Hindawi

Journal of Sensors
Volume 2024, Article ID 9853625, 1 page https://doi.org/10.1155/2024/9853625

Retraction
Retracted: Construal Attacks on Wireless Data Storage Applications
and Unraveling Using Machine Learning Algorithm
Journal of Sensors
Received 23 January 2024; Accepted 23 January 2024; Published 24 January 2024

Copyright © 2024 Journal of Sensors. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly
cited.
This article has been retracted by Hindawi following an investigation undertaken by the publisher [1]. This investigation has
uncovered evidence of one or more of the following indicators of systematic manipulation of the publication process:

(1) Discrepancies in scope

1 Journal of Sensors
(2) Discrepancies in the description of the research reported
(3) Discrepancies between the availability of data and the research described
(4) Inappropriate citations
(5) Incoherent, meaningless and/or irrelevant content included in the article
(6) Manipulated or compromised peer review

The presence of these indicators undermines our confidence in the integrity of the article’s content and we cannot,
therefore, vouch for its reliability. Please note that this notice is intended solely to alert readers that the content of this article
is unreliable. We have not investigated whether authors were aware of or involved in the systematic manipulation of the
publication process.
Wiley and Hindawi regrets that the usual quality checks did not identify these issues before publication and have since
put additional measures in place to safeguard research integrity.
We wish to credit our own Research Integrity and Research Publishing teams and anonymous and named external
researchers and research integrity experts for contributing to this investigation.
The corresponding author, as the representative of all authors, has been given the opportunity to register their
agreementordisagreementtothisretraction.Wehavekeptarecordof any response received.
References

[1] P. R. Kshirsagar, H. Manoharan, H. A. Alterazi, N. Alhebaishi,

O. B. J. Rabie, and S. Shitharth, “Construal Attacks on
Wireless Data Storage Applications and Unraveling Using
Machine Learning Algorithm,” Journal of Sensors, vol. 2022,
Article ID 9386989, 13 pages, 2022.
Hindawi
Journal of Sensors
Volume 2022, Article ID 9386989, 13 pages https://doi.org/10.1155/2022/9386989

Research Article
Construal Attacks on Wireless Data Storage Applications and
Unraveling Using Machine Learning Algorithm
Pravin R. Kshirsagar, Hariprasath Manoharan, Hassan A. Alterazi , Nawaf
Alhebaishi , Osama Bassam J. Rabie , and S. Shitharth 1 Department of Artificial
Intelligence, G. H. Raisoni College of Engineering, Nagpur, India
2
Department of Electronics and Communication Engineering, Panimalar Engineering College, Poonamallee, Chennai, India
3
Department of Information Technology, Faculty of Computing and Information Technology,
King Abdulaziz University, Saudi Arabia
4
3 Journal of Sensors
Department of Information Systems, Faculty of Computing and Information Technology, King Abdulaziz University, Saudi
Arabia 5
Department of Computer Science & Engineering, Kebri Dehar University, Kebri Dehar, Ethiopia

Correspondence should be addressed to S. Shitharth; [email protected]

Received 8 July 2022; Accepted 3 August 2022; Published 16 August 2022

Academic Editor: Sweta Bhattacharya

Copyright © 2022 Pravin R. Kshirsagar et al. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly
cited.

