Received 20 December 2023, accepted 7 January 2024, date of publication 24 January 2024, date of current version 12 February 2024.

Digital Object Identifier 10.1109/ACCESS.2024.3357993

FNN for Diabetic Prediction Using Oppositional

Whale Optimization Algorithm
RAJESH CHATTERJEE 1 , MOHAMMAD AMIR KHUSRU AKHTAR 1 ,
DINESH KUMAR PRADHAN 2 , FALGUNI CHAKRABORTY 2 , MOHIT KUMAR 3,

SAHIL VERMA 4 , (Senior Member, IEEE), RUBA ABU KHURMA5,6 ,

AND MARIBEL GARCÍA-ARENAS 7,8
1 Faculty of Computing & IT, Usha Martin University, Ranchi 835103, India
2 Dr.B. C. Roy Engineering College, Durgapur 713206, India
3 Department of IT, MIT Art, Design and Technology University, Pune 412201, India
4 Department of Computer Science and Engineering, Chandigarh Group of Colleges, Jhanjeri, Mohali, Punjab 140307, India
5 Faculty of Information Technology, MEU Research Unit, Middle East University, Amman 11831, Jordan
6 Applied Science Research Center, Applied Science Private University, Amman 11931, Jordan
7 Department of Computer Engineering, Automatics and Robotics, University of Granada, 18071 Granada, Spain
8 Centro de Investigación en Tecnologías de la Información y las Comunicaciones de la Universidad de Granada (CITIC-UGR), 18071 Granada, Spain

Corresponding author: Maribel García-Arenas ([email protected])

This work was supported in part by the Ministerio Español de Ciencia e Innovación under Grant PID2020-115570GB-C22,
Grant PID2022-137461NB-C31, and Grant MCIN/AEI/10.13039/501100011033; in part by Programa Operativo FEDER 2021-2027
under Grant C-ING-027-UGR23; and in part by the Cátedra Fujitsu Tecnología para las Personas (UGR-Fujitsu).

ABSTRACT The medical field is witnessing rapid adoption of artificial intelligence (AI) and machine
learning (ML), revolutionizing disease diagnosis and treatment management. Researchers explore how
AI and ML can optimize medical decision-making, promising to transform healthcare. Feed Forward
Neural Networks (FNN) are widely used to create predictive disease models, cross-validated by medical
experts. However, complex medical data like diabetes leads to multi-modal search spaces prone to local
minima, affecting optimal solutions. In this study, we focus on optimizing a diabetes dataset from the
Pima Indian community, evaluating decision-making performance in diabetes management. Employing
multimodal datasets, we compare various optimization algorithms, including the Whale Optimization
Algorithm (WOA) and Particle Swarm Optimization (PSO). The test results encompass essential metrics like
best-fit value, mean, median, and standard deviation to assess the impact of different optimization techniques.
The findings highlight the superiority of the Oppositional Whale Optimization Algorithm (OWOA) over
other methods employed in our research setup. This study demonstrates the immense potential of AI and
metaheuristic algorithms to revolutionize medical diagnosis and treatment approaches, paving the way
for future advancements in the healthcare landscape. Results reveal the superiority of OWOA over other
methods. AI and metaheuristics show tremendous potential in transforming medical diagnosis and treatment,
driving future healthcare advancements.

INDEX TERMS Feed forward neural network (FNN), oppositional learning, artificial intelligence,
meta-heuristic algorithms, whale optimization algorithm (WOA).

I. INTRODUCTION helps increase neural network’s power and use cases. Out
Machine learning is a branch of science that learns from data of many different types of neural networks available, feed
and provides insight. One discipline is Artificial Neural Net- forward is one of the popular oldest neural networks used
works (ANN), inspired by neurons in the human brain. The today.
advent of humongous data and excellent computing power In the feed forward neural network, the connection from
the input to the hidden to-output layer is one-directional.
The associate editor coordinating the review of this manuscript and There might be multiple input nodes, each with a specific
approving it for publication was Yu-Huei Cheng . weight associated and multiplied individually to sum up the
2024 The Authors. This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License.
20396 For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/ VOLUME 12, 2024
R. Chatterjee et al.: FNN for Diabetic Prediction Using Oppositional Whale Optimization Algorithm

value. Weight is used to assign a certain importance level to model mentioned above, it is still stuck in local optima in
each neuron. These are then added with a bias value. The finding the global minima. This study has incorporated oppo-
overall summed value is passed to the activation function sitional learning with the WOA to overcome the possibility of
to obtain the results. Many highly used activation functions getting stuck into local optima and decide the best weight and
include sigmoid, linear function, tanh, softmax, ReLU and bias set, which gives the global optima.
leaky ReLU.
Therefore, each neuron is connected with other neurons
and helps each other to come up with a given output. The A. RESEARCH REVIEW
activation value helps neuron to decide if it needs to get fire Metaheuristic algorithms are popular optimization tech-
or not to achieve a given output. niques that show encouraging results to solve complex prob-
Figure 1, an FNN [1], [2] shows standard input, hidden lems that are highly non-linear, have many optima and take
and output nodes architecture. These nodes are analogues to time to converge. Some notable optimization techniques are
neurons of Homo sapiens mind. The provided input is the GA (Genetic Algorithm) [5], PSO [6], [7], BAT [8], DE (Dif-
constituent of an input layer, and it captures all the features ferential Evolution) [9], ACO [10], CS (Cuckoo Search) [11],
available as part of the input. We process the input values ABC (Artificial Bee Colony) [12], and WOA (Whale Opti-
and go to the hidden layer, which captures the hidden pattern mization Algorithm) [13]. These optimization techniques are
with the non-linearity of data. The output layer is the one effective during various studies to find optimal control param-
that provides the desired result(s) that we want to predict or eters. Thereby, it provides faster convergence and avoids
classify. The most important aspect of an FNN understands local optima issues, assisting in solving real-world use cases.
the non-linearity, insight available with the data and providing Like the population-based algorithm, gradient-based back
parallelism. There is an increasing number of real-life appli- propagation (BP) is popular for training feed forward neural
cations coming up, like classifying pictures [3], analyzing networks and other network architectures. The essence of
texts [4], and forecasting with time-series analysis. back-propagation is to propagate the error backwards. It first
The potential of a feed forward network arrives with the captures the error or loss incurred in one iteration, and the
ability to learn hidden patterns with the data when an FNN is error is propagated back. This propagation of error helps
trained well. A well-trained model can solve a given regres- fine-tune control parameters, i.e. weight and bias. This helps
sion or classification problem with optimal fitness value. The in more generalized learning of the FNN model and thereby
training depends on a good model architecture, the selection helps find the global optima. The process keeps repeating
of hyper parameters, and the choice of the proper optimiza- until a satisfied error value is met.
tion algorithm. In the study [14], the author has used the back propagation
The current study focuses on the optimization pillar of algorithm combined with differential evolution to take mul-
these three pillars and understands the impact on the model tiple parameters like the gross domestic product of gas, total
with a better optimization algorithm. The algorithm aims to population, total import etc., as the input and rightly predict
converge faster and provide global minima that help in faster the energy consumption. The back-propagation is found to
model training and prediction with less error rate. have a considerable impact on predicting accurately. The BP
In the literature, several techniques have been proposed works by internally handling the weight and slowly improv-
to optimize the model. A few notable are gradient descent ing to reach closer to predicting correct value. BP is used in
and meta-heuristic algorithms. When a model starts learning, gradient descent that mathematically reduces the fitness value
the issue to address is the scenarios when the model is stuck (error rate) and helps with learning the model better.
in local minima and takes longer to converge. There comes The author Hosseinioun et al. [15], [16] has developed
the importance of optimization algorithms. It helps find the dynamic voltage and frequency scaling (DVFS) technique
optimal control parameters that help in reducing the error rate which is an energy aware method to reduce the usage of
with faster convergence and find the global minima. energy. This is done applying meta-heuristic technique by
Multi-modal search spaces are the kind of issues which mix of Invasive Weed optimization and Culture (IWO-CA).
do not have one optimum but have many optima which may This helps in improved usage of energy.
be a mix of local optima or global optima. In such search Malik and Kim [17] has shown the usage of Particle swarm
spaces, at times the modal stuck in localized minimal value optimization with regeneration based PSO neural network
and cannot reach global minimal value. In many studies, and velocity boost-based PSO NN to manage and minimize
optimization techniques are put forward to address aforemen- the energy consumption. Neural network combined with PSO
tioned problem. The variables that dictate faster convergence is found to be effective to get a better accuracy in predicting.
are optimal metric of weight and bias. In the study from Li et al. [18] has done the research to
The WOA is a nature-inspired population-based meta- see the relation between crack depth and surface acoustic
heuristic algorithm based on noteworthy trapping workings wave (SAW). The study uses features generated using laser
called the bubble-net feeding procedure of humpback whales. source scan to the neural network and uses particle swarm
In recent years, WOA has been applied in different engineer- optimization as the optimization algorithm. The outcome is
ing fields. Although the WOA is best suited for training the very positive, and the model created using neural network

