy Optim. Eng 8(1) (Jan. 2019), 88–114. doi: 10.4018/IJEOE.2019010105.

83.

Heraguemi

K.E.

Kamel

and Drias

, Multi-objective bat algorithm for mining numerical association rules, Int. J. Bio-Inspired Comput 11(4) (2018), 239. doi: 10.1504/IJBIC.2018.092797.

84.

Shan

Liu

and Sun

P.-L.

, Modified bat algorithm based on lévy flight and opposition based learning, Sci. Program 2016 (2016), 1–13. doi: 10.1155/2016/8031560.

85.

and Zhou

, A novel complex-valued bat algorithm, Neural Comput. Appl 25(6) (2014), 1369–1381.

86.

Saji

and Riffi

M.E.

, A novel discrete bat algorithm for solving the travelling salesman problem, Neural Comput. Appl 27(7) (Oct. 2016), 1853–1866. doi: 10.1007/s00521-015-1978-9.

87.

Cai

Wang

Kang

and Wu

, Adaptive bat algorithm for coverage of wireless sensor network, Int. J. Wirel. Mob. Comput 8(3) (2015), 271. doi: 10.1504/IJWMC.2015.069411.

88.

Osaba

Yang

X.-S.

Diaz

Lopez-Garcia

and Carballedo

, An improved discrete bat algorithm for symmetric and asymmetric Traveling Salesman Problems, Eng. Appl. Artif. Intell 48 (Feb. 2016), 59–71. doi: 10.1016/J.ENGAPPAI.2015.10.006.

89.

Alihodzic

and Tuba

, Improved Hybridized Bat Algorithm for Global Numerical Optimization, in: 2014 UKSim-AMSS 16th International Conference on Computer Modelling and Simulation, Mar. 2014, pp. 57–62. doi: 10.1109/UKSim.2014.97.

90.

Cai

Gao

and Xue

, Improved bat algorithm with optimal forage strategy and random disturbance strategy, Int. J. Bio-Inspired Comput 8(4) (2016), 205–214.

91.

Perez

Valdez

Castillo

Melin

Gonzalez

and Martinez

, Interval type-2 fuzzy logic for dynamic parameter adaptation in the bat algorithm, Soft Comput 21(3) (Feb. 2017), 667–685. doi: 10.1007/s00500-016-2469-3.

92.

H.Z.J.

Zhao

and Li

, Ai-based two-stage intrusion detection for software defined iot networks, IEEE Internet Things J 6(2) (2019), 2093–2102.

93.

Ahmadi

A.H.

and Nikravesh

S.K.Y.

, A novel instantaneous exploitation based bat algorithm, in: 2016 24th Iranian Conference on Electrical Engineering (ICEE), May 2016, pp. 1751–1756. doi: 10.1109/IranianCEE.2016.7585804.

94.

Cui

and Kang

, Bat algorithm with inertia weight, in: 2015 Chinese Automation Congress (CAC), Nov. 2015, pp. 792–796. doi: 10.1109/CAC.2015.7382606.

95.

Perez

Valdez

and Castillo

, Proposed augmentation of the Bat Algorithm using fuzzy logic for dynamic parameter adaptation, in: 2015 Annual Conference of the North American Fuzzy Information Processing Society (NAFIPS) Held Jointly with 2015 5th World Conference on Soft Computing (WConSC), Aug. 2015, pp. 1–6. doi: 10.1109/NAFIPS-WConSC.2015.7284132.

96.

Chowdhury

Rakshit

Konar

and Nagar

A.K.

, A modified bat algorithm to predict Protein-Protein Interaction network, in: 2014 IEEE Congress on Evolutionary Computation (CEC), Jul. 2014, pp. 1046–1053. doi: 10.1109/CEC.2014.6900518.

97.

Naik

S.M.

Jagannath

R.P.K.

and Kuppili

, Bat algorithm-based weighted laplacian probabilistic neural network, Neural Comput. Appl 32(4) (Feb. 2020), 1157–1171. doi: 10.1007/s00521-019-04475-4.

98.

Zhou

Zhao

and Liu

, A multiobjective discrete bat algorithm for community detection in dynamic networks, Appl. Intell 48(9) (Sep. 2018), 3081–3093. doi: 10.1007/s10489-017-1135-5.