Cloud services are a popular concept used to describe how internet-based services are delivered and maintained. The computer
technology environment is being restructured with respect to information preservation. Data protection is of critical importance
when storing huge volumes of information. In today’s cyber world, an intrusion is a significant security problem. Services,
information, and services are all vulnerable to attack in the cloud due to its distributed structure of the cloud. Inappropriate
behavior in the connection and in the host is detected using intrusion detection systems (IDS) in the cloud. DDoS attacks are
difficult to protect against since they produce massive volumes of harmful information on the network. This assault forces the
cloud services to become unavailable to target consumers, which depletes computer resources and leaves the provider exposed
to massive financial and reputational losses. Cyber-analyst data mining techniques may assist in intrusion detection. Machine
learning techniques are used to create many strategies. Attribute selection techniques are also vital in keeping the dataset ’s
dimensionality low. In this study, one method is provided, and the dataset is taken from the NSL-KDD dataset. In the first
strategy, a filtering method called learning vector quantization (LVQ) is used, and in the second strategy, a dimensionality-
simplifying method called PCA. The selected attributes from each technique are used for categorization before being tested
against a DoS attack. This recent study shows that an LVQ-based SVM performs better than the competition in
 detecting threats. defenses do not apply. Transaction and demand floods
assaults, delayed performance assaults, and asymmetrical
assaults may all be referred to as DDoS attacks in the cloud.
A flood of these assaults not only creates traffic but also
1. Introduction imitates that of a genuine user [2]. This makes it difficult for
the target to tell the difference between such a flood of
Cloud computing is the ecosystem in which individuals
attacks and legal traffic, and therefore, they need to provide
pool information, services, and knowledge using system
services to the genuine user. A denial-of-service assault on a
resources offered through the internet. It creates a
commodity causes it to become unavailable or service to
convenient and dynamic infrastructure for computing for
legitimate customers to degrade.
business organisations. There are a variety of dangers and
A physical device, a collection of machines, or a system
difficulties that have emerged with the increased use of the
of computers may constitute a source. An attacker may
computing environment. One of the greatest difficulties to place authorized customers in a state of denial if they can
cloud computing environments is keeping consumer successfully deny the access of the specific part [3]. The
privacy, information leakage, and identification concerns means by which this assault is conducted out varies based
under control [1]. As a result of the unique cloud computing on how far into the OSI and TCP/IP models it is carried out.
infrastructure, old problems have been successfully The implementation of any type of denial-of-service attack
combated, but new issues with infrastructure distribution has a variety of variables at play, including the assault
have emerged. When it comes to cloud computing security, instrument that is used to create bandwidth, the protocol
a major concern is that networking and security systems being targeted, the communications layer, and the kind of
information, cloud architecture, and individual security victim. Assailant motivation is to reduce the amount of
requirements all vary. App layer carries out responses resources available to the legitimate customers to the
implementing the interprocess communication. These minimum needed to deny them. Although many protections
patterns resemble genuine responses, thus conventional may be used to shield vital resources from being attacked in
5 Journal of Sensors
this manner, the flaws that are present in the systems are a detection system (IDS) in order to fulfill the monitoring
fact of computing. An assault against computation’s needs and allow for the differentiation between a regular
confidentiality, trustworthiness, and authenticity is traffic flow and an attack. GOIDS is finding out the specific
underway. Threats such as unauthorized users, asset theft, characteristics in the initial IDS dataset that are best suited
and doing beyond the permitted limits are all often used by to identify DDoS assaults of this low pace. Once the
attackers for information security purposes [4]. The attributes have been chosen, they become inputs to
abovementioned problems may be addressed by the use of classifiers. These machine learning models, namely, the
IDS, which identifies and evaluates whether internet traffic SVM, DT, NB, and MLP, is utilized to identify the assault
is regular or unusual in order to find a solution. The that occurred in the system. According to Prathyusha et al.
emergence of many different intrusion detection systems is (2020) [2], in this article, a novel DDoS detection method
attributed to network setup variability. There are distinct has been proposed by using artificial immune systems. This
benefits and drawbacks to every kind of IDS. IDS are suggested approach can identify dangers and modulate the
disseminated IDS because it use hypervisors to identify biological resistance mechanism to react accordingly. Wang
network hosts and disseminate the results. To investigate et al. (2019) [3], in order to pick the best possible attributes
DDoS attacks in the cloud, machine learning is used to the during the training phase, offer a multilayer perceptions
NSL-KDD dataset [5]. The attributes chosen by both LVQ (MLP) coupled sequential attribute selection. Once it is
and PCA attribute selection approaches are essential for a determined that substantial identification mistakes have
successful implementation of mining algorithms. Attribute been made, the feedback mechanism is built to update the
selection is a classification algorithm. assault detectors to prevent future breaches. Rabbani et al.
(2019) [4] proposed probabilistic-neural network (PSO-
1.1. Review of Literature. Dwivedi et al. (2020) [1], using a PNN) for developing a new attack detector. The first step is
machine learning technique, make a proposal for a new to organize the data such that it is easy to interpret. Then,
grasshopper optimization algorithm (GOA) with a machine the multilayer neural network was used to distinguish
learning algorithm (GOIDS). The plan of action is harmful activities. According to Punitha and Indumathi
implemented based on the implementation of an intrusion (2020) [5], entrusting our data security to a central cloud
database, which uses an algorithm that generates the the DDoS attack detection rate in a power system. The
empire’s own security keys, puts our data security at risk. identification rate is increased utilizing CNNs which are
The suggested system is also capable of detecting and trained and tested in stages known as the training and
monitoring how information is used. ICKGA and trapdoor testing process. According to Shitharth et al. (2020) [11], a
creator are used to generate secret keys for every user, novel DDoS detection method that leverages machine
whereas CP-ABE and key creation use the ICKGA and learning-based classifiers is proposed in a cloud
trapdoor generator. Once the trapdoor generator has verified environment. As input to the classifier, people have gathered
the integrity of the user data in the cloud as well as on the and categorized characteristics that they believe to be
user level, the trapdoor generator kicks in. Using a helpful. Kishirsagar et.al [6, 10] elaborate the use of
dynamically weighted ensemble neural network (DWENN), different algorithms for classification and prediction of
a dynamic classifier that adjusts its sensitivity dynamically benchmark datasets and real time dataset which were useful
to identify DDoS attacks with more strength is finally used. in the emerging all fields and elaborate the use of hybrid
According to Wani et al. (2019) [7], in order to identify artificial intelligence along with optimization techniques for
the DDoS assault in the cloud environment, they developed classification and prediction of various datasets with high
a novel detection technique using SVM. According to the accuracy [12]. The algorithms used in various research
plan, it is compared to NB and RF. According to Shitharth worked were useful in cyber security, mobile computing,
and Sangeetha (2020) [8], a number of distributed denial of and cloud computing for more accurate results with
service (DDoS) assaults has been identified using machine different evaluation parameters.
learning-based models. Attribute selection is utilized to Deepa et al. [13] have devised an ensemble approach to
come up with the optimum attributes. The characteristics combat DDoS assaults. They used four distinct machine
chosen have been trained and evaluated using support learning algorithms in the SDN environment to identify
vector machines (SVM), naive Bayes (NB), ANN, and suspicious network traffic. SVM-SOM method obtained
KNN classifiers. Ghanbari et al. (2020) [9] presented a new superior results, with 98.12% accuracy, than the other ML
DDoS attack detection system that was intended to increase algorithms. A DDoS attack-detection system for SDN was
7 Journal of Sensors
presented by the authors. Two separate security steps were
used. Signature-based attacks were detected by Snort,
which
Journal of Sensors 8