VOLUME 12, 2024 20397

 R. Chatterjee et al.: FNN for Diabetic Prediction Using Oppositional Whale Optimization Algorithm

with PSO is able to quickly evaluate the crack depths and able world. The hybrid model is able to work well in getting the
correctly use the Scanning Laser Source (SLS) data. Xu et al. performance accuracy and is able to converge fast.
[19] did a great work on flood management with accuracy on From the outcome of the simulated results, it is observed
forecasting rain beforehand. that proposed approach is more effective than the conven-
An approach using LSMT network and PSO helped learn- tional firefly algorithm.
ing time series data and be better prepared to predict rainfall The Table 1 shows a comparation between different opti-
thereby doing better flood management. mization techniques using various factors like convergence
Ilbeigi et al. [20] have proposed method using genetic rate, avoidance of local optima, dependence on input control
algorithm in neural network. In the proposal, the author(s) parameters, time complexity etc.
worked to come up with a workable method to optimize the Mirjalili et al. [13] developed WOA, an efficient optimizer
energy consumption of building. First important features are method. It is based on the preying traits of whales. This
taken for the purpose numerically and this is then put to artifi- follows the phase of exploration and exploitation. Some of
cial neural network which is trained and tested to see how its the notable points are as follows:
working with varied iterations. Then energy optimization is • WOA follows a helix path for hunting the victim.
done using genetic algorithm using those critical variables. It • This is a distinct feature of the humpback whale that it
shows a significant improvement and a significant reduction follows during the exploration phase.
in energy consumption. • The next phase is to exploit the target with a randomized
In the research from Wang et al. [21], the author used whale value that helps catch the prey in the given phase.
optimization technique with artificial neural network to get • Overall, WOA does a better job of finding global optima
the best weight and bias for image segregation. Similarly, and avoiding local optima.
Li et al [22] showed a whale optimization technique (WOA) WOA is increasingly becoming a very popular optimization
which is a modified one which helped in faster global con- algorithm, and many use cases are coming up in various
vergence. Kaladevi et al. [23] is able to identify cancer genus studies.
cells using meta-heuristic optimization technique. For a more accurate measure of the renewable power
Musheer et al. [24] employed uses ABC for selecting resources and the amount of the loss of power, Prakash et al.
best features for microarray data. In this study, author has [31] have implemented WOA. Similarly, Haghnegahdar and
conducted statistical hypothesis test on six cancer datasets Wang [32] have used neural networks with WOA as an opti-
to compare the efficiency of the proposed technique to other mization algorithm to detect and prevent cyber-attacks for
algorithms. Aziz et al. [25] proposed a new feature selection Smart grids, which are intelligent power systems. The author
approach for better classification result in ANNs. Indepen- of the paper [33] has used WOA in an ANN model to predict
dent component analysis (ICA) is used as an extraction fly rock in 5 different categories, which provides a better
method and ABC as an optimizer. Aziz et al. [26] improved result.
the ICA feature selection using genetic bee colony algorithm. The success of meta-heuristic algorithm with oppositional-
Zhang et al. [27] show the approach for selecting fea- based learning have inspired to provide a novel optimization
tures of a problem. Feature selection is one of the most algorithm known as OWOA to train FFNN to predict diabetic
important problems in optimization. The binary differential possibility in a woman based on specific parameters.
evolution is based on binary operator to focus on feasible The proposed OWOA method is then applied to a neural
optimal spaces and removing the less effective features to network with a standard database diabetic problem. The per-
help in less computational complexity. The outcome with formance is evaluated with other optimization methods like
different dataset shows that it is helping in avoiding local WOA, ABC and PSO. The primary focus of measurement
optima and is able to do global exploration. Similarly, Gao is how effectively these algorithms can avoid local optimum
et al. [28] show a novel approach using directional permu- and converge faster. It is good to note that the OWOA opti-
tational differential evolution algorithm to address photo- mization algorithm uses the advantage of combining WOA
voltaic generation model. The PV generation is a complex along with OBL. Some well-defined diabetes datasets are
process as well as very challenging due to its non-convex, considered to evaluate the capability of OWOA to predict
non-linear property with multiple parameters involved. The diabetes, and outcomes are compared with other standard
proposed algorithm is evaluated with six experimental groups optimization models. The improved accuracy and the fitness
on singular, double and triple diode models along with PV value we can get using OWOA show that it does better than
models. The PV models are showing a much better result many other meta-heuristic algorithms for diabetic prediction.
and are able to outperforms the performance accuracy of Using oppositional base learning with the whale optimiza-
others. tion algorithm increases the capability of the exploration
Lenin et al. [29] used DFO with improved squirrel search and exploitation phase to enhance the convergence ability
optimization (ESSO) algorithm to optimize power losses. In of WOA, which in turn helps in finding the Global Optima
the study [30] a combined WOA and sine cosine algorithm more effectively. The efficacy of OWOA has not yet been
to identify the features of the cellular topology in the chaotic established for diabetic problems.

