Defense Strategies For Epidemic Cyber Security Threats Modeling and Analysis by Using A Machine Learning Approach
Defense Strategies For Epidemic Cyber Security Threats Modeling and Analysis by Using A Machine Learning Approach
ABSTRACT This paper investigates the mathematical modelling of cybercrime attacks on multiple devices
connected to the server. This model is a very successful way for cybercrime, bio-mathematics, and artificial
intelligence to investigate and comprehend the behaviour of mannerisms with harmful intentions in a
computer system. In this computational model, we are studying the factors (i.e., computer viruses, disease
infections, and cyberattacks) that affect connected devices. This compartmental model, SEIAR, represents
the various hardware utilised during the cyberattack. The letters S, E, I, A, and R are used to represent
different stages or groups of individuals in epidemiological models, helping to understand the spread
and control of infectious diseases. The dynamics of the previous model are determined by a series of
differential equations. The dynamics of the preceding model are determined by a system of differential
equations. Numerical solutions of the model are calculated using backpropagated Levenberg-Marquardt
algorithm (BLMA) and a specific optimization algorithm known as the Levenberg-Marquardt algorithm
(LMA). Reference solutions were obtained by using the Runge-Kutta algorithm of order 4 (RK-4). The
backpropagated Levenberg-Marquardt algorithm (BLMA), commonly known as the damped least-squares
(DLS) method. Subsequently, we endeavor to analyze the surrogate solutions obtained for the system and
determine the stability of our approach. Moreover, we aim to ascertain fitting curves to the target solutions
with minimum errors and achieve a regression value of 1 for all the predicted solutions. The outcome of our
simulations ensures that our approach is capable of making precise predictions concerning the behavior of
real-world phenomena under varying circumstances. The testing, validation, and training of our technique
concerning the reference solutions are then used to determine the accuracy of the surrogate solutions obtained
by BLMA. Convergence analysis, error histograms, regression analysis, and curve fitting were used for each
differential equation to examine the robustness and accuracy of the design strategy.
will be a lot of delay. Technology has a significant role usage behaviours as well as cutting-edge technology, such
in increasing efficiency [2]. This means that people in an as surveys, monitoring applications, public opinion [19],
industry must be technologically prepared and accomplish network providers, 3D Point Cloud Upsampling [20], [21]and
their responsibilities significantly more quickly. Software app stores.
is used to administer and regulate the country’s entire Various mathematical models for malware spread have
economy [3]. Computer software is used in different been suggested by various researchers [22], [23]. In the
industries in our lives. In the robotics industry [4], highly field of machine learning, especially when working with
sensitive software is used to boost worker productivity sensitive or proprietary models [24], model stealing attacks
in the robotic universe, agriculture [5], and medicine by highlight the necessity for strong security measures. The
enhancing the ability to analyse medical information [6], agri- improvement of the security of uplink non-orthogonal
cultural production, communication, and easy connection and multiple access (NOMA) systems is the primary emphasis
sharing. of Physical Layer Security [25]. It suggests utilising energy-
But one of the major drop bags of software is hacking harvesting yammers to increase physical layer security
or cyberattacks. In paradigms, the risk associated with while preserving system functionality. It also emphasises
these cyberattacks, such as the industry 4.1, is just named the significance of continued research and development of
in [7] and [8]. To present an industrial security solution appropriate remedies to protect against such assaults and
tailored exclusively for virtual reality (VR) devices. The preserve model producers’ intellectual property [26]. These
emphasis is on solving the security difficulties and risks mathematical models represent the different scenarios that
associated with virtual reality technology, with the goal of accrue in the system during a cyberattack. This model is
ensuring the integrity, confidentiality, and availability of data mostly compartmental and deterministic. Because gadgets
and systems in industrial settings [9]. A cyberattack is a are categorised into separate sections, they are compartmental
cybercriminal attack carried out with one or more computers models: (S) Susceptible(Total), (I) Infected,(A) Asymp-
against a single or several computers or networks [10]. tomatic, (E) Exposed, and (R) Recovered among others.
Cybercriminals utilize a variety of methods to launch These compartments can be obtained by different types
their assaults, including DoS and DDoS attacks, middle of dynamics: SIS, SEIR, SCIR, SI, and SIR just a few
(MitM) attacks, man-in-the-phishing, drive-by threats, spear- names from [23]. These models are important in a variety of
phishing attacks, SQL injection attacks, cross-site scripting domains because differential equation theory may be used to
(XSS) attacks, password attacks, eavesdropping attacks, investigate the performance and interaction of their solutions.
malware attacks, and the use of carefully picked appropriate Discrete techniques, including the Euler, Crank-Nicolson,
information. Framework for the Internet of Things (IoT) and Runge-Kutta methods, among others [27] are frequently
functioning in heterogeneous small cell networks [11], that used to solve ODEs. But in this case, the solutions to a system
is energy-efficient [12]. In order to improve the overall are obtained by the machine learning method. This procedure
performance and sustainability of IoT applications [13], in a achieves a solution that is much more closely related to a
variety of network environments, the key goal is to optimize practical mathematical model. To overcome the constraints of
energy consumption by effectively managing communication solutions acquired via the machine learning method, artificial
resources and network deployment [14]. As a result of neural network (ANN) [28], [29] witch approaches have been
a cyberattack Corrupting address data and papers might suggested in [30] and [31]. It has been shown that an ANN
have extended repercussions for your company’s financial with a hidden layer and a linear activation function may
health. A commonly used control approach, model predictive approximate any function when there are a lot of hidden
control (MPC) [15], may address optimisation control issues neurons [32]. Furthermore, the model will be solved and
with constraints. They have been widely used in modern the results will be explained using a deep learning-based
industrial control systems. Even though some damaged data approach [33].