is a tool designed to spot them. Using the SVM classifier The proposed methodology is used for preventing denial
and the DNN machine learning method, they launched an of service attack using a quantization model which
attack classification scheme thereafter. eliminates all attacks using step processing procedures. By
Mašetić et al. and Rao et al. [14, 15] and developed an incorporating the proposed method, unidentified attributes
automated DoS attack categorization method for cloud are directly removed from the system, thus making all data
computing. This research is conducted in stages, such as to revolve in a hassle free environment. Moreover, the
conducting an assault simulation, collecting information, losses that are present in this type of system are reduced
and choosing attributes, before applying categorization to even if the data is stored in the cloud. Furthermore, volume
the results. For this research, data is acquired via mimicking of information in presence of large data set is prevented
the cloud environment and DoS assault, together with using machine learning algorithm where ten initial
Wireshark’s Tshark capability. One of the categorization attributes are completely knowledgeable; thus, it is used as
reference data for preventing external attacks in the system.
models for DoS attacks and standard network activity is the
support vector machine (SVM).
1.4. Objectives. The major objective of proposed work
focuses on deciphering three objectives which is considered
1.2. Research Gap and Motivation. In addition to different
as minimization problem as follows:
methods that are provided in earlier sections, some recent
articles also focused on detecting DoS attacks using (i) To minimize the denial of service attack on
different data set where [18] used CAIDA for experimental datathat is included within the systems and to
verification cases. However, if CAIDA is used, large data provide potential defence for large data set
set cannot be stored in the system thus high case external
attacks is not prevented. In [19–23], data detection in (ii) To incorporate machine learning algorithms
industrial applications is analysed as data in entire segment byrationalization process without describing any
inside the industry must be protected in reaching external dimensions for entire data set
users. Thus, the protection is provided using machine (iii) To categorize and allocate resources based on
learning algorithm with two directional data flow targetcustomers, thus increasing the security of
procedures. Even though bidirectional flow is provided, the data that is provided to all users
amount of data traffic in the system can be handled with
single traffic flow itself, thus preventing less amount of
users. There is clearly a need of a strategic plan to use
2. Distributed DoS
machine learning methods in a methodical manner in order The malicious distributed-denial-of-service (DDoS) assaults
to make comprehensive evaluations possible, as otherwise that plague the internet these days are a major worldwide
built-in issues like collinearity, multicollinearity, and threat. These assaults are deftly executed and use the same
duplication would present in machinemined data. methods of conventional denial of service (DoS) attacks,
Additionally, the use of machine learning methods in data but they are implemented on a larger scale due to the usage
science-driven ways requires integrating all of the key of botnets. In order to spread quickly, a botnet may spread
needs of data science-driven approaches. A modeling may by taking use of malware that infects tens or even hundreds
not fulfill its goal, but if that is the case, the model will of computers which are then used to further spread the
always incorporate aspects of classifier. Integrating machine malware by being managed by an attacker that is targeting a
learning and attribute engineering techniques in a single victim [16]. Attacks on the internet provide an exciting
framework also has a significant impact on the current potential for attackers to take control of users computers
research. In other words, all inclusive experimentation and and generate zombies. By infecting people through worms,
trustworthy results need joint consideration. Trojan horses, or backdoors, the zombies use the tricks of
their trade: compelling links, e-mail content, or trustworthy
1.3. Proposed Methodology. Many existing methods [1–15] sender addresses. Computers linked to the Internet, such as
emphases only on basic attacks where data is processed Web servers, have vulnerabilities and flaws that may be
with low security features. Even many methods does not exploited by attackers using a range of different hacker
incorporate learning techniques for avoiding attacks from methods. This leads to malicious malware being placed on
external users. It is always necessary that a user must these systems, and subsequently to these computers being
acquire knowledge from existing data and unnecessary data placed in a vulnerable position, giving malevolent
must be eliminated using attribute engineering procedures. programmes full control over them. These machines are
The abovementioned technique is carried out in case of often known as “handlers” and “zombies.” The attackers,
intrusion prevention systems where different machine under control of the controllers, have the command of the
learning techniques can be allocated. To overcome the gap zombie army.
that is present in existing methods, proposed method is
incorporated by reducing dimensionality of entire data
handling systems.
9 Journal of Sensors