20398 VOLUME 12, 2024

R. Chatterjee et al.: FNN for Diabetic Prediction Using Oppositional Whale Optimization Algorithm

TABLE 1. Comparative study of popular meta-heuristic optimization •To integrate oppositional based learning with WOA
algorithms.
algorithm to create larger search space and convergence
avoidance.
• The performances of OWOA along with other
nature-inspired algorithms are evaluated for faster con-
vergence and time to converge.
• The output is compared and analyzed with other opti-
mization algorithms like CSSO, WOA, and SSO.
• The Diabetes test scenarios are used for the diabetic
prediction to carry out.
• The algorithm is evaluated with statistical methods.
• Different metrics are compared to evaluate fitness of the
current proposed algorithm.
The rest of the paper is divided into different parts. The
mathematical formulation is explained in Section II. Dif-
ferent metaheuristic algorithms are shown in Section III.
Section IV shows different oppositional based learning. In
section V, we proposed OWOA algorithm. In Section VI, the
datasets are described. The experimental setup is discussed
in Section 7. Section 8 provides the experimental results
obtained from different datasets. Finally, Section 9 discusses
the conclusion of the proposed study.

II. PROBLEM FORMULATION

In Figure 1, here xn = x1 , x2 , x3 ,. . . are set of inputs, wn =
w1 , w2 , w3 , . . . is weight. The output can be shown as the
cumulative value of weight and bias as:
Xm
Ol = xm wmn + bn where n = 1, 2, 3, . . . (1)
j=1
This is the value for a given neuron. Such a set of transferred
functions we use to get our desired output which is called
activation function. Let us show the activation function as:
O = f (At ) (2)
where, f = method to activate neurons. In our study, sigmoid
function is used for that purpose. The sigmoid method can be
shown as:
f (x) = 1/(1 + e−t ) (3)
B. NOVELTY AND CONTRIBUTION
The primary objective for OWOA is to find optimal set of This is a special case of single layered network. Likewise,
hyper parameters to help with an optimized neural network. we can extend the concept to multi-layered network, assume.
By hyper-parameters, we are mainly addressing weights Input nodes are (I1 , I2 , . . . , Im ), H = H1 , H2 , . . . , Hk hidden-
and biases that we can do for WOA, PSO, stochastic gra- nodes and X = X1 , X2 , . . . , Xs are output-nodes as described
dient descent (SGD), and Adam optimization algorithm in the Figure 1, the activation method of a given hidden node
(AOA) optimization techniques that are used to optimize say nth hidden node can be shown as:
Xm 
control parameters (weights and biases). The proposed In = f Iv rvj + bj where j = 1, 2, 3, . . . k (4)
OWOA has shown a better performance in the diabetes v=1
dataset. The output of the node t th can be shown as:
To showcase the performance of OWOA, the algorithm X
k

will be assessed with finding global optima, fitness value and Ot = f pn ant + et where t = 1, 2, 3, . . . r (5)
n=1
some statistical metrics.
Here are the primary contributions of the research: where rvj is the weight parameter between r th input node and
• To propose a novel optimization technique called Whale jth hidden node; ant denotes the weight between nth hidden
Optimization algorithm to identify the optimize weight node and t th output node; bj represents the bias of jth hidden
and bias control parameters. node and et is the bias of t th output node; m is the number

VOLUME 12, 2024 20399

 R. Chatterjee et al.: FNN for Diabetic Prediction Using Oppositional Whale Optimization Algorithm

FIGURE 1. Architecture of a) Single layer b) Multilayer FNN.

of input nodes; k is the number of hidden nodes and r is the in smart home, accuracy improvement for cancer diagnosis
number of output nodes. [37], etc.
The fitness value be obtained as: The steps to follow for ABC are:
Xn  2 1. To initialize the number of scout bees looking for food
F= Oli − Sil (6) origin.
i=1
2. To define employed bees that searches new food origin
here n is the samples size and Sil refers to ith sample for l th within the neighbor.
output. 3. To define on looker bees to evaluate the quality of food
as per the search done by source bees by the employed
III. OPTIMIZATION TECHNIQUES TO TRAIN AN FNN bees.
The feed forward networks have a very close relationship 4. To scout bees to randomly initiate new solutions.
with the optimization techniques used in computational sci- 5. To keep re-iterating steps 2, 3, and 4 to achieve best
ence. Optimization works to find the control fields that result solution.
in minimized fitness value. The FNN model is trained to
reduce the error rate in each iteration. There are multiple mod- B. PARTICLE SWARM OPTIMIZATION
els used to optimize the fitness function. There are numerous Particle swarm optimization (PSO) is a nature inspired
optimization techniques [34] used in literature. algorithm to achieve global. Wang et al. [6] proposed the
Gradient descent uses the differential function to find popular algorithm. This algorithm is developed to mimic the
the global minima and minimizes the function, whereas the intelligent behavior demonstrated by bird and fishes. Like
meta-heuristic algorithm uses a population set to find the any other population-based search algorithm, it uses a set of
global minima. The control variables assist the model in population to search for the best result. The search consists
learning the underlying model well. The model minimizes of individual result as well it compares with search from
control parameters using a better choice of network architec- the around candidate. These search agents keep the informa-
ture, a better optimization technique and proper tuning with tion about velocity and position which is updated in every
hyper parameters. Many optimization techniques are pro- iteration. After a number of iterations, it is able to get the
posed in the literature to get minimum control parameters for best location and position which is considered as the global
the best fitness function. The population-based meta-heuristic optimum value.
optimization techniques show promising results with faster This can be shown mathematically as:
convergence and finding global minima. The current litera-
fns+1 = lfns + d1 .rand 1 . qbest sn − xns


ture will discuss the population-based optimization algorithm
+ d2 .rand 2 . qbest − xns


to train the feed forward neural network. (7)
xns+1 = xns + us+1
n (8)
A. ARTIFICIAL BEE COLONY OPTIMIZATION
Artificial Bee Colony optimization also known as ABC [35], where fns is the velocity of nth particle at iteration s; l is
[36] is a popular meta heuristic algorithm used to solve vari- the weight; d1 and d2 are acceleration coefficient; rand 1 and
ous scientific and real world problems. The algorithm is based rand 2 are the random numbers between0 and 1; xns is current
on the behavior of bee swarms to intelligently find their food position of agent n at iteration s, qbest is the nth agent at
source. The algorithm has three parts as the natural behavior iteration t and gbest is the overall best solution so far.
of bees: find all possible location of available food sources
which are scout bees, exploiting the sources coming from first C. WHALE OPTIMIZATION ALGORITHM
set which are employed bees and then evaluate quality of food Mirjalilli et al. [13], [21], proposed a new optimization
which are done by onlooker bees. algorithm called WOA. This is based on the hunting behav-
The application of ABC is seen in multiple areas of health ior of humpback whale. The hunting behavior is known
care, telecommunication, infrastructure like monitoring load as bubble-net hunting algorithm. The given whale travels