could be recovered, doing so frequently involves the help
of IT specialists and consumes time and resources that
your company could be using elsewhere. Trojan horses, A. RESEARCH CONTRIBUTION
or worms, are computer programs that resemble how epi- Our research contribution demonstrates the development of
demics propagate among people [16]. A computer virus, such a sophisticated computational method capable of providing
as the human virus, may propagate across an interconnected exact surrogate solutions for complex mathematical models
computer system and transmit viruses from one person to relevant to real-world cases. The invention and use of a
another. Data loss prevention is crucial since erased files and compartmental model, SEIAR, to describe the propagation
information often cannot be recovered. With the ability to and control of cybercrime assaults on many devices linked
access Internet services, smartphones have developed into to a server appears to be the research contribution of
indispensable personal gadgets [17]. App use traces can your paper. This model is designed to assist academics
be gathered by app developers and service providers to and practitioners in better understanding the behavior and
show links between users, applications, and handsets [18]. dynamics of cybercrime assaults, particularly those involving
This survey compiles major trends in smart phone app contagious illnesses. In this research, we use mathematical
B. TECHNICAL DETAIL
In the previous work, the author use a machine-learning
technique for solving the system of equations that repre- Runge-Kutta methods. The backpropagated Levenberg-
sent the real-world phenomena of cyber assault. But the Marquardt algorithm (BLMA) is also used to solve
author can’t mention the particular type of cyber-attack. numerical problems using feed-forward neural networks
And this analysis is not more efficient for the particular (FNNs).
type of cyber-attack. In this work, we employ a modern • Statistical analysis: The outcomes of the trials should
machine learning technique, Artificial Neural Networks be analyzed statistically to see if the variations in
(ANNs) for a particular type of cyber-attack (DDOS attack). performance across models are statistically significant.
ANNs are composed of interconnected nodes that perform • The study presents the testing, validation, and training
mathematical operations on input data to generate outputs, of the reference data set in order to give the value of
known as the Feed-Forward Neural Network (FNN) [35]. the approximate solution using BLMA. Additionally,
The intention of comparing an ANN to the Runge-Kutta a thorough description is given of the testing, validation,
technique is to assess the ANN’s performance in solving and training procedures.
ODEs [36]. Because the Runge-Kutta technique is a well- • The study employs a number of methods to assess
established and frequently used numerical method for solving the robustness and correctness of the design strat-
ODEs, comparing the performance of the ANN to this egy, including convergence analysis, error histograms,
method may give insights into the efficacy and accuracy regression analysis, and curve fitting. These meth-
of the ANN approach. An ANN is a machine-learning ods are used for the analysis of each differential
strategy that is useful for handling and processing linear equation.
situations, converges quicker than other approaches, and
is regarded to be an efficient optimization method. The D. COMPARATIVE EXPERIMENTS FOR SEIAR MODEL
article should provide a detailed description of the methods
Comparative experiments can give important insights into
and their implementations, including the neural network
a proposed model’s complex nature, but they are not
design, the training procedure, and the precise parameters
always feasible or practicable. In our example, we created
employed.
a revolutionary SEAIR cyber security model based on a
machine learning process that especially targets DDOS
C. EXPERIMENTAL EVALUATION GUIDELINES assaults. We think that our model offers a novel and
• The SEIAR compartmental model is used to depict effective strategy for mitigating such threats and that it
the various hardware employed during the cyberat- has the potential to dramatically improve computer system
tack. The variables utilized in the differential equa- security. We also tested our approach against the well-
tions are included in the detailed description of the known RK-4 method and obtained good results. While we
model. recognize the value of comparable tests, we are confident
• The study presents the numerical solution strategies in the advanced nature of our suggested model based on
used to discover the general solution to differential rigorous mathematical analysis and detailed performance
equations, including the Machine learning, ANN, and evaluation.
FIGURE 8. Analysis of performance function in terms of mean square error for the system of (ODE) of case 1.
software to remove the action of a DDOS attack, such a device damage rates, µ is the normal device damaged. After
time-sensitive immunity will exist at the rate δA and ψI, losing their temporal susceptibility, recovered devices finally
accordingly. In the absence of this, devices will be eliminated return at an η rate towards the unprotected compartment.
from the computer at the rate of (µ + ξ1 ) and (µ +ξ2 )I where The corresponding system of ordinary differential equations
ξ1 and ξ2 the attack on the A and I compartment results in represents the mathematical dynamic model of the preceding
Figure (2): The total number of devices (N) represents infectious and
dS infected devices in these compartments.