When an attack is first begun, the assailant controls as

many computer systems as possible, enabling him to initiate
the assault. An estimate for the number of zombies may be
anything from a few hundred to a few thousand. In the
figure below, Figure 1, you can see how a botnet of zombie-
related attacks develops [17]. The size of the botnet impacts
the amount of damage, the intensity, and the range of an
attack. A botnet that may inflict debilitating and catastrophic

Zombies

Handler

Zombies
Handler

Zombies
Handler

Zombies
Attacker Victim
Handler

Zombies

Handler
Zombies

Zombies

Zombies
Control traffic
Flood traffic

Figure 1: DDoS attack architecture.

attacks is a serious threat. For example, just a little amount
of information is given by one zombie. In contrast, on user
devices, meanwhile, the huge amount of zombies that have
risen depletes computer resources. When single connection
speed traffic looks as normal, traffic floods using low packet
rates that are part of a DDoS attack are especially difficult
to detect. Attacks that inflict extreme damage may happen
due to existing detection methods tending to increase the
speed of DDoS attacks. At the present, DDoS assaults are
done through link and packet flooding. This kind of attack
has increased drastically on the Internet because hackers
know where and how data is obtained [17]. This kind of
assault may be carried out because weaknesses in the
protocols, operating systems, and web applications
constantly surface. In such attacks, the most common
motives include money gain, blackmail, hacking, or
personal problems. This usually happens when web-based
media, such as internet poker, social media sites, or internet
shopping, are attacked.
Journal of Sensors 10

Distributed denial of service (DDoS) attack

Attribute engineering

Attribute elimination Attribute adjustment

Attribute selection Attribute normalization

Machine learning
(NB, KNN, SVM, RF)

Evaluation

Figure 2: Strategic level framework for DDoS attack detection.

Attribute selection

NSL-KDD data set Rank based Dimensionally reduction

Selected attribute by LVQ Selected attribute by PCA

Training data Classification models (NB, KNN, SVM, RF, DT)

Test data Predict the label of the test set

Performance evaluation

Figure 3: IDS in cloud environment.