99.

Mirjalili

S.M.

and Yang

X.-S.

, Binary bat algorithm, Neural Comput. Appl 25(3–4) (Sep. 2014), 663–681. doi: 10.1007/s00521-013-1525-5.

100.

Wang

Zhou

Gao

and Liu

, A self-adaptive bat algorithm for the truck and trailer routing problem, Eng. Comput 35(1) (2018), 108–135.

101.

Gan

J.E.

and Lai

W.K.

, Automated Grading of Edible Birds Nest Using Hybrid Bat Algorithm Clustering Based on K-Means, in: 2019 IEEE International Conference on Automatic Control and Intelligent Systems (I2CACIS), 2019, pp. 73–78.

102.

Zade

B.M.H.

and Aliahmadipour

, A boosting approach based on bat optimization in MLP neural networks: Classification task, in: 2018 6th Iranian Joint Congress on Fuzzy and Intelligent Systems (CFIS), 2018, pp. 83–86.

103.

Jaddi

N.S.

Abdullah

and Hamdan

A.R.

, Optimization of neural network model using modified bat-inspired algorithm, Appl. Soft Comput 37 (2015), 71–86.

104.

Enache

A.-C.

and Sgârciu

, A feature selection approach implemented with the binary bat algorithm applied for intrusion detection, in: 2015 38th International Conference on Telecommunications and Signal Processing (TSP), 2015, pp. 11–15.

105.

Huang

Zeng

and Han

, Dynamic inertia weight binary bat algorithm with neighborhood search, Comput. Intell. Neurosci 2017 (2017).

106.

Taha

A.M.

Chen

S.-D.

and Mustapha

, Bat algorithm based hybrid filter-wrapper approach, Adv. Oper. Res 2015 (2015).

107.

Yang

Zhu

Yang

Liu

and Yin

, Feature selection based on modified bat algorithm, IEICE Trans. Inf. Syst 100(8) (2017), 1860–1869.

108.

Lin

J.-H.

Chou

C.-W.

Yang

C.-H.

and Tsai

H.-L.

, A chaotic Levy flight bat algorithm for parameter estimation in nonlinear dynamic biological systems, Comput. Inf. Technol 2(2) (2012), 56–63.

109.

Rizk-Allah

R.M.

and Hassanien

A.E.

, New binary bat algorithm for solving 0–1 knapsack problem, Complex Intell. Syst 4(1) (2018), 31–53.

110.

Messaoudi

and Kamel

, A multi-objective bat algorithm for community detection on dynamic social networks, Appl. Intell 49(6) (2019), 2119–2136.

111.

Enache

A.-C.

and Sgârciu

, Anomaly intrusions detection based on support vector machines with an improved bat algorithm, in: 2015 20th International Conference on Control Systems and Computer Science, 2015, pp. 317–321.

112.

Srividhya

and Mallika

, Multimodal feature selection using invasive weed optimization and improved BAT for high dimensional imbalanced datasets, Int. J. Appl. Eng. Res 13(2) (2018), 960–966.

113.

Murugan

Mohan

M.R.

Rajan

C.C.A.

Sundari

P.D.

and Arunachalam

, Hybridizing bat algorithm with artificial bee colony for combined heat and power economic dispatch, Appl. Soft Comput 72 (2018), 189–217.

114.

Pan

J.-S.

Dao

T.-K.

Kuo

M.-Y.

and Horng

M.-F.

, Hybrid bat algorithm with artificial bee colony, in: Intelligent Data analysis and its Applications, Volume II, Springer, 2014, pp. 45–55.

115.

Rozlini

Munirah

M.Y.

Nawi

N.M.

and Wahid

, Bat with Dempster theory for feature selection: A framework 2(14) (2015), 158–165.

116.

Enache

A.C.

Sgarciu

and Togan

, Comparative Study on Feature Selection Methods Rooted in Swarm Intelligence for Intrusion Detection, in: Proceedings – 2017 21st International Conference on Control Systems and Computer, CSCS 2017, 2017, pp. 239–244. doi: 10.1109/CSCS.2017.40.

117.

Enache

A.-C.

and Sgârciu

, Enhanced Intrusion Detection System Based on Bat Algorithm-support Vector Machine, in: Proceedings of the 11th International Conference on Security and Cryptography, 2014, pp. 184–189. doi: 10.5220/0005015501840189.