= A − βS(I + λA − µS + ηR), S(0) = S0
dt N (t) = S(t) + I (t) + E(t) + R(t) + A(t). (2)
dE
= βS(I + λA) − (α + µ)E,
E(0) = E0 The behaviour of the model was simulated with the
dt
following 1000 connected devices: S(t0 ) = 998, E(t0 ) = 0,
dI
= P1 αE − (ψ + µ + ξ2 )I , I (0) = I0 I(t0 ) = 1, A(t0 ) = 1, R(t0 ) = 0. All devices were assumed to
dt
be susceptible at time t = t0 .
dA
= (1 − P1 − P2 )αE − (δ + µ + ξ1 )A, A(0) = A0
dt
dR = ψI + δA + δP2 αE − (η + µ).
III. DESIGN METHODOLOGY
R(0) = R0
dt This section discusses a machine learning strategy based on
(1) an artificial neural network and emphasises guided neural
FIGURE 13. Analysis of performance function in terms of mean square error for the system of (ODE) of case 2.
processes (ANN). An ANN is a network of interconnected Neural Network (FNN). In this instance, the hidden layer
neurons that can analyse several inputs but only generate contains neutrons from the [32]. This system of differential
one output. Using a multi-layer perceptron, we may optimise equations is solved using a different strategy, such as the
the number of hidden units (MLP). MLPs are artificial Metropolis-Hastings technique derived from the Markov
neural networks composed of interconnected nodes that Chain Monte Carlo discussed in [41]. To reduce the log loss
perform mathematical operations on input data to generate function and achieve a solution closer to the real system
outputs [40]. It is frequently referred to as the Feed-Forward of ordinary differential equations (ODEs), In the previous
work DDOS model is solve by employed three optimization an effective machine learning approach for handling and
techniques such as Conjugate-Gradient method (CG) [42], processing linear scenarios, converging at a faster rate than
Broyden-Fletcher-Goldfarb-Shanno method (BFGS) [43], other methods. ANNs are sophisticated nonlinear statistical
and Limited-Memory BFGS for Bound-constrained methods models that leverage complex interactions between inputs
(L-BFGS-B) [44]. and outputs to identify new surrogate solutions System
In this paper, we employ a modern machine learning of ODEs. Similar to neurons in the nervous system,
technique, Artificial Neural Networks (ANNs). ANN is ANNs consist of input, hidden, and output layers and
utilize deep learning methods with the help of activation hidden layer.
functions. The algorithm receives input data, processes it, n
X
and generates an output. These algorithms are used for Nj = (Wij Xi + bj ), (3)
different optimization problems, including neural networks, i=1
and are effective for training different types of neural where, respectively, wij represents connection weights, bj
networks. The ANN is thought to be an efficient optimization denotes biassed vectors, and xi denotes inputs. A log-sigmoid
method. Following the usual MLP architecture with one activation function is used in the feed-forward neural network
model, which is defined as: used to obtain the surrogate solution. And how to Create an
1 initial data set using a standard machine learning technique,
fj (x) = . (4) and compare it with the fourth-order Runge-Kutta algorithm.
1 + e−Nj
Train a feedforward neural network (FNN) using the initial
The implementation of FNN-LMA is carried out in two data set [45]. Use the FNN to predict the solution to the
stages, and the complete workflow of the design algorithms ODE system at future time steps using an Adams-Bashforth
is depicted in Figure (1). Figure (1) depicts the mechanism predictor. Validate and refine the FNN using a testing data
set. The FNN is refined using an error backpropagation • In the second step, the BLM method is executed with
algorithm to minimize the error between the predicted and the appropriate hidden neuron settings and test data in the
actual solution. This figure displays the whole workflow of second phase using the ‘‘nftool’’ programme included in the
the design methods. In the first stage, a numerical solution MATLAB package. Furthermore, BLM exploits a reference
is generated using the Runge-Kutta algorithm of fourth order solution and the testing, validation, and training processes
in Mathematica’s ‘‘ND Solve’’ module to generate an initial to provide approximations for various system of equations
data set (Rk4). instances. The NNs-LMT technique uses a single neural
FIGURE 18. Analysis of performance function in terms of mean square error for the system of (ODE) of case 3.
network, as shown in Figure (5). Figure (3) demonstrates the numerical solutions to the system of differential equations
architecture of the system that is connected to the domain specified in equation (1). The solution is stored as a neural
controller. network on the base of the Levenberg–Marquardt algorithm
(BLMA) [46]. Figure (5) depicts a neural network for
IV. RESULTS the system of ODEs. For this compartmental mathematical
We tackle the mathematical problem by using the Feed- model, we use one intermediate layer for each input and
Forward Artificial Neural Network technique to find output as shown in Figure (5). To verify the resilience of the
ANN-Approach, the model was trained 34 times, as shown researchers have proposed Artificial Neural networks, such
in Figure (17), 34 percent of the fitting results are reasonably as Lagares et al.in [27] as an approach to solving boundary
excellent, with an average error of 1.23 × 10−07 . More data and initial value problems. The design algorithm BLMA is
points should be retained to reduce error. From [27] ass in used to investigate the affect variations of the damaging rate
Solution techniques for initial and boundary value problems of connected devices as shown in Figure (2). According to the
by using Levenberg–Marquardt algorithm (BLMA). It has universal approximation theory [47], which uses a twins ANN
been demonstrated how the ANN-approach may be used to provide approximation solutions for the system ODE [48].
to solve a set of ordinary differential equations. Many A single hidden layer feed-forward neural network should be
able to approximate any continuous function. To accomplish The solution of that differential equations can be expressed in
this, the system of ODE’s should be written as follows: term of
S (t) = F1 (S(t), E(t), I (t), A(t), R(t), t), S(t0 ) = S0 , y′ (t) = F(y(t), y(t)) = y0 .