20400 VOLUME 12, 2024

R. Chatterjee et al.: FNN for Diabetic Prediction Using Oppositional Whale Optimization Algorithm

a certain distance original position to new position with heli- ‘Max_Iter’, the helix shape is constructed based on two cru-
cal path by encircling them. cial values: the current search value ‘X’ and the best-known
The search agent (individual agents) assumes current posi- value ’X^∗’. These values play a pivotal role in determining
tion as the optimal solution and shares the current location the trajectory and overall shape of the path, as described in
and based on that updates the optimal solution information. equation (14).
This helps the hump-back whales to reach closer to the prey −
→
−
→
at each iteration by updating information of the optimal loca- X (i + 1) = E ′ .pkr . cos (2πl) X ∗ (i) (14)
tion. This process continues until the whale hunts its trap.
This is a nature inspired algorithm aims at obtaining the The distance the whale needs to travel to reach its prey is
global minima with a good number of agents. We initialize represented by the vector E’ = (X∗)(i)-X(i). To ensure the
the agents with a randomly defined control parameters which desired shape of the path, a constant ‘k’ is utilized, which
are used as the current best position for each of the whale. As controls the exponential value. Additionally, the variable ‘r’
we keep iterating, we come close to the optimal solution for takes random values from −1 to 1.
all the agents. This is an iterative process that continues until Considering an equal probability for choosing between
we do not reach close to satisfactory solution. The approach the processes of reducing encirclement and creating a spiral-
makes WOA a ‘goto’ algorithm. shaped path, the next value can be computed as follows:
In nature, humpback whales hunts fishes using the same (
method that inspired WOA algorithm. It uses the given −
→ Reducing Spiral path, if t = .5
X (i + 1) = (15)
method to search for global minima and thereby avoids stuck helical path, if ̸= .5
forever into a local minimum.
The process of hunting a prey is a two-step process, first The algorithm comprises two main components: the first part
is to exploit and then to explore. The exploitation step uses of the equation represents the bubble-net technique, while the
reduced spiral path to reduce the distance and exploration second part defines the helical path.
continues to find the prey. In the WOA algorithm, similar to other population-based
The exploitation phase starts when a whale sees a victim methods, a balance between exploitation and exploration is
and makes spiral path to reach out to the victim. The mathe- achieved through the appropriate adjustment of the values ‘e’
matical expression can be shown as: and ‘s’.
Extensive literature demonstrates that WOA is effective in
−
→ →−
− →
D = | C .X ∗ (i) − X (i)| (9) solving diverse optimization problems. However, one persist-
−→∗ −
→− → ing challenge is the potential trapping in local minima.
X (i + 1) = X (i) − B . E (10) Stepwise summary involved in WOA algorithm is as men-
i Represents the latest iteration, X∗ Denotes the optimal tioned below:
value, X refers to the latest value, | | represents the absolute 1. To initialize WOA input parameters.
value symbol, (\ dot) denotes element-wise multiplication, 2. To initialize search agent values randomly.
Y and R constants used in the process, B and C variables that 3. To find fitness value of each search agent.
determine the shape of the curve in the helical path followed 4. The search agents are sorted in descending order and
by whales in each iteration to reach the target. B & C are two mark best solution as the ‘‘pbest’’.
coefficients that are computed using following equation 5. Each iteration, search agents are updated with,
i.e., searching and encircling the victim. This done by
B α r⃗ − ⃗e
⃗ = 2⃗ (11) updating the current position and happens in spherical
⃗ = 2⃗r
C (12) shape.
6. The fitness value of a given search is compared with the
The value of e is determined by: previous values and we update if current value is best.
7. The step 4 to 7 is in continuation till exit criteria are not
α = 2 − n ∗ (2/Max_iter) (13)
met. We have taken maximum iteration value as 100 for
During each iteration, the parameter α takes values from the current study.
set {0, 2} and decreases gradually. Additionally, r is a random
vector uniformly distributed in the range [0, 1]. The values D. OPPOSICIONAL BASED LEARNING
assigned to B and C directly impact the shape of the curve, In evolutionary algorithms, the initial population is randomly
leading the whales on their helical path towards the target. generated within the search space, and the aim is to increase
Notably, as the value of B decreases during each iteration, the population’s values iteratively to reach better solutions at
the curve continuously tightens and descends further as a each step. The search process continues until a predefined
result, the algorithm achieves a reduced encirclement in each benchmark is achieved. In cases where prior information is
iteration. not available, an initial solution is used, leading to random
In the context of the current run, denoted by ‘n’, and predicted value that creates a distance between these pre-
with the knowledge of the maximum iteration, represented by dicted values and the best solution. This distance plays a role

VOLUME 12, 2024 20401

 R. Chatterjee et al.: FNN for Diabetic Prediction Using Oppositional Whale Optimization Algorithm

in determining the computational time required for execut- Algorithm 1 Opposition Value Initialization
ing the search algorithm. When the initial predicted value FOR J = 1 to nx :
is closer to the optimal solution, the algorithm converges FOR l = 1 to ny :
quickly. If not, it takes more time to converge. OPj,l = uj + vl − Xj,l
Madhavi et al. [38] introduced a highly efficient compu- END FORLOOP
tational intelligence concept to increase optimization algo- END FORLOOP
rithms, known as oppositional-based learning (OBL). OBL Here, nx = population of the sample, and ny = total n◦ of
has been incorporated into various evolutionary algorithms control variables
for better convergence and enables finding the global optima
for optimization problems. The fundamental idea behind TABLE 2. Training data to train the FNN with diabetes dataset.
OBL is to utilize the current value and compare it with its
opposite value. This comparison allows the rapid identifica-
tion of better solutions. By evaluating the opposite values at
the same time, the algorithm has a better chance of converging
to positions closer to global solutions quickly. Therefore,
OBL is a powerful way to improve the final solution in
many optimization tasks. The definition of OBL is mentioned
below.
with each iteration and randomly jump to a new solution value
1) OPPOSITE NUMBER that may be a better solution than the present one.
Opposite number x , of a real number x is mathematically But using randomized value, here we use jumping rate Yt as
defined as: described in TABLE 2 for opposite population initialization:

x , = x + y − v where v ∈ [x, y] (16)