118.

Hasan

S.S.

et al., A novel fuzzy inspired bat algorithm for multidimensional function optimization problem, Int. J. Fuzzy Syst. Appl 8(1) (Jan. 2019), 83–100. doi: 10.4018/IJFSA.2019010105.

119.

Shan

and Cheng

, Modified bat algorithm based on covariance adaptive evolution for global optimization problems, Soft Comput 22(16) (Aug. 2018), 5215–5230. doi: 10.1007/s00500-017-2952-5.

120.

Chen

Y.-T.

Liao

B.-Y.

Lee

C.-F.

Tsay

W.-D.

and Lai

M.-C.

, An Adjustable Frequency Bat Algorithm Based on Flight Direction to Improve Solution Accuracy for Optimization Problems, in: 2013 Second International Conference on Robot, Vision and Signal Processing, Dec. 2013, pp. 172–177. doi: 10.1109/RVSP.2013.47.

121.

Topal

A.O.

and Altun

, Dynamic Virtual Bats Algorithm (DVBA) for Global Numerical Optimization, in: 2014 International Conference on Intelligent Networking and Collaborative Systems, Sep. 2014, pp. 320–327. doi: 10.1109/INCoS.2014.40.

122.

Wang

and Wang

, An Adaptive Bat Algorithm, Springer, Berlin, Heidelberg, 2013, 216–223.

123.

Heraguemi

K.E.

Kamel

and Drias

, Multi-swarm bat algorithm for association rule mining using multiple cooperative strategies, Appl. Intell 45(4) (Dec. 2016), 1021–1033. doi: 10.1007/s10489-016-0806-y.

124.

Seyman

M.N.

, Adaptive arrangement of cyclic prefix length for MC-CDMA systems via multi-objective bat algorithm, Neural Comput. Appl 30(7) (Oct. 2018), 2319–2326. doi: 10.1007/s00521-017-3188-0.

125.

Chakri

Yang

X.-S.

Khelif

and Benouaret

, Reliability-based design optimization using the directional bat algorithm, Neural Comput. Appl 30(8) (Oct. 2018), 2381–2402. doi: 10.1007/s00521-016-2797-3.

126.

Reddy

M.P.

Mukherjee

and Ganguli

, Optimal design of damage tolerant composite using ply angle dispersion and enhanced bat algorithm, Neural Comput. Appl., Sep. 2019, 1–20. doi: 10.1007/s00521-019-04455-8.

127.

Bavafa

Azizipanah-Abarghooee

and Niknam

, New self-adaptive bat-inspired algorithm for unit commitment problem, IET Sci. Meas. Technol 8(6) (Nov. 2014), 505–517. doi: 10.1049/iet-smt.2013.0252.

128.

Zhao

G.W.M.L.X.

, An improved bat algorithm with variable neighborhood search for global optimization, in: 2016 IEEE Congress on Evolutionary Computation (CEC), 2016, pp. 1773–1778.

129.

Rahimi

Bavafa

Aghababaei

Khooban

M.H.

and Naghavi

S.V.

, The online parameter identification of chaotic behaviour in permanent magnet synchronous motor by Self-Adaptive Learning Bat-inspired algorithm, Int. J. Electr. Power Energy Syst 78 (2016), 285–291.

130.

Banati

and Chaudhary

, Enhanced shuffled bat algorithm (EShBAT), in: 2016 International Conference on Advances in Computing, Communications and Informatics (ICACCI), Sep. 2016, pp. 731–738. doi: 10.1109/ICACCI.2016.7732134.

131.

Fister

Fong

and Brest

, A novel hybrid self-adaptive bat algorithm, Sci. World J 2014 (2014).

132.

Bansal

and Sahoo

, Full model selection using bat algorithm, in: 2015 International Conference on Cognitive Computing and Information Processing (CCIP), 2015, pp. 1–4.

133.

Reddy

P.V.B.

Nandyala

S.P.

and Devi

J.S.

, Speaker verification with optimized feature subset using MOBA, in: 2016 IEEE Distributed Computing, VLSI, Electrical Circuits and Robotics (DISCOVER), 2016, pp. 101–106.

134.