′
(6)
E (t) = F2 (S(t), E(t), I (t), A(t), R(t), t), E(t0 ) = E0 ,
′
In this solution, y’(t) symbolizes the left-hand side, F(y(t),(t))
I ′ (t) = F3 (S(t), E(t), I (t), A(t), R(t), t), I (t0 ) = I0 , indicates the right-hand side, and y(t0 )=y0 represents the
system’s initial condition. We apply an ANN approach with
A′ (t) = F4 (S(t), E(t), I (t), A(t), R(t), t), A(t0 ) = A0 ,
10 hidden layers to fit the function y(t), which may be
R (t) = F5 (S(t), E(t), I (t), A(t), R(t), t), R(t0 ) = R0 .
′
represented in matrix multiplication form:
(5) N (t, w) = Wk2 u(Wk1 + b1 ) + b2 , (7)
where W is weights, t are the inputs and Wk1 to Wkn are machine learning algorithm (FNN-BLM) [51] to illustrate the
the weight matrices and b1 ,b2 are bias term,u is the linear correctness and efficiency of the design algorithm, as stated
activation function Sigmoid and Tanh are two examples [49]. in Tables 3 and 4. The statistics show that the FNN-BLM
All the parameters are represented by W [Wk1 ,b1 ,Wk2 ,b2 ]. technique is legitimate, and the solutions overlap with the
The given solution is also rewrite as; numerical results with minimal absolute errors in the range
of 10−5 to 10−8 .
y(t, w) = y0 + t − t0 N (t, w), (8)
where V. DISCUSSION
The study is carried out by detecting cybercrime in Jordanian
N (t0 , w) ̸ = y0 , (9)
adopting COVID-19 problems and timeline mapping of
such that the solution function y(t0 ,w)=y0 and it’s derivation significant events and cyber attacks to assess targeted
is y’(t0 ,w)=y0 By minimizing the loss function [49], such as sectors and their cyber security [52]. This article discusses
Sigmoid or tanh, among others, the ideal parameter (w) can cybercrime and its issues in Jordan’s criminal justice system,
be obtained. Jordanian legislation dealing with cybercrime and computer-
The FNN-BLM algorithm’s results are compared with the enabled offenses and as well as other crimes committed
Runge-Kutta technique, the least square method [50], and a using electronic devices, [53], [54]. The impact of cybercrime
TABLE 3. Comparing machine-learning-derived solutions with RK-4 for TABLE 5. Comparing machine-learning-derived solutions with RK-4 for
Susceptible devices for case 1. Infected devices for case 1.
includes the theft of personal information from hundreds followed by Karnataka (40), Kerala (38), and the states of
of millions of people. During the last year, events have Andhra Pradesh and Rajasthan (16 each). 99 of the 217 total
touched more than 40 million individuals in the United States, incidents registered under the IT Act 2000 constituted
54 million in Turkey, 20 million in Korea, 16 million in obscene publishing or transmission in electronic form,
Germany, and more than 20 million in China, to name a sometimes known as cyber pornography. This accounts for
few [55]. According to the National Social Crime Records 45.6 percent of all cases [56]. The Keller-Segel system [57],
Bureau, 217 occurrences were registered under the IT Act which models the group behaviour of chemotactic organisms
in 2007, compared to 142 incidents the previous year with both attraction and repulsion terms, may be controlled
(2006), signifying a 52.8 percent rise in 2007.The state of using a mechanism that ensures stability. Its main focus
Maharashtra had the largest percentage of incidents (22.3%), is on finding control mechanisms that can stabilise the
TABLE 7. Comparing machine-learning derived solutions with RK-4 for TABLE 9. Comparing machine-learning-derived solutions with RK-4 for
Recovered device for case 1. Exposed devices for case 2.
TABLE 8. Comparing machine-learning-derived solutions with RK-4 for TABLE 10. Comparing machine-learning-derived solutions with RK-4 for
Susceptible devices for case 2. Infected devices for case 2.
system globally which implies that the population densi- which are commonly represented as a time series of numbers.
ties of the organisms reach a stable state regardless of After creating the dataset, the next step is to use it to train
the initial circumstances. This project will construct an an ANN. The results of RK-4 are then utilized as ‘‘target’’
effective and precise mathematical model to understand the data to train the ANN. The significance of training an ANN
behavior of numerous possible threats outside of a device is to make predictions on previously unseen input data the
network. training process begins with input data being presented to
The first stage is to generate a dataset of solutions to the the network, and the network makes a prediction based on
system of differential equations using a numerical approach, its current weights and biases. The predicted output is then
such as the Runge-Kutta method of order 4 (RK-4). The compared to the target output, and the difference between the
RK-4 method is used in the system of differential equations two is calculated as the error. The backpropagation algorithm
at this stage to provide a collection of discrete solutions, is used to propagate the error back through the network by
TABLE 11. Comparing machine-learning derived solutions with RK-4 for TABLE 13. Comparing machine-learning-derived solutions with RK-4 for
Asymptomatic device for case 2. Susceptible devices for case 3.