3. Data Set Description and Attributes depending on protocols, traffic, and topologies, as well as evaluating
network protocol tests [14]. The sets of data that In a dispersed test environment, wired network is extremely are accessible
are those that are built from the ground up, costly. Modeling is a widely-used technique in network like a direct data set, and
those that have been obtained from research. It is useful for studying network issues that vary public sources, like a public
data set. When open source
11 Journal of Sensors

Start

Input data Normalization

Analysis

Attribute in
Attack type Normal data
threshold level
Yes

Total data

Update the class attribute

Randomly selected for training and testing data

Save in file

End

Figure 4: Flowchart of SVM algorithm for attack detection.

Table 1: Results of LVQ method. 3.1. Classification Technique. SVM is being used
Parameters NB DT SVM successfully for multiple-class classification, but
Accuracy 0.9286 0.9176 0.9985 researchers are still trying to figure out how to expand it.
The two predominant kinds of multiclass SVM methods at
Recall 0.9176 0.9142 0.9768
this time are hypothesis-based and algorithm-based. The
Precision 0.9814 0.9886 0.9928
first method uses several binary classifiers to construct the
F-measure 0.9486 0.9571 0.9940 overall classifier, whereas the second method directly
software is used to generate a direct dataset, the resulting incorporates all training examples to derive the classifier.
dataset is termed direct data set. If the dataset is made By choosing examples at the edges of the class descriptors,
available to the public, it is called public data set. This the SVM may choose the optimal separating hyper plane
study makes use of a public dataset, NSL-KDD, which is for training inside the attribute space. The SVM model that
deliberated in Figure 3. we create has the number of classes equal to k. All of the
Attribute selection method is a strategy that uses several positive instances are used in training an SVM with
parameters, selecting the ones that are the most significant classifier set I, and all other examples are used in training
and have the greatest effect on the anticipated variable. The an SVM with classifier set II.
data used in attribute selection is not the whole data set,
,
with regard to attribute selections, the addition and deletion Thus, given l training data ðx1, y1Þ ðx2, y2Þ, ⋯:ðxl, ylÞ, i =
of information have no effect on the entire collection.
Attribute selection is done out in the proposed study using 1,2,3, ⋯:l where xi ∈ Rl and yi ∈f1, 2, ⋯kg are the
two different approaches. They are a technique of filtering
and a means of reducing complexity.
Journal of Sensors 12

Comparison
Comparison
of precision
of accuracy
usingusing
LVQ LVQ
31

2.8
0.99 70
Precision measurement
Accuracy
2.6 measurement
0.98
60
2.4
0.97
2.2 50
0.96
2 40
0.95
1.8
0.94 30
1.6
0.93 20
1.4
0.92
1.2 10
0.91
1 80 82 84 86 88 90 92 94 96 98 100
1 1.2 1.4 1.6 1.8 2
Number of iterations
Best epoch
NB DT
SVM
(a)
Comparison of recall using LVQ
0.98

0.97
Recall measurement
0.96

0.95

0.94

0.93

0.92

0.91
80 82 84 86 88 90 92 94 96 98 100
Number of iterations

NB DT
SVM
(b)

Figure 5: Continued.
(c)
 13 Journal of Sensors

Comparison of F-measure using LVQ

100

F-measure (existing)
80

60

40

20

0
3

2.5 3
F-measure (proposed) 2.5
2
Number
2 of total epoch
1.5
1.5
1 1

(d)

Figure 5: Results of LVQ method. (a) Accuracy. (b) Recall. (c) Precision. (d) F-measure.

ass of xj the jth SVM solves the following optimization problem wj T∅ð Þxi + bj ≤−1 + ξij if yi ≠ j, 3
ð Þ

: ð4Þ

wjbjξij 2 i=1 i
ðÞ bias, and slack variables, respectively, thenmapped into a
higher dimensional space by the function.∅ðxiÞ may be

wj T∅ð Þxi + bj ≥ 1 −ξij if yi = j, 2 There is a constant, established a priori, which is C. Qua-

ð Þ
min ( 1 wj Twj + c !〠l ξj ), 1 Since the nonlinear function, w, b, and ξ have weight,

dratic programming issue (shown as equation (1) in the

Journal of Sensors 14

Comparison of accuracy using PCA

3
x = 90
Accuracy measurement y = 2.86 (stacked )
2.5 y = 0.997 (segment)

1.5

0.5

0
80 90 100
Number of iterations
(a)
Comparison of recall using PCA
1

0.9

0.8measurement
Recall
0.7

0.6

0.5

0.4

0.3

0.2

0.1

0
80 85 90 95 100
Number of iterations

NB DT
SVM
(b)

Figure 6: Continued.
 15 Journal of Sensors

Comparison of precision using PCA

Precision measurement (existing)

80

60

40

20

Precision1measurement (proposed)

2
3
Best
2 epoch
3
1

(c)
Comparison of F-measure using PCA
120
F-measure (existing vs proposeed)

100

80

60

40

20

0
1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 3
Number of total epoch

(d)

Figure 6: Results of PCA method. (a) Accuracy. (b) Recall. (c) Precision. (d) F-measure.