Algorithm 2 Jumping Based Initialization
2) OPPOSITE POINT IF randomVal = randomly selected value between 0 and 1
Let X = x1 + x2 + . . . + xn be a point in n dimensional space, THEN
where xk ∈ X and rk ∈ [ak , bk ], opposite number can be FOR J = 1 to nx :
evaluated as: FOR l = 1 to ny :
OPj,l = uj + vl − Xj,l
xk, = ak + bk − xk (17) END FORLOOP
END FORLOOP
Using aforementioned opposite point concept, oppositional
ENDIF
based learning (OBL) concept is described.
Here, nx = population of the sample, and ny = total n◦ of
control variables.
3) OPPOSITION BASED OPTIMIZATION
Let X = x1 , x2 , . . . , xn be possible solution space in n-dim
space with g(X ) with the function to evaluate fitness and X ‘
is the opposite value of X whose function to evaluate fitness is IV. PROPOSED OPPOSITIONAL LEARNING BASED WOA
g(X ‘ ). As per maximization theorem, if g (X)≤ g (X ‘ ), then FOR FNN
X can be replaced by X ‘ . The combined data points which The accuracy and convergence to diagnose a diabetic person
are existing space and opposite space can be used to find the using the Pima Indian data set shows a better result when
better fitness function. OBL is introduced with the WOA. The model presents the
Therefore, oppositional based learning has primarily two opposite learning space in the whale optimization algorithm
important parts (a) how to initialize the opposite space and and the current learning to improve the model accuracy. The
(b) at each iteration, create opposite points by generation algorithm follows the defined steps herewith:
jumps. Here is how we can describe it. • Step 1: A given number of search agents are chosen
using the OBL concepts.
a: OPPOSITION BASED INITIALIZATION OF OPPOSITE SPACE • Step 2: To choose an Artificial neural network with a
The initialization is done using algorithm shown in set of input nodes, one hidden layer and one output
Algorithm 1. layer. The network architecture that we use is dependent
on input and output values. The hidden layers should
b: OPPOSITION BASED GENERATION BY JUMPING be chosen wisely to avoid over fitting or under-fitting
PROBABILITY problems. The control parameters are weight and bias,
In the case of jumping probability, the process is similar to which depend on the number of inputs and hidden and
earlier one, here to make the learning more optimized we output nodes we choose. The weights and biases (control
forcibly changes the present population to the opposite one parameters) are randomly initialized.

20402 VOLUME 12, 2024

R. Chatterjee et al.: FNN for Diabetic Prediction Using Oppositional Whale Optimization Algorithm

TABLE 3. Least fitness metric of the proposed algorithms for diabetes TABLE 3. (Continued.) Least fitness metric of the proposed algorithms for
dataset. diabetes dataset.

• Step 4: Fitness values are calculated using Eq.6. The

• Step 3: The upper, lower and maximum iteration values lowest fitness value is chosen in each iteration to min-
are initialized. imize the overall error.

VOLUME 12, 2024 20403

 R. Chatterjee et al.: FNN for Diabetic Prediction Using Oppositional Whale Optimization Algorithm

TABLE 4. Efficiency evaluation of OWOA, ABC, WOA and PSO for diabetic classification.

• Step 5: The WOA is used for the exploration and Y = number of hidden nodes. In current setup, Y = 7. The
exploitation phase. The humpback whale follows a data is used to provide the diagnosis of diabetic occurrence
reducing helical path to reach the prey each time. using this dataset. The dataset has 768 total data, 500 are
• Step 6: The best value obtained for each iteration is non-diabetic, 268 diabetic cases. The dataset has a total of
updated as the best solution available globally. With the 8 fields and one output field as diabetic prediction. The
best solution, the distance is calculated to see how close research is performed on ladies to evaluate the impact of the
we are to the prey. given input parameters on diabetes whether it’s impacting
• Step 7: Based on the jumping value on every iteration, negatively or positively. The output class is divided into 0 and
candidate solution and its opposite points values are cal- 1 respectively mean non-diabetic, diabetic. All the optimiza-
culated and kept as the candidate values for the FNN, and tion algorithms are evaluated in Python-v.3.7.xx versions,
the fitness value of each OWOA solution is evaluated 4GB-RAM and 500 Gigs Ram, and Intel Core i52450N CPU.
using equation (6). The maximum iteration for the proposed OWOA model is
• Step 8: All the fitness values are sorted in ascend- 100, and the total number of candidates is 50.
ing order using control parameters for all the available This dataset is used to train FNN to evaluate the model
whales (agents). accuracy. Four standard meta-heuristic algorithms (OWOA,
• Step 9: To choose twenty-five best fitness values using WOA, ABC, and PSO) are used to train a feed forward neural
OWOA and use those best values for next iteration. network (FNN) to measure the suitability and supremacy of
• Step 10: The weight and bias values are checked for the proposed OWOA. We have considered the same number
duplicate occurrence and if a duplicate value is found, of iterations for all the algorithms to check the proposed
it will be replaced with a randomized value. algorithm’s performance unbiasedly.
• Step 11: To repeat 3rd to 6th step for maximum iteration
to train feed forward neural network. A. SIMULATION RESULTS AND DISCUSSION
• Step 12: When termination criteria is met, iteration is In current setup, we used OWOA to evaluate diabetes
stopped and go to step 4 for next iteration. detection efficiency with standard dataset. The impact and
Flow chart of the algorithm is shown in figure 2. improvement of the proposed OWOA model are thoroughly
examined via a comprehensive comparison with several other
models for example OWOA, WOA, ABC and PSO for dia-
betes detection.
To check the efficiency of the model, Table 3 and Table 5
present a comparison of fitness value. Furthermore, accuracy
is measured using another method known as the confusion
matrix. This analysis provides valuable insights into the effec-
tiveness of the proposed OWOA technique and its potential
advantages over other algorithms for diabetes diagnosis.
The performance evaluation technique used in this study is
essential for machine learning classification. With the help of
the confusion matrix, various computations for the model can
be conducted, including the accuracy of the model.
The accuracy of classification is evaluated using eqn. (18)

Accuracy = (X + X ∼ )/(X + X ∼ + A + A∼ ) (18)

FIGURE 2. Flow chart of the proposed algorithm. where X (True Positive), X ∼ (True Negative), A (False Neg-
ative), and A∼ (False Positive).
V. EXPERIMENT CONFIGURATION SETUP It provides details about the input-nodes, hidden-nodes,
In this study, the Pima Indian dataset has been used. The and training data utilized. Notably, the results of TABLE 3
model architecture follows 1 - Y - 1 configuration, with and Table 5 displays the least fitness metric and improved

20404 VOLUME 12, 2024

R. Chatterjee et al.: FNN for Diabetic Prediction Using Oppositional Whale Optimization Algorithm

TABLE 5. Least fitness metric of the proposed algorithms for diabetes TABLE 5. (Continued.) Least fitness metric of the proposed algorithms for
dataset. diabetes dataset.

FIGURE 3. Boxplot of the statistical results for OWOA, ABC, WOA and PSO
algorithms using Diabetes dataset. The results obtained for the OWOA are
clearly better than those obtained for the rest of the tested algorithms.

efficiency which indicate the OWOA algorithm outperforms

the other algorithms (OWOA, ABC, WOA and PSO).