TABLE 15. Comparing machine-learning-derived solutions with RK-4 for security experts can acquire insights into the patterns of cyber
Infected devices for case 3.
assaults and identify places where extra security measures
may be required by analyzing the histogram. And the root
of that error divided by the total sample points is known as
the mean square error as shown in Figure (16). And fitting
of the Figure shows the accuracy. Figure (17) illustrates
what happens when the points in this plot are converging
toward zero and are extremely close to zero: this indicates
that the result is more accurate. Demonstrating that the points
are convergent to one (R=1) and that the best validation
performance is 8.1687×10−12 at epoch 106 is given in
Figure (19). The training Figure (20) represents the behavior
of the gradient. The gradient in the optimization method is
for network performance concerning network weights. The
regression Figure compares the target data with RK-4 to
determine the error, as illustrated in Figure (10). We choose
random setups for case 1. Furthermore use the same process
to arrive at the surrogate solution. And using Matlab to plot
the curve solution. In the second case the birth rate, recovery
rate, infected recovered stage, the rate at which the device
crashes (Natural death), and the rate at which devices leave
respectively are variate. And keep constant the infection
rate, devices leave Unprotected either to the infected, contact
TABLE 16. Comparing machine-learning derived solutions with RK-4 for
Recovered device for case 3. rate, and the rate of damaging devices due to attack as
shown in table (17). In the third case the birth rate, recovery
rate, infected recovered stage, the rate at which the device
crashes (Natural death), and the rate at which devices leave
respectively are kept constant. And attribute other variables
such as: infection rate, devices leave exposed either to the
infected, contact rate, and the rate of damaging devices due
to attack as shown in Figure (4). The Figure histogram
calculates the error by comparing the output data to the target.
As illustrated in Figure (6), the mean square error is defined
as the root of the error divided by the total number of
sample points. And the fitting of the Figure demonstrates
accuracy. Figure (6) can help to understand the frequency
and distribution of cyber assaults over time. The histogram
can depict the number of assaults per time interval as well
as the distribution of attack attributes such as duration, kind,
and intensity. Cyber security experts can acquire insights
into the patterns of cyber assaults and identify places where
extra security measures may be required by analyzing the
histogram. Figure (7) depicts what happens when the points
on this plot converge toward zero or are extremely near
zero, indicating that the result is more accurate. Figure (8)
indicates that the points are converging to one (R=1) (10)
samples) of validation data, 15% (150 samples) of testing and that the best validation performance at epoch 77 is
data, and 10 hidden neurons in the fitting network’s hidden 8.1687×1012 . The training Figure (9) depicts the gradient’s
layer for each input as shown in Figure (17). In figures, behavior. The gradient of the optimizing approach is for
the histogram Figure compares the output data with the network performance in terms of network weights. Moreover,
target and computes the error. A histogram analysis can we aim to ascertain curves to the target solutions, intending
help to understand the frequency and distribution of cyber to achieve a regression value of 1 for all the predicted
assaults over time as shown in Figure 16. The histogram can solutions as shown in Figure (10). The same analysis
depict the number of assaults per time interval as well as the histograms (11), fitting (12), performance analysis (13),
distribution of attack attributes such as duration, kind, and regression analysis (14) and training analysis (15) are revised
intensity. The importance of histogram analysis for Cyber for the second and third scenario.
[3] A. S. Alsafran, ‘‘A feasibility study of implementing IEEE 1547 and IEEE [26] H. Jin and Z. Wang, ‘‘Asymptotic dynamics of the one-dimensional
2030 standards for microgrid in the kingdom of Saudi Arabia,’’ Energies, attraction-repulsion Keller–Segel model,’’ Math. Methods Appl. Sci.,
vol. 16, no. 4, p. 1777, Feb. 2023. vol. 38, no. 3, pp. 444–457, Feb. 2015.
[4] R. Pinciroli and C. Trubiani, ‘‘Performance analysis of fault-tolerant multi- [27] I. E. Lagaris, A. Likas, and D. I. Fotiadis, ‘‘Artificial neural networks for
agent coordination mechanisms,’’ IEEE Trans. Ind. Informat., vol. 19, solving ordinary and partial differential equations,’’ IEEE Trans. Neural
no. 9, pp. 9821–9832, Sep. 2023. Netw., vol. 9, no. 5, pp. 987–1000, Sep. 1998.
[5] M. Aizat, A. Azmin, and W. Rahiman, ‘‘A survey on navigation approaches [28] Z. Qu, X. Liu, and M. Zheng, ‘‘Temporal–spatial quantum graph
for automated guided vehicle robots in dynamic surrounding,’’ IEEE convolutional neural network based on Schrödinger approach for traffic
Access, vol. 11, pp. 33934–33955, 2023. congestion prediction,’’ IEEE Trans. Intell. Transp. Syst., vol. 24, no. 8,
[6] R. Chengoden, N. Victor, T. Huynh-The, G. Yenduri, R. H. Jhaveri, pp. 8677–8686, Aug. 2023.