Xie

et al., A novel test-cost-sensitive attribute reduction approach using the binary bat algorithm, Knowledge-Based Syst 186 (Dec. 2019), 104938. doi: 10.1016/J.KNOSYS.2019.104938.

135.

Taghian

Nadimi-Shahraki

M.H.

and Zamani

, Comparative Analysis of Transfer Function-based Binary Metaheuristic Algorithms for Feature Selection, in: 2018 International Conference on Artificial Intelligence and Data Processing (IDAP), 2018, pp. 1–6.

136.

Rodrigues

et al., A wrapper approach for feature selection based on bat algorithm and optimum-path forest, Expert Syst. Appl 41(5) (2014), 2250–2258.

137.

Deshpande

Doke

Deshpande

and Chaudhari

A.N.

, Expert system for retrieval of documents using evolutionary approaches incorporating clustering, in: 2017 International conference of Electronics, Communication and Aerospace Technology (ICECA), Vol. 2, 2017, pp. 414–418.

138.

Liu

Yan

and Lu

, Feature selection for image steganalysis using binary bat algorithm, IEEE Access, 2019.

139.

Kang

Kim

and Kim

J.-M.

, Reliable fault diagnosis for incipient low-speed bearings using fault feature analysis based on a binary bat algorithm, Inf. Sci. (Ny) 294 (Feb. 2015), 423–438. doi: 10.1016/J.INS.2014.10.014.

140.

Deng

and Duan

, Chaotic mutated bat algorithm optimized edge potential function for target matching, in: 2015 IEEE 10th Conference on Industrial Electronics and Applications (ICIEA), 2015, pp. 1049–1053.

141.

Raj

S.P.S.

Raja

N.S.M.

Madhumitha

M.R.

and Rajinikanth

, Examination of Digital Mammogram Using Otsu’s Function and Watershed Segmentation, in: 2018 Fourth International Conference on Biosignals, Images and Instrumentation (ICBSII), 2018, pp. 206–212.

142.

Priyadharshini

F.R.A.

Hariprasad

Asvitha

Anandhi

and Priyadarshini

A.P.S.

, An Approach to Segment Computed Tomography Images using Bat Algorithm, in: 2018 International Conference on Recent Trends in Electrical, Control and Communication (RTECC), 2018, pp. 6–9.

143.

Tharwat

V.S.A.

Zawbaa

H.M.

Gaber

and Hassanien

A.E.

, Automated zebrafish-based toxicity test using Bat optimization and AdaBoost classifier, in: 2015 11th International Computer Engineering Conference (ICENCO), 2015, pp. 169–174.

144.

Hamidzadeh

Sadeghi

and Namaei

, Weighted support vector data description based on chaotic bat algorithm, Appl. Soft Comput 60 (Nov. 2017), 540–551. doi: 10.1016/J.ASOC.2017.07.038.

145.

Al-Betar

M.A.

Awadallah

M.A.

Faris

Yang

X.-S.

Khader

A.T.

and Alomari

O.A.

, Bat-inspired algorithms with natural selection mechanisms for global optimization, Neurocomputing 273 (2018), 448–465.

146.

Satapathy

S.C.

Sri Madhava Raja

Rajinikanth

Ashour

A.S.

and Dey

, Multi-level image thresholding using otsu and chaotic bat algorithm, Neural Comput. Appl 29(12) (Jun. 2018), 1285–1307. doi: 10.1007/s00521-016-2645-5.

147.

Kotteeswaran

and Sivakumar

, A Novel Bat Algorithm Based Re-tuning of PI Controller of Coal Gasifier for Optimum Response, Springer, Cham, 2013, 506–517.

148.

Ghanem

W.A.H.M.

and Jantan

, An enhanced bat algorithm with mutation operator for numerical optimization problems, Neural Comput. Appl 31(S1) (Jan. 2019), 617–651. doi: 10.1007/s00521-017-3021-9.

149.

Meng

X.-B.

Gao

X.Z.

Liu

and Zhang

, A novel bat algorithm with habitat selection and doppler effect in echoes for optimization, Expert Syst. Appl 42(17–18) (2015), 6350–6364.

150.

Ibrahim

D.R.

Ghnemat

and Hudaib

, Software defect prediction using feature selection and random forest algorithm, in: 2017 International Conference on New Trends in Computing Sciences (ICTCS), 2017, pp. 252–257.