VOLUME 12, 2024 20405

 R. Chatterjee et al.: FNN for Diabetic Prediction Using Oppositional Whale Optimization Algorithm

Algorithm 3 Algorithm for OWOA standard deviation, best value, and least value. These statis-
Initialize input parameters of WOA like population size (nx ), tical results clearly demonstrate that the best, mean, median,
number of control variables (ny ), maximum iteration cycles and worst fitness values obtained by the OWOA method are
(nmax , jumping rate, etc.) closer and therefore assures the advantage of the OWOA
Create initial population of WOA as follows: method. The tuning efficiency of the proposed OWOA has
FOR i in range (1, nx ): do
FOR j in range(1, ny ): do been measured in terms of convergence mobility. The conver-
F(i,j) = xi,min + rand ∗(xi,max -xi,min ) gence graph of the proposed algorithm in comparison with the
END FORLOOP other algorithms are shown in the figure 4. This graph shows
END FORLOOP that the proposed algorithm can more easily escape the local
Generate opposite initial population as follows: optima than other algorithms.
FOR i in range (1, nx ): do
FOR j in range(1, ny ): do
OP(i,j) = xi,min + (xi,max )+ F(i,j) TABLE 6. Efficiency evaluation of OWOA, ABC, WOA and PSO for diabetic
END FORLOOP classification.
END FORLOOP
Evaluate fitness value of current population and opposite
population of all whale search agent using (4)
Based of fitness values, (Mp ) a number of best solutions are
selected.
J=1
WHILE J < Jmax do
FOR j in range(1,mx ):do
Update jth whale particle using two phases namely
searching phase and encircling phase.
END FORLOOP
Chaotic map is used to generate non-repetitive set of random
numbers to cover all possible states in search domain. VI. CONCLUSION
FOR j in range(1,mx ):do
Generate opposite particle of the jth whale particle using In this study, a hybrid OWOA is developed to train the
non-repetitive chaos based jumping rate. FNN to predict diabetic patients based on some parameters.
END FORLOOP This study combines the oppositional learning technique to
Calculate fitness value of current population and opposite address two significant bottlenecks observed in the original
population of all whale agent using (4). WOA: slow convergence and the tendency to get stuck in
Based on fitness values, Mp number of best solutions are
selected from current and opposite population. local minima. By incorporating OBL, the proposed approach
J = J +1 achieves smoother convergence, resulting in an improved
END WHILELOOP convergence rate and enhanced search capabilities. The study
Display optimal results. applies the OWOA technique to solve a diabetic dataset prob-
lem. The results show the efficacy of OWOA as convergence
accuracy and ability to find global optima, which makes it a
robust algorithm for various scientific and research domains.

REFERENCES
[1] D. Svozil, V. Kvasnicka, and J. Pospichal, ‘‘Introduction to multi-layer
feed-forward neural networks,’’ Chemometric Intell. Lab. Syst., vol. 39,
no. 1, pp. 43–62, Nov. 1997, doi: 10.1016/S0169-7439(97)00061-0.
[2] G. Bebis and M. Georgiopoulos, ‘‘Feed-forward neural networks,’’ IEEE
Potentials, vol. 13, no. 4, pp. 27–31, Oct. 1994, doi: 10.1109/45.329294.
[3] B. Zhang, W. Huang, L. Gong, J. Li, C. Zhao, C. Liu, and D. Huang,
‘‘Computer vision detection of defective apples using automatic light-
ness correction and weighted RVM classifier,’’ J. Food Eng., vol. 146,
pp. 143–151, Feb. 2015, doi: 10.1016/j.jfoodeng.2014.08.024.
[4] M. Arora and V. Kansal, ‘‘Character level embedding with deep convo-
lutional neural network for text normalization of unstructured data for
Twitter sentiment analysis,’’ Social Netw. Anal. Mining, vol. 9, no. 1, p. 12,
Mar. 2019, doi: 10.1007/s13278-019-0557-y.
FIGURE 4. Convergence graph of OWOA, WOA, ABC and PSO for diabetes [5] J. Zhang, H. Li, Z. Tang, Q. Lu, X. Zheng, and J. Zhou, ‘‘An improved
dataset. quantum-inspired genetic algorithm for image multilevel thresholding seg-
mentation,’’ Math. Problems Eng., vol. 2014, pp. 1–12, Jan. 2014.
[6] S. Pervaiz, Z. Ul-Qayyum, W. H. Bangyal, L. Gao, and J. Ahmad,
‘‘A systematic literature review on particle swarm optimization techniques
To assess the advantage of the suggested OWOA model for for medical diseases detection,’’ Comput. Math. Methods Med., vol. 2021,
improvement, the statistical values of OWOA are compared pp. 1–10, Sep. 2021, doi: 10.1155/2021/5990999.
[7] R. Poli, J. Kennedy, and T. Blackwell, ‘‘Particle swarm optimization,’’
with ABC, WOA and SSO in Table 4 and Table 6. Figure 3 Swarm Intell., vol. 1, no. 1, pp. 33–57, Jun. 2007, doi: 10.1007/s11721-
presents a boxplot of statistical variables like mean, median, 007-0002-0.