M. Alazab, S. Bhattacharya, P. Hegde, P. K. R. Maddikunta, and [29] Y. Deng, W. Zhang, W. Xu, Y. Shen, and W. Lam, ‘‘Nonfactoid question
T. R. Gadekallu, ‘‘Metaverse for healthcare: A survey on potential answering as query-focused summarization with graph-enhanced multihop
applications, challenges and future directions,’’ IEEE Access, vol. 11, inference,’’ IEEE Trans. Neural Netw. Learn. Syst., pp. 1–15, Mar. 2023.
pp. 12765–12795, 2023. [30] R. J. LeVeque, Finite Difference Methods for Ordinary and Partial
[7] J. Callenes and M. Poshtan, ‘‘Dynamic reconfiguration for resilient state Differential Equations: Steady-State and Time-Dependent Problems.
estimation against cyber attacks,’’ IEEE Trans. Emerg. Topics Comput., Philadelphia, PA, USA: SIAM, 2007.
pp. 1–12, Apr. 2023. [31] P. Ramuhalli, L. Udpa, and S. S. Udpa, ‘‘Finite-element neural networks for
solving differential equations,’’ IEEE Trans. Neural Netw., vol. 16, no. 6,
[8] D. P. Möller, ‘‘Cyberattacker profiles, cyberattack models and scenarios,
pp. 1381–1392, Nov. 2005.
and cybersecurity ontology,’’ in Guide to Cybersecurity in Digital
[32] H. Sug, ‘‘The effect of training set size for the performance of neural
Transformation: Trends, Methods, Technologies, Applications and Best
networks of classification,’’ WSEAS Trans. Comput., vol. 9, pp. 306–1297,
Practices. Berlin, Germany: Springer, 2023, pp. 181–229.
Nov. 2010.
[9] Z. Lv, D. Chen, R. Lou, and H. Song, ‘‘Industrial security solution for [33] Z. Lv, L. Qiao, J. Li, and H. Song, ‘‘Deep-Learning-Enabled security
virtual reality,’’ IEEE Internet Things J., vol. 8, no. 8, pp. 6273–6281, issues in the Internet of Things,’’ IEEE Internet Things J., vol. 8, no. 12,
Apr. 2021. pp. 9531–9538, Jun. 2021.
[10] A. K. Dangi, K. Pant, J. Alanya-Beltran, N. Chakraborty, S. V. Akram, and [34] N. Kanwal, ‘‘Analysis of cybercrimes: A critical perspective,’’ Dept.
K. Balakrishna, ‘‘A review of use of artificial intelligence on cyber security Comput. Sci., NCBA&E Lahore, Pakistan, Tech. Rep., Mar. 2023, vol. 1,
and the fifth-generation cyber-attacks and its analysis,’’ in Proc. Int. Conf. no. 1.
Artif. Intell. Smart Commun. (AISC), Jan. 2023, pp. 553–557. [35] Z. Xiong, X. Li, X. Zhang, M. Deng, F. Xu, B. Zhou, and M. Zeng,
[11] Y. Chen, L. Zhu, Z. Hu, S. Chen, and X. Zheng, ‘‘Risk propagation in ‘‘A comprehensive confirmation-based selfish node detection algorithm for
multilayer heterogeneous network of coupled system of large engineering socially aware networks,’’ J. Signal Process. Syst., pp. 1–19, Apr. 2023.
project,’’ J. Manag. Eng., vol. 38, no. 3, May 2022, Art. no. 04022003. [36] X. Qin, Z. Liu, Y. Liu, S. Liu, B. Yang, L. Yin, M. Liu, and W. Zheng,
[12] H. Jiang, Z. Xiao, Z. Li, J. Xu, F. Zeng, and D. Wang, ‘‘An energy-efficient ‘‘User OCEAN personality model construction method using a BP neural
framework for Internet of Things underlaying heterogeneous small cell network,’’ Electronics, vol. 11, no. 19, p. 3022, Sep. 2022.
networks,’’ IEEE Trans. Mobile Comput., vol. 21, no. 1, pp. 31–43, [37] M. Malik and M. Dutta, ‘‘Feature engineering and machine learning
Jan. 2022. framework for DDoS attack detection in the standardized Internet of
[13] B. Cheng, D. Zhu, S. Zhao, and J. Chen, ‘‘Situation-aware IoT service Things,’’ IEEE Internet Things J., vol. 10, no. 10, pp. 8658–8669,
coordination using the event-driven SOA paradigm,’’ IEEE Trans. Netw. May 2023.
Service Manag., vol. 13, no. 2, pp. 349–361, Jun. 2016. [38] H. Jiang, M. Wang, P. Zhao, Z. Xiao, and S. Dustdar, ‘‘A utility-
[14] P. Chen, H. Liu, R. Xin, T. Carval, J. Zhao, Y. Xia, and Z. Zhao, aware general framework with quantifiable privacy preservation for
‘‘Effectively detecting operational anomalies in large-scale IoT data destination prediction in LBSs,’’ IEEE/ACM Trans. Netw., vol. 29, no. 5,
infrastructures by using a GAN-based predictive model,’’ Comput. J., pp. 2228–2241, Oct. 2021.
vol. 65, no. 11, pp. 2909–2925, Nov. 2022. [39] H. Zhu, M. Xue, Y. Wang, G. Yuan, and X. Li, ‘‘Fast visual tracking with
[15] B. Li, X. Zhou, Z. Ning, X. Guan, and K.-F.-C. Yiu, ‘‘Dynamic event- Siamese oriented region proposal network,’’ IEEE Signal Process. Lett.,
triggered security control for networked control systems with cyber- vol. 29, pp. 1437–1441, 2022.
attacks: A model predictive control approach,’’ Inf. Sci., vol. 612, [40] J. Lee and J. Jin, ‘‘Thickness and refractive index measurements of a thin-
pp. 384–398, Oct. 2022. film using an artificial neural network algorithm,’’ Metrologia, vol. 60,
[16] H. Saini, Y. S. Rao, and T. C. Panda, ‘‘Cyber-crimes and their impacts: no. 2, 2023, Art. no. 025001.