151.

Hafezi

Shahrabi

and Hadavandi

, A bat-neural network multi-agent system (BNNMAS) for stock price prediction: Case study of DAX stock price, Appl. Soft Comput 29 (Apr. 2015), 196–210. doi: 10.1016/J.ASOC.2014.12.028.

152.

Pravesjit

, A hybrid bat algorithm with natural-inspired algorithms for continuous optimization problem, Artif. Life Robot 21(1) (2016), 112–119.

153.

Bento

P.M.R.

Pombo

J.A.N.

Calado

M.R.A.

and Mariano

S.J.P.S.

, Optimization of neural network with wavelet transform and improved data selection using bat algorithm for short-term load forecasting, Neurocomputing 358 (Sep. 2019), 53–71. doi: 10.1016/J.NEUCOM.2019.05.030.

154.

Al-Betar

M.A.

Alomari

O.A.

and Abu-Romman

S.M.

, A TRIZ-inspired bat algorithm for gene selection in cancer classification, Genomics 112(1) (Jan. 2020), 114–126. doi: 10.1016/J.YGENO.2019.09.015.

155.

Yaseen

Z.M.

et al., A hybrid bat-swarm algorithm for optimizing dam and reservoir operation, Neural Comput. Appl 31(12) (Dec. 2019), 8807–8821. doi: 10.1007/s00521-018-3952-9.

156.

Paiva

F.A.P.

Silva

C.R.M.

Leite

I.V.O.

Marcone

M.H.F.

and Costa

J.A.F.

, Modified bat algorithm with cauchy mutation and elite opposition-based learning, in: 2017 IEEE Latin American Conference on Computational Intelligence (LA-CCI), Nov. 2017, pp. 1–6. doi: 10.1109/LA-CCI.2017.8285715.

157.

Salgotra

and Singh

, A novel bat flower pollination algorithm for synthesis of linear antenna arrays, Neural Comput. Appl 30(7) (Oct. 2018), 2269–2282. doi: 10.1007/s00521-016-2833-3.

158.

Kaur

Saini

B.S.

and Gupta

, A novel feature selection method for brain tumor MR image classification based on the fisher criterion and parameter-free bat optimization, Neural Comput. Appl 29(8) (Apr. 2018), 193–206. doi: 10.1007/s00521-017-2869-z.

159.

Arunarani

A.R.

and Manjula

, BABC task scheduler: Hybridisation of BAT and artificial bee colony for deadline constrained task scheduling, Int. J. Bus. Intell. Data Min 11(4) (2016), 379. doi: 10.1504/IJBIDM.2016.082216.

160.

Kiran

and Reddy

G.R.M.

, Bat-termite: A novel hybrid bio inspired routing protocol for mobile ad hoc networks, Int. J. Wirel. Mob. Comput 7(3) (2014), 258. doi: 10.1504/IJWMC.2014.062032.

161.

Shehab

Khader

A.T.

Laouchedi

and Alomari

O.A.

, Hybridizing cuckoo search algorithm with bat algorithm for global numerical optimization, J. Supercomput 75(5) (May 2019), 2395–2422. doi: 10.1007/s11227-018-2625-x.

162.

Saha

S.K.

Kar

Mandal

Ghoshal

S.P.

and Mukherjee

, A new design method using opposition-based BAT algorithm for IIR system identification problem, Int. J. Bio-Inspired Comput 5(2) (2013), 99–132.

163.

Alomari

O.A.

Tajudin

Mohammed

and Mohammed

A.A.

, A novel gene selection method using modified MRMR and hybrid bat-inspired algorithm with $

\beta

$-hill climbing, 2018.

164.

Chen

et al., Distributed Text Feature Selection Based On Bat Algorithm Optimization, in: 2019 10th IEEE International Conference on Intelligent Data Acquisition and Advanced Computing Systems: Technology and Applications (IDAACS), Vol. 1, 2019, pp. 75–80.

165.

and Le

, Hybrid Binary Bat Algorithm with Cross-Entropy Method for Feature Selection, in: 2019 4th International Conference on Control and Robotics Engineering (ICCRE), 2019, pp. 165–169.

166.