Table 2: PCA method results. Table 3: Comparable results of LVQ and PCA.

arameters NB Detection accuracy

Classification algorithms
ccuracy 0.8832 0.9758 LVQ PCA
ecall 0.9673 NB
0.9815 0.9289 0.8832
recision 0.8672 DT
0.9753 0.9397 0.9756
-M 0.9143 SVM
0.9786 0.9985 0.9951
Journal of Sensors 16

Comparison of detection accuracy for classification algorithms

Detection accuracy (existing)

80

60

40

20

0
3
2.5 3
Detection accuracy (proposed) 2.5
2
2
1.5 Best epoch
1.5
1 1

Figure 7: Comparative results of LVQ and PCA.

graphic below) involves searching for the best hyperplane In this section, we will be using the Gaussian kernel

in want to increaseequation (1). Minimizing2/kwjk the Kðx, xiÞ and the Lagrange multiplier. We will change
thekernel function Kðx, xiÞ in a data-dependent
di1/2ðwfference between assault cate-jÞTwj, therefore, Gaussian manner to enhance SVM classifier classification

researchers gories. Data do not exist in a linear format, accuracy. In SVM, the four common functions are linear,
polynomial of degree d, RBF, and MLP. A flowchart
therefore, there is a cost. c j
i Þ. SVM tries to find a
depicting the algorithm’s steps is given in Figure 4. The
procedure of a simulation method that uses support vector
compromise between the regularization term and training machines is shown using this flowchart. The origins from
both equations (1) and (7) are provided in such a way it is
mistakes 1/2 integrated in a single equation for defining the objective
functions as follows,
determinedðwjÞTwj corrections and training mistakes. Once
you havek decision functions from equation (1), you are
n
finished solving for k.
Oi = min 〠 DoSi, Ai, ð8Þ
〠i=1 αijK xð , xiÞ + b1, 5 i=1
ð Þ

where DoSi indicates various attack process. Ai describes

6 different attribute in a system.
j
〠i=1 αi K xð , xiÞ + b : k ð Þ

We state that the value of the choice function for class xi 4. Outcomes
is in the class with the greatest value:
Attribute selection techniques are employed, and the
attribute set that results from this is used for classification.
l Verification measurements are computed by using these
〠 7
class of x = argmaxi=1⋯k αijK xð l, xiÞ + bj: ð Þ i=1
17 Journal of Sensors

theoretical method, which relate to accuracy, precision, thus solving the necessary properties for defining the
recall, and f -measure. learning rate. Figure 8 illustrates the simulation outcomes
and comparison of robustness that is present in both LVQ
4.1. Assessment of Characteristics: LVQ Process. These and PCA.
results in Table 1 and Figure 5 have been obtained from From Figure 8, it is pragmatic that robustness of LVQ is
experiments that follow the research set of data. much reduced as compared to PCA due to dimensionless
Applications of different classifiers like NB, SVM, and DT characteristics. To validate the robustness of LVQ and PCA
are made possible with the deployment of LVQ. With five best epoch is considered but original ranges are chosen
respect to malicious records, the SVM classifier has a from 10 to 100. Due to presence of vector quantization, the
higher performance level as compared to NB and DT. step size is chosen as 20, thus, the following best epoch
such as 20, 40, 60, 80, and 100 is considered. During the
4.2. PCA Strategy: Explore Various Qualities. PCA is used
abovementioned variations, it is much clear that robustness
for dimensionality reduction. Figure 6 shows the findings. of LVQ reduces from 283 to 107 and further reduces for
SVM method from Table 2 does better than NB and DT remaining periods. On the other hand, even though PCA
reduces
300 100% the
Robustness (existing)

amount of
200 67%

100 33%

0 0%
269 304 347 399 433
Best epoch
Robustness (proposed)