A review,’’ Int. J. Eng. Res. Appl., vol. 2, no. 2, pp. 202–209, 2012. [41] C. P. Robert, G. Casella, C. P. Robert, and G. Casella, ‘‘The metropolis—
[17] T. Li, T. Xia, H. Wang, Z. Tu, S. Tarkoma, Z. Han, and P. Hui, ‘‘Smartphone Hastings algorithm,’’ in Monte Carlo Statistical Methods. Berlin,
app usage analysis: Datasets, methods, and applications,’’ IEEE Commun. Germany: Springer, 1999, pp. 231–283.
Surveys Tuts., vol. 24, no. 2, pp. 937–966, 2nd Quart., 2022. [42] X. Xia, W. Xue, P. Wan, H. Zhang, X. Wang, and Z. Zhang, ‘‘FCGSM:
[18] T. Li, Y. Li, M. A. Hoque, T. Xia, S. Tarkoma, and P. Hui, ‘‘To what extent Fast conjugate gradient sign method for adversarial attack on image
we repeat ourselves? Discovering daily activity patterns across mobile classification,’’ in Innovative Computing Vol 2—Emerging Topics in Future
app usage,’’ IEEE Trans. Mobile Comput., vol. 21, no. 4, pp. 1492–1507, Internet. Berlin, Germany: Springer, 2023, pp. 709–716.
[43] W. Liu, Y. He, X. Wang, Z. Duan, W. Liang, and Y. Liu, ‘‘BFG: Privacy
Apr. 2022.
protection framework for Internet of Medical Things based on blockchain
[19] F. Meng, X. Xiao, and J. Wang, ‘‘Rating the crisis of online public opinion
and federated learning,’’ Connection Sci., vol. 35, no. 1, Dec. 2023,
using a multi-level index system,’’ 2022, arXiv:2207.14740.
Art. no. 2199951.
[20] H. Liu, H. Yuan, J. Hou, R. Hamzaoui, and W. Gao, ‘‘PUFA-GAN: [44] M. Balaaditya and S. D. Dunston, ‘‘Analysis of the effect of adversarial
A frequency-aware generative adversarial network for 3D point cloud training in defending EfficientNet-B0 model from DeepFool attack,’’ in
upsampling,’’ IEEE Trans. Image Process., vol. 31, pp. 7389–7402, 2022. Proc. 3rd Int. Conf. Intell. Commun. Comput. Techn. (ICCT), Jan. 2023,
[21] S. Lu, Y. Ding, M. Liu, Z. Yin, L. Yin, and W. Zheng, ‘‘Multiscale feature pp. 1–7.
extraction and fusion of image and text in VQA,’’ Int. J. Comput. Intell. [45] W. Lyu and Z.-A. Wang, ‘‘Logistic damping effect in chemotaxis models
Syst., vol. 16, no. 1, p. 54, Apr. 2023. with density-suppressed motility,’’ Adv. Nonlinear Anal., vol. 12, no. 1,
[22] D. K. Saini, ‘‘Cyber defense: Mathematical modeling and simulation,’’ Int. pp. 336–355, Sep. 2022.
J. Appl. Phys. Math., vol. 2, no. 5, pp. 312–315, 2012. [46] X. Xie, B. Xie, D. Xiong, M. Hou, J. Zuo, G. Wei, and J. Chevallier, ‘‘New
[23] M. Martcheva, An Introduction to Mathematical Epidemiology, vol. 61. theoretical ISM-K2 Bayesian network model for evaluating vaccination
Berlin, Germany: Springer, 2015. effectiveness,’’ J. Ambient Intell. Humanized Comput., vol. 14, no. 9,
[24] J. Zhang, S. Peng, Y. Gao, Z. Zhang, and Q. Hong, ‘‘APMSA: Adversarial pp. 12789–12805, Sep. 2023.
perturbation against model stealing attacks,’’ IEEE Trans. Inf. Forensics [47] E. Negrini, ‘‘Universal approximation theorem, G. Cybenko,’’ Worcester
Security, vol. 18, pp. 1667–1679, 2023. Polytech. Inst., Tech. Rep., Oct. 2019.
[25] K. Cao, B. Wang, H. Ding, L. Lv, R. Dong, T. Cheng, and F. Gong, [48] Z. Lv, D. Chen, H. Feng, W. Wei, and H. Lv, ‘‘Artificial intelligence in
‘‘Improving physical layer security of uplink NOMA via energy harvesting underwater digital twins sensor networks,’’ ACM Trans. Sensor Netw.,
jammers,’’ IEEE Trans. Inf. Forensics Security, vol. 16, pp. 786–799, 2021. vol. 18, no. 3, pp. 1–27, Aug. 2022.
[49] S. Chakraverty and S. Mall, Artificial Neural Networks for Engineers and [60] J. Naskath, G. Sivakamasundari, and A. A. S. Begum, ‘‘A study on different
Scientists: Solving Ordinary Differential Equations. Boca Raton, FL, USA: deep learning algorithms used in deep neural nets: MLP SOM and DBN,’’
CRC Press, 2017. Wireless Pers. Commun., vol. 128, no. 4, pp. 2913–2936, Feb. 2023.