Parashar

Senthilnath

and Yang

X.S.

, A novel bat algorithm fuzzy classifier approach for classification problems, Int. J. Artif. Intell. Soft Comput 6(2) (2017), 108. doi: 10.1504/IJAISC.2017.084579.

167.

Tufail

Zafar

and Baig

, Digital watermarking for relational database security using mRMR based binary bat algorithm, in: 2018 17th IEEE International Conference On Trust, Security And Privacy In Computing And Communications/12th IEEE International Conference On Big Data Science And Engineering (TrustCom/BigDataSE), 2018, pp. 1948–1954.

168.

Kavitha

and Christopher

, Heart rate variability classification using sade-elm classifier with bat feature selection, ICTACT J. Soft Comput 7(4) (2017).

169.

Hou

Zhang

and Yang

, Application of bat algorithm for texture image classification, Int. J. Intell. Syst. Appl 10(5) (2018), 42–50.

170.

Imane

and Nadjet

, Bat algorithm for overlapping community detection, in: 2015 SAI Intelligent Systems Conference (IntelliSys), Nov. 2015, pp. 664–667. doi: 10.1109/IntelliSys.2015.7361211.

171.

Cheng

and Bao

, A Kernelized Fuzzy C-means Clustering Algorithm based on Bat Algorithm, in: Proceedings of the 2018 10th International Conference on Computer and Automation Engineering – ICCAE 2018, 2018, pp. 1–5. doi: 10.1145/3192975.3193009.

172.

Qiao

Kewen

Panpan

and Wang

, Lung nodule classification using curvelet transform, LDA algorithm and BAT-SVM algorithm, Pattern Recognit. Image Anal 27(4) (Oct. 2017), 855–862. doi: 10.1134/S1054661817040228.

173.

Khennak

and Drias

, Bat-inspired algorithm based query expansion for medical web information retrieval, J. Med. Syst 41(2) (Feb. 2017), 34. doi: 10.1007/s10916-016-0668-1.

174.

Srivastava

P.R.

Bidwai

Khan

Rathore

Sharma

and Yang

X.S.

, An empirical study of test effort estimation based on bat algorithm, Int. J. Bio-Inspired Comput 6(1) (2014), 57. doi: 10.1504/IJBIC.2014.059966.

175.

Oshaba

A.S.

Ali

E.S.

and Abd Elazim

S.M.

, PI controller design for MPPT of photovoltaic system supplying SRM via BAT search algorithm, Neural Comput. Appl 28(4) (Apr. 2017), 651–667. doi: 10.1007/s00521-015-2091-9.

176.

Sambariya

D.K.

and Prasad

, Design of optimal proportional integral derivative based power system stabilizer using bat algorithm, Appl. Comput. Intell. Soft Comput 2016 (2016), 1–22. doi: 10.1155/2016/8546108.

177.

Das

Mandal

Ghoshal

S.P.

and Kar

, An efficient side lobe reduction technique considering mutual coupling effect in linear array antenna using BAT algorithm, Swarm Evol. Comput 35 (Aug. 2017), 26–40. doi: 10.1016/J.SWEVO.2017.02.004.

178.

Jalal

Mukhopadhyay

A.K.

and Goharzay

, Bat algorithm as a metaheuristic optimization approach in materials and design: Optimal design of a new float for different materials, Neural Comput. Appl 31(10) (Oct. 2019), 6151–6161. doi: 10.1007/s00521-018-3430-4.

179.

Gandomi

A.H.

Yang

X.-S.

Alavi

A.H.

and Talatahari

, Bat algorithm for constrained optimization tasks, Neural Comput. Appl 22(6) (May 2013), 1239–1255. doi: 10.1007/s00521-012-1028-9.

180.

Sahlol

A.T.

Suen

C.Y.

Zawbaa

H.M.

Hassanien

A.E.

and Elfattah

M.A.

, Bio-inspired BAT optimization algorithm for handwritten Arabic characters recognition, in: 2016 IEEE Congress on Evolutionary Computation (CEC), 2016, pp. 1749–1756.

181.

Mohamed

et al., The Effectiveness of Bat Algorithm for Data Handling in Various Applications, no. November, 2016, 25–27.

182.

