Optimizing Network Security and Performance Through the Integration of Hybrid GAN-RNN Models in SDN-based Access Control and Traffic Engineering

Abstract

By offering flexible and adaptable infrastructures Software-Defined Networking (SDN) has emerged as a disruptive technology that has completely changed network provisioning and administration. By seamlessly integrating Hybrid Generative Adversarial Network-Recurrent Neural Network (GAN-RNN) modeling into the foundation of SDN-based traffic engineering and accessibility control methods, this work presents a novel and comprehensive method to improve network efficiency and security. The proposed Hybrid GAN-RNN models address two important aspects of network management: traffic optimization and access control. They combine the benefits of Generative Adversarial Networks (GANs) and Recurrent Neural Networks (RNNs). Traditional traffic engineering techniques frequently find it difficult to quickly adjust to situations that are changing quickly within today's dynamic networking environments. The models' capacity to generate synthetic traffic patterns that nearly perfectly replicate the complexity of real network traffic demonstrates the power of GANs. Network administrators can now allocate resources and routing methods more dynamically, as well as in responding to real-time network inconsistencies, due to this state-of-the-art technology. The technique known as Hybrid GAN-RNN addresses the enduring problem of network security. With their reputation for continuous learning and by utilizing Python software, recurrent neural networks (RNNs) are at the forefront of developing flexible management of access rules. With an incredible 99.4% accuracy rate, the "Proposed GAN-