20406 VOLUME 12, 2024

R. Chatterjee et al.: FNN for Diabetic Prediction Using Oppositional Whale Optimization Algorithm

[8] W. H. Bangyal, A. Hameed, J. Ahmad, K. Nisar, M. R. Haque, [28] S. Gao, K. Wang, S. Tao, T. Jin, H. Dai, and J. Cheng, ‘‘A state-of-
A. A. A. Ibrahim, J. J. P. C. Rodrigues, M. A. Khan, D. B. Rawat, the-art differential evolution algorithm for parameter estimation of solar
and R. Etengu, ‘‘New modified controlled bat algorithm for numeri- photovoltaic models,’’ Energy Convers. Manage., vol. 230, Feb. 2021,
cal optimization problem,’’ Comput., Mater. Continua, vol. 70, no. 2, Art. no. 113784, doi: 10.1016/j.enconman.2020.113784.
pp. 2241–2259, 2022. [29] K. Lenin, ‘‘Real power loss reduction by duponchelia fovealis optimiza-
[9] K. V. Price, ‘‘Differential evolution,’’ in Handbook of Optimization. Berlin, tion and enriched squirrel search optimization algorithms,’’ Soft Comput.,
Germany: Springer, 2013, pp. 187–214, doi: 10.1007/978-3-642-30504- vol. 24, no. 23, pp. 17863–17873, Dec. 2020, doi: 10.1007/s00500-020-
7_8. 05036-x.
[10] M. Dorigo, M. Birattari, and T. Stutzle, ‘‘Ant colony optimization,’’ [30] M. S. Turgut, H. M. Sağban, O. E. Turgut, and Ö. T. Özmen, ‘‘Whale
IEEE Comput. Intell. Mag., vol. 1, no. 4, pp. 28–39, Nov. 2006, doi: optimization and sine–cosine optimization algorithms with cellular topol-
10.1109/MCI.2006.329691. ogy for parameter identification of chaotic systems and Schottky barrier
[11] X.-S. Yang and S. Deb, ‘‘Cuckoo search: Recent advances and applica- diode models,’’ Soft Comput., vol. 25, no. 2, pp. 1365–1409, Jan. 2021,
tions,’’ Neural Comput. Appl., vol. 24, no. 1, pp. 169–174, Jan. 2014, doi: doi: 10.1007/s00500-020-05227-6.
10.1007/s00521-013-1367-1. [31] D. B. Prakash and C. Lakshminarayana, ‘‘Optimal siting of capaci-
[12] D. Karaboga, ‘‘Artificial bee colony algorithm,’’ Scholarpedia, vol. 5, tors in radial distribution network using whale optimization algorithm,’’
no. 3, p. 6915, Mar. 2010, doi: 10.4249/scholarpedia.6915. Alexandria Eng. J., vol. 56, no. 4, pp. 499–509, Dec. 2017, doi:
[13] S. Mirjalili and A. Lewis, ‘‘The whale optimization algorithm,’’ 10.1016/j.aej.2016.10.002.
Adv. Eng. Softw., vol. 95, pp. 51–67, May 2016, doi: [32] L. Haghnegahdar and Y. Wang, ‘‘A whale optimization algorithm-trained
10.1016/j.advengsoft.2016.01.008. artificial neural network for smart grid cyber intrusion detection,’’ Neu-
[14] Y.-R. Zeng, Y. Zeng, B. Choi, and L. Wang, ‘‘Multifactor-influenced ral Comput. Appl., vol. 32, no. 13, pp. 9427–9441, Jul. 2020, doi:
energy consumption forecasting using enhanced back-propagation 10.1007/s00521-019-04453-w.
neural network,’’ Energy, vol. 127, pp. 381–396, May 2017, doi: [33] H. Guo, J. Zhou, M. Koopialipoor, D. Jahed Armaghani, and M. M. Tahir,
10.1016/j.energy.2017.03.094. ‘‘Deep neural network and whale optimization algorithm to assess flyrock
[15] P. Hosseinioun, M. Kheirabadi, S. R. K. Tabbakh, and R. Ghaemi, induced by blasting,’’ Eng. Comput., vol. 37, no. 1, pp. 173–186, Jan. 2021,
‘‘A new energy-aware tasks scheduling approach in fog computing using doi: 10.1007/s00366-019-00816-y.
hybrid meta-heuristic algorithm,’’ J. Parallel Distrib. Comput., vol. 143, [34] S. H. Yeung and K. F. Man, ‘‘Multiobjective optimization,’’
pp. 88–96, Sep. 2020, doi: 10.1016/j.jpdc.2020.04.008. IEEE Microw. Mag., vol. 12, no. 6, pp. 120–133, Oct. 2011, doi:
10.1109/MMM.2011.942013.
[16] N. Manikandan, N. Gobalakrishnan, and K. Pradeep, ‘‘Bee optimization
[35] D. Karaboga, B. Gorkemli, C. Ozturk, and N. Karaboga, ‘‘A comprehen-
based random double adaptive whale optimization model for task schedul-
sive survey: Artificial bee colony (ABC) algorithm and applications,’’ Artif.
ing in cloud computing environment,’’ Comput. Commun., vol. 187,
Intell. Rev., vol. 42, no. 1, pp. 21–57, 2014, doi: 10.1007/s10462-012-
pp. 35–44, Apr. 2022, doi: 10.1016/j.comcom.2022.01.016.
9328-0.
[17] S. Malik and D. Kim, ‘‘Prediction-learning algorithm for efficient energy
[36] S. Ghosh and D. Chatterjee, ‘‘Artificial bee colony optimization based
consumption in smart buildings based on particle regeneration and velocity
non-intrusive appliances load monitoring technique in a smart home,’’
boost in particle swarm optimization neural networks,’’ Energies, vol. 11,
IEEE Trans. Consum. Electron., vol. 67, no. 1, pp. 77–86, Feb. 2021, doi:
no. 5, p. 1289, May 2018, doi: 10.3390/en11051289.
10.1109/TCE.2021.3051164.
[18] K. Li, Z. Ma, P. Fu, and S. Krishnaswamy, ‘‘Quantitative evaluation [37] P. Stephan, T. Stephan, R. Kannan, and A. Abraham, ‘‘A hybrid artificial
of surface crack depth with a scanning laser source based on particle bee colony with whale optimization algorithm for improved breast cancer
swarm optimization-neural network,’’ NDT & E Int., vol. 98, pp. 208–214, diagnosis,’’ Neural Comput. Appl., vol. 33, no. 20, pp. 13667–13691,
Sep. 2018, doi: 10.1016/j.ndteint.2018.05.011. Oct. 2021, doi: 10.1007/s00521-021-05997-6.
[19] Y. Xu, C. Hu, Q. Wu, S. Jian, Z. Li, Y. Chen, G. Zhang, Z. Zhang, [38] S. Mahdavi, S. Rahnamayan, and K. Deb, ‘‘Opposition based learning:
and S. Wang, ‘‘Research on particle swarm optimization in LSTM neural A literature review,’’ Swarm Evol. Comput., vol. 39, pp. 1–23, Apr. 2018,
networks for rainfall-runoff simulation,’’ J. Hydrol., vol. 608, May 2022, doi: 10.1016/j.swevo.2017.09.010.
Art. no. 127553, doi: 10.1016/j.jhydrol.2022.127553.
[20] M. Ilbeigi, M. Ghomeishi, and A. Dehghanbanadaki, ‘‘Prediction and
optimization of energy consumption in an office building using artificial
neural network and a genetic algorithm,’’ Sustain. Cities Soc., vol. 61,
Oct. 2020, Art. no. 102325, doi: 10.1016/j.scs.2020.102325. RAJESH CHATTERJEE received the M.C.A.
[21] J. Wang, J. Bei, H. Song, H. Zhang, and P. Zhang, ‘‘A whale opti- degree from Jawaharlal Nehru University. He is
mization algorithm with combined mutation and removing similarity currently pursuing the Ph.D. degree with Usha
for global optimization and multilevel thresholding image segmenta- Martin University, Ranchi, Jharkhand, India. He is
tion,’’ Appl. Soft Comput., vol. 137, Apr. 2023, Art. no. 110130, doi: currently an Assistant Vice President in an inter-
10.1016/j.asoc.2023.110130. national bank. He has published extensively and
[22] M. Li, X. Yu, B. Fu, and X. Wang, ‘‘A modified whale optimiza- notably in the field of data science and machine
tion algorithm with multi-strategy mechanism for global optimization intelligence. His research interests include mobile
problems,’’ Neural Comput. Appl., vol. 2023, pp. 1–14, Jan. 2023, doi: meta heuristic algorithm, natural language pro-
10.1007/s00521-023-08287-5. cessing, machine learning, deep learning, and
[23] P. Kaladevi, V. V. Punitha, D. Muthusankar, and R. Praveen, ‘‘Breast cancer computer vision.
diagnosis using orca predation optimization algorithm,’’ J. Intell. Fuzzy
Syst., vol. 45, no. 3, pp. 3855–3873, Aug. 2023, doi: 10.3233/jifs-231176.
[24] R. A. Musheer, C. K. Verma, and N. Srivastava, ‘‘Novel machine
learning approach for classification of high-dimensional microarray
MOHAMMAD AMIR KHUSRU AKHTAR
data,’’ Soft Comput., vol. 23, no. 24, pp. 13409–13421, Dec. 2019, doi:
received the M.Tech. and Ph.D. degrees from the
10.1007/s00500-019-03879-7.
Department of Computer Science and Engineer-
[25] R. Aziz, C. K. Verma, M. Jha, and N. Srivastava, ‘‘Artificial neural network
classification of microarray data using new hybrid gene selection method,’’ ing, Birla Institute of Technology, Mesra, Ranchi,
Int. J. Data Mining Bioinf., vol. 17, no. 1, p. 42, 2017. Jharkhand, India, in 2009 and 2015, respectively.
[26] R. Aziz, C. K. Verma, and N. Srivastava, ‘‘Artificial neural network He is currently an Associate Professor with the
classification of high dimensional data with novel optimization approach Faculty of Computing and Information Technol-
of dimension reduction,’’ Ann. Data Sci., vol. 5, no. 4, pp. 615–635, ogy, Usha Martin University, Ranchi. He is the
Dec. 2018, doi: 10.1007/s40745-018-0155-2. author of over 20 peer-reviewed publications. His
[27] Y. Zhang, D.-W. Gong, X.-Z. Gao, T. Tian, and X.-Y. Sun, ‘‘Binary differ- research interests include mobile ad-hoc networks,
ential evolution with self-learning for multi-objective feature selection,’’ natural language processing, machine learning, deep learning, metaverse,
Inf. Sci., vol. 507, pp. 67–85, Jan. 2020. and feature engineering.