Wang

Chen

and Zhao

, Texture image classification based on support vector machine and bat algorithm, in: 2015 IEEE 8th International Conference on Intelligent Data Acquisition and Advanced Computing Systems: Technology and Applications (IDAACS), Vol. 1, 2015, pp. 309–314.

183.

Sathananthavathi

and Indumathi

, BAT algorithm inspired retinal blood vessel segmentation, IET Image Process 12(11) (2018), 2075–2083.

184.

Aasha

S.S.M.

, Multi-objective effective enhanced adaptive fusion technique using BAT algorithm for effective gait-based recognition, in: 2017 4th International Conference on Advanced Computing and Communication Systems (ICACCS), 2017, pp. 1–6.

185.

Tang

Fong

and Deb

, Integrating Nature-inspired Optimization Algorithms to K-means Clustering, in: Digital Information Management (ICDIM), 2012 Seventh International Conference on, 2012, pp. 116–123.

186.

Agarwal

and Mehta

, Comparative analysis of nature inspired algorithms on data clustering, in: Research in Computational Intelligence and Communication Networks (ICRCICN), 2015 IEEE International Conference on, 2015, pp. 119–124.

187.

Reddy

M.P.

Mukherjee

and Ganguli

, Optimal design of damage tolerant composite using ply angle dispersion and enhanced bat algorithm, Neural Comput. Appl., Sep. 2019, 1–20. doi: 10.1007/s00521-019-04455-8.

188.

Wang

G.-G.

Chu

H.E.

and Mirjalili

, Three-dimensional path planning for UCAV using an improved bat algorithm, Aerosp. Sci. Technol 49 (Feb. 2016), 231–238. doi: 10.1016/J.AST.2015.11.040.

189.

Chaudhary

and Banati

, Swarm bat algorithm with improved search (SBAIS), Soft Comput., 2018, 1–31.

190.

Aleti

and Moser

, Studying feedback mechanisms for adaptive parameter control in evolutionary algorithms, in: 2013 IEEE Congress on Evolutionary Computation, 2013, pp. 3117–3124.

191.

Zhang

et al., A survey on algorithm adaptation in evolutionary computation, Front. Electr. Electron. Eng 7(1) (2012), 16–31.

192.

Parpinelli

R.S.

Plichoski

G.F.

Da Silva

R.S.

and Narloch

P.H.

, A review of techniques for online control of parameters in swarm intelligence and evolutionary computation algorithms, Int. J. Bio-Inspired Comput 13(1) (2019), 1–20.

193.

Trivedi

I.N.

Pradeep

Narottam

Arvind

and Dilip

, Novel adaptive whale optimization algorithm for global optimization, Indian J. Sci. Technol 9(38) (2016), 319–326.

194.

Parpinelli

R.S.

and Narloch

P.H.

, A review of techniques for online control of parameters in swarm intelligence and evolutionary computation algorithms A Review of Techniques for On-line Control of Parameters in Swarm Intelligence and Evolutionary Computation Algorithms Rafael Stubs Parpi, no. January, 2019. doi: 10.1504/IJBIC.2019.097731.

Improvements of bat algorithm for optimal feature selection: A systematic literature review

Abstract

Keywords

1. Introduction

Table 1 Focus and limitations of existing surveys

Table 3 Scientific databases and results for literature search

2.3.1 Exclusion criteria

2.3.2 Inclusion criteria

3.2 Advantages and disadvantages of bat algorithm

Table 4 Advantages and disadvantages of bat algorithm

Table 5 Definitions and notations of summary’s concepts

3.3.1.1 Continuous encoding scheme

3.3.1.2 Binary encoding scheme

3.3.1.3 Discrete encoding scheme

3.3.1.4 Mix encoding scheme

Table 7 Techniques used for improving components of bat algorithm

Table 8 Recent applications of bat for optimal feature selection

Table 9 Dataset size used in evaluating BA performance for feature selection

7. Conclusion

Footnotes

Acknowledgments

References

Table 1
Focus and limitations of existing surveys

Table 3
Scientific databases and results for literature search

Table 4
Advantages and disadvantages of bat algorithm

Table 5
Definitions and notations of summary’s concepts

Table 7
Techniques used for improving components of bat algorithm

Table 8
Recent applications of bat for optimal feature selection

Table 9
Dataset size used in evaluating BA performance for feature selection