300 100%

200 67%

100 33%

0 0%
107 142 187 225 283
Best epoch

Figure 8: Comparison of robustness characteristics.

when it comes to detection accuracy (0.9971 vs. 0.9965). robustness, it is much higher for all epoch periods as
When using the attribute selection technique in this dimension process for data is defined in existing method.
attribute-based selection process, 10 out of 21 attributes are 5. Conclusions
used.
This page attempts to give a basic overview of the different
4.3. Comparative Results. Attribute selection techniques, DDoS attack methods in use, while also offering an indepth
such as SVM, were used to classifier performance, and the look at potential defenses. An essential part in the overall
findings are summarized in Table 3. Table 3 and Figure 7 data protection process is played by intrusion prevention. A
indicate that SVM performs better for both attribute benchmarking set of NSL-KDD standards is used to
selection methods. Classifying harmful records is best identify intruders for internet information. The study only
performed using an SVM-based approach. uses information that pertain to DDoS attacks. Attributes
such as LVQ and PCA were utilized to categories the
4.4. Robustness Characteristics. In this comparative attacks based on machine learning approaches such as
outcome section, the robustness characteristics with respect SVM, NB, and DT. To verify whether the DDoS attack was
to LVQ and PCA are observed for different iteration occurring, the algorithms’ performance was monitored. Ten
periods, and their changes are simulated. Since more attributes were selected using LVQ, and the remaining ten
amount of data set is present in this process for preventing attributes were selected using PCA. Using an LVQbased
DoS, it is essential to find individual robustness for attribute selection in an SVM model was shown to be more
attributes. Further, the robustness of an algorithm successful in identifying attacks. When compared to other
determines the association between two distinct data set, algorithms, it comes out to be more accuracy, has greater
recall, is more precise, and has a higher F-score.
Journal of Sensors 18