VOLUME 12, 2024 20407

 R. Chatterjee et al.: FNN for Diabetic Prediction Using Oppositional Whale Optimization Algorithm

DINESH KUMAR PRADHAN received the Ph.D. SAHIL VERMA (Senior Member, IEEE) received
degree in computer science and engineering the B.Tech. and M.Tech. degrees in computer
from NIT Durgapur. He is currently an Assis- science and engineering from Maharishi Markan-
tant Professor in information technology with deshwar University, Mullana, Ambala, India,
the Dr. B. C. Roy Engineering College, Durgapur, in 2012, and the Ph.D. degree in computer sci-
India, where he has been a Faculty Member, since ence and engineering from Jaipur National Uni-
2009. He has published extensively and notably in versity, Jaipur, India, in 2017. He is currently with
the field of data science and machine intelligence. the Department of Computer Science and Engi-
Outside of professional interests, he travels widely, neering, Chandigarh Group of Colleges, Chandi-
reads, writes, sails, and believes in happy living. garh University, Mohali, India. He has authored
or coauthored articles in reputed top-cited journals, including the IEEE
TRANSACTIONS ON INDUSTRIAL INFORMATICS, IEEE INTERNET OF THINGS JOURNAL,
ACM Transactions on Internet Technology, and IEEE ACCESS. He is
a Reviewer of top-cited journals, including the IEEE TRANSACTIONS ON
INTELLIGENT TRANSPORT SYSTEMS, IEEE TRANSACTIONS ON NETWORK SCIENCE AND
ENGINEERING, IEEE ACCESS, Neural Computing and Applications (Springer),
Human-Centric Computing and Information Sciences (Springer), Mobile
Networks and Applications (Springer), Journal of Information Security and
Applications (Elsevier), Mobile Information Systems (Hindawi), Interna-
tional Journal of Communication Systems (Wiley), and Security and Com-
munication Networks (Hindawi). He is a Professional Member of ACM and
IAENG. He has many research contributions in the area of cloud computing,
the Internet of Things, vehicular ad-hoc networks, WSN, and MANET. Some
of his research findings are published with top-cited publishers, including
IEEE, Wiley, Springer, ACM, Elsevier, MDPI, and various international
conferences of repute. He has chaired many sessions in international con-
FALGUNI CHAKRABORTY received the M.Tech. ferences. He is also a member of editorial board of many journals, a TPC
degree in operations research from NIT Durgapur, member, and an organizing committee member in various international
West Bengal, India, in 2014, and the Ph.D. degree conferences. He has also participated as an author in various international
from the Department of Computer Science and conferences. He is involved in many research and development activities. He
Engineering, NIT Durgapur, in 2021. He is cur- has visited two international universities as VSB, Technical University of
rently an Assistant Professor with the Master of Ostrava, Ostrava, Czech Republic; and University of Milan, Milan, Italy, for
Computer Application Department, Dr. B. C. Roy research collaborations. He has guided 14 PG students, three Ph.D. students,
Engineering College, Durgapur, West Bengal. He and six are ongoing.
has 14 years of experience in academics and three
years IT industry. He is the author of over ten
peer-reviewed publications in reputed journals like Applied Soft Computing
(Elsevier) and Soft Computing (Springer). His research interests include RUBA ABU KHURMA received the bachelor’s
computer vision, evolutionary algorithms, AI & machine learning, and deep and master’s degrees (Hons.) in computer sci-
learning. ence from Yarmouk University, in 2004 and 2007,
respectively, and the Ph.D. degree (Hons.) in
computer science from the University of Jordan,
in 2021. She was an Assistant Professor with
the Computer Science Department, Al-Ahliyya
Amman University. She was a Researcher in
cybersecurity and machine learning with the
Department of Cybersecurity and Digital Foren-
sics, Al-Balqa Applied University. She has published many articles in
reputable journals, book chapters, and international conferences, such as
CEC, ICPRAM, ECTA, ITISE, PICIT, and Evo-applications.

MARIBEL GARCÍA-ARENAS received the Ph.D.

degree in computer science from the University of
Granada, in 2003. From 2003 to 2007, she was
MOHIT KUMAR received the Ph.D. degree with the University of Jaén, teaching computer
from Jaipur National University. He has pub- and telecommunication engineering. She has been
lished articles in reputed top-cited journals like with the University of Granada, since 2007, where
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, she is currently a Senior Lecturer with the Archi-
IEEE TRANSACTIONS ON NETWORK SCIENCE AND tecture and Technology of Computers Department,
ENGINEERING, HCIS (Springer), Journal of Infor- teaching in computer science. Her main research
mation Science and Engineering, Sensors (MDPI), interests include evolutionary computation and
and Symmetry. He also has been granted two high-performance computing. She has worked on several European, national,
patents. He is currently guiding four Ph.D. stu- regional projects being the head researcher for three projects, and two of
dents. He has also worked as a consultant in them related to mobility and traffic at national and regional level. She also
government funded projects. His research interests include WSN, the IoT, has more than 25 publications in international journals and a higher number
optimization, ML, AI, DL, and data science. He is a reviewer of top-cited in conferences.
journals.

20408 VOLUME 12, 2024