[50] A. Abazari, M. Zadsar, M. Ghafouri, and C. Assi, ‘‘Detection of cyber- [61] J. Yu, L. Lu, Y. Chen, Y. Zhu, and L. Kong, ‘‘An indirect eavesdropping
physical attacks using optimal recursive least square in an islanded attack of keystrokes on touch screen through acoustic sensing,’’ IEEE
microgrid,’’ in Proc. IEEE Power Energy Soc. Gen. Meeting (PESGM), Trans. Mobile Comput., vol. 20, no. 2, pp. 337–351, Feb. 2021.
Jul. 2022, pp. 1–5. [62] S. Hemavathi and B. Latha, ‘‘FRHO: Fuzzy rule-based hybrid optimization
[51] A. Handa, A. Sharma, and S. K. Shukla, ‘‘Machine learning in for optimal cluster head selection and enhancing quality of service in wire-
cybersecurity: A review,’’ Wiley Interdiscipl. Rev., Data Mining Knowl. less sensor network,’’ J. Supercomput., vol. 79, no. 11, pp. 12238–12265,
Discovery, vol. 9, no. 4, p. e1306, 2019. Jul. 2023.
[52] A. H. Amarullah, A. J. S. Runturambi, and B. Widiawan, ‘‘Analyzing [63] V. Pantelakis, ‘‘Adversarial machine learning attacks against network
cyber crimes during COVID-19 time in Indonesia,’’ in Proc. 3rd Int. Conf. intrusion detection systems,’’ M.S. thesis, School Inf. Technol. Commun.,
Comput. Commun. Internet (ICCCI), Jun. 2021, pp. 78–83. Dept. Digit. Syst., 2023.
[53] R. S. Faqir, ‘‘Cyber crimes in Jordan: A legal assessment on the [64] N. Y.-R. Douha, M. Bhuyan, S. Kashihara, D. Fall, Y. Taenaka, and
effectiveness of information system crimes law no (30) of 2010,’’ Int. J. Y. Kadobayashi, ‘‘A survey on blockchain, SDN and NFV for the smart-
Cyber Criminolog, vol. 7, no. 1, pp. 1–10, 2013. home security,’’ Internet Things, vol. 20, Nov. 2022, Art. no. 100588.
[54] P. Datta, S. N. Panda, S. Tanwar, and R. K. Kaushal, ‘‘A technical review [65] D. Arivudainambi, K. A. V. Kumar, and P. Visu, ‘‘Malware traffic classifi-
report on cyber crimes in India,’’ in Proc. Int. Conf. Emerg. Smart Comput. cation using principal component analysis and artificial neural network for
Informat. (ESCI), Mar. 2020, pp. 269–275. extreme surveillance,’’ Comput. Commun., vol. 147, pp. 50–57, Nov. 2019.
[55] E. F. G. Ajayi, ‘‘Challenges to enforcement of cyber-crimes [66] I. B. Mijoya, S. Khurana, and N. Gupta, ‘‘Performance analysis of hard
laws and policy,’’ J. Internet Inf. Syst., vol. 6, no. 1, pp. 1–12, voting and soft voting techniques on Android malware detection,’’ School
Aug. 2016. Eng. Technol., Sharda Univ., Greater Noida, India, Tech. Rep., Mar. 2023,
[56] R. Montasari, ‘‘Cyber threats and the security risks they pose to national vol. 58, no. 1.
security: An assessment of cybersecurity policy in the United Kingdom,’’ [67] R. Ali, A. Ali, F. Iqbal, A. M. Khattak, and S. Aleem, ‘‘A systematic
Countering Cyberterrorism. 2023, pp. 7–25. review of artificial intelligence and machine learning techniques for cyber
[57] H.-Y. Jin and Z.-A. Wang, ‘‘Global stabilization of the full attraction- security,’’ in Big Data and Security. Berlin, Germany: Springer, 2020,
repulsion Keller–Segel system,’’ 2019, arXiv:1905.05990. pp. 584–593.
[58] K. K. Jean-Claude, ‘‘Understanding the worldwide paths towards the [68] J. Ma and J. Hu, ‘‘Safe consensus control of cooperative-competitive
creation of true intelligence for machines,’’ Faculty Comput. Sci. Distance multi-agent systems via differential privacy,’’ Kybernetika, vol. 58, no. 3,
Learn., Bircham Int. Univ., Madrid, Spain, Tech. Rep., Feb. 2023, vol. 15, pp. 426–439, Sep. 2022.
no. 1. [69] C. Qin, Y. Jin, Z. Zhang, H. Yu, J. Tao, H. Sun, and C. Liu, ‘‘Anti-noise
[59] A. Cohen, N. Nissim, and Y. Elovici, ‘‘Novel set of general diesel engine misfire diagnosis using a multi-scale CNN-LSTM neural
descriptive features for enhanced detection of malicious emails using network with denoising module,’’ CAAI Trans. Intell. Technol., vol. 8,
machine learning methods,’’ Exp. Syst. Appl., vol. 110, pp. 143–169, no. 3, pp. 963–986, Sep. 2023.
Nov. 2018.