5.1. Policy Implications and Biomedical Engineering: Imaging & Visualization, vol.
8, no. 4, pp. 426–435, 2020.
(i) The proposed DoS model can be incorporated in [7] A. R. Wani, Q. P. Rana, U. Saxena, and N. Pandey, “Analysis
allindustries even with large amount of data set and detection of DDoS attacks on cloud computing
where new security features are enabled environment using machine learning techniques,” in 2019
Amity International Conference on Artificial Intelligence
(ii) By using the enhanced security features, more (AICAI), 2019.
amount of data overflow can be prevented and [8] S. Shitharth and K. Sangeetha, “Enhanced SCADA IDS
even worst type of attacks can be prevented using security by using MSOM hybrid unsupervised algorithm,”
loop formatting procedures International Journal of Web-Based Learning and Teaching
Technologies (IJWLTT), vol. 17, no. 3, 2021.
(iii) All the target systems can process different type of
packets inside a particular device where less [9] M. Ghanbari and W. Kinsner, “Detecting DDoS attacks using
resources are allocated in productions polyscale analysis and deep learning,” International Journal
of Cognitive Informatics and Natural Intelligence (IJCINI),
Data Availability vol. 14, no. 1, pp. 17–34, 2020.
The data that support the findings of this study are available [10] P. Kshirsagar and S. Akojwar, “Optimization of BPNN
from the corresponding author, upon reasonable request. parameters using PSO for EEG signals,” in
ICCASP/ICMMD-2016. Advances in Intelligent Systems
Research, vol. 137, pp. 385– 394, 2016.
Conflicts of Interest
[11] S. Shitharth, N. Satheesh, B. Praveen Kumar, and K.
The authors declare that there are no conflicts of interest Sangeetha, “IDS detection based on optimization based on
regarding the publication of this paper. WI-CS and GNN algorithm in SCADA network,” in
Architectural Wireless Networks Solutions and Security
Issues, vol. 196, no. 1pp. 247– 266, Springer, 2021.
Acknowledgments
[12] S. Shitharth, K. M. Prasad, K. Sangeetha, P. R. Kshirsagar, T.
This work has been done remotely at DAAI Lab, Thu Dau S. Babu, and H. H. Alhelou, “An enriched RPCO-BCNN
Mot University, Vietnam, and i3 LABs, Techno India NJR mechanisms for attack detection and classification in
Institute of Technology, India. SCADA systems,” IEEE Access, vol. 9, pp. 156297–156312,
2021.
[13] V. Deepa, K. M. Sudar, and P. Deepalakshmi, “Design of
References
ensemble learning methods for DDoS detection in SDN
[1] S. Dwivedi, M. Vardhan, and S. Tripathi, “Defense against environment,” in Proceedings of the International
distributed DoS attack detection by using intelligent Conference on Vision Towards Emerging Trends in
evolutionary algorithm,” International Journal of Computers Communication and Networking (ViTECoN), pp. 30-31,
and Applications, vol. 44, no. 3, pp. 219–229, 2022. Vellore, India, 2019.
[2] D. J. Prathyusha and G. Kannayaram, “A cognitive [14] Z. Mašetić, D. Kečo, N. Doǧru, and K. Hajdarević, “SYN
mechanism for mitigating DDoS attacks using the artificial flood attack detection in cloud computing using support
immune system in a cloud environment,” Evolutionary vector machine,” TEM Journal, vol. 6, no. 4, 2017.
Intelligence, vol. 14, no. 2, pp. 607–618, 2021. [15] N. S. Rao, K. C. Sekharaiah, and A. A. Rao, “A survey of
[3] M. Wang, Y. Lu, and J. Qin, “A dynamic MLP-based DDoS distributed denial-of-service (DDoS) defense techniques in
attack detection method using feature selection and ISP domains,” Innovations in Computer Science and
feedback,” Computers & Security, vol. 88, article 101645, Engineering, vol. 32, pp. 221–230, 2019.
2020. [16] G. Bhageerath Chakravorthy, R. Aditya Vardhan, K. Karthik
[4] M. Rabbani, Y. L. Wang, R. Khoshkangini, H. Jelodar, R. Shetty, K. Mahesh, and S. Shitharth, “Handling tactful data
Zhao, and P. Hu, “A hybrid machine learning approach for in cloud using Pkg encryption technique,” in 4th Smart
malicious behaviour detection and recognition in cloud Cities Symposium (SCS 2021), pp. 338–343, 2021.
computing,” Journal of Network and Computer Applications,
[17] R. Aluvalu, V. U. Maheswari, K. K. Chennam, and S.
vol. 151, article 102507, 2020.
Shitharth, “Data security in cloud computing using Abe-
[5] A. A. A. Punitha and G. Indumathi, “RETRACTED based access control,” in Architectural Wireless Networks
ARTICLE: A novel centralized cloud information Solutions and Security Issues, vol. 196, no. 1pp. 47–62,
accountability integrity with ensemble neural network based Lecture notes in network and systems, Springer, 2021.
attack detection approach for cloud data,” Journal of
[18] K. K. Chennam, R. Aluvalu, and S. Shitharth, “An
Ambient Intelligence and Humanized Computing, vol. 12,
authentication model with high security for cloud database,”
no. 5, pp. 4889–4900, 2021.
in Architectural Wireless Networks Solutions and Security
[6] P. Kshirsagar, N. Balakrishnan, and A. D. Yadav, “Modelling Issues, vol. 196, no. 1pp. 13–26, Lecture notes in network
of optimised neural network for classification and prediction and systems,
of benchmark datasets,” Computer Methods in Biomechanics Springer, 2021.
19 Journal of Sensors

[19] A. Bandi, L. Sherpa, and S. M. Allu, “Machine learning

algorithms for DDoS attack detection in cybersecurity,”
Studies in Computational Intelligence, vol. 1027, pp. 269–
281, 2022.
[20] F. Musumeci, A. C. Fidanci, F. Paolucci, F. Cugini, and M.
Tornatore, “Machine-learning-enabled DDoS attacks
detection in P4 programmable networks,” Journal of
Network and Systems Management, vol. 30, no. 1, 2022.
[21] M. Liyanage, Q. V. Pham, K. Dev et al., “A survey on zero
touch network and service management (ZSM) for 5G and
beyond networks,” Journal of Network and Computer
Applications, vol. 203, article 103362, 2022.
[22] T. Shakeel, S. Habib, W. Boulila et al., “A survey on
COVID-19 impact in the healthcare domain: worldwide
market implementation, applications, security and privacy
issues, challenges and future prospects,” in Complex &
Intelligent Systems, pp. 1–32, Springer International
Publishing, 2022.
[23] G. Srivastava, R. H. Jhaveri, S. Bhattacharya et al., “XAI for
cybersecurity: state of the art, challenges, open issues and
future directions,” vol. 1, no. 1, pp. 1–33, 2022, http://arxiv
.org/abs/2206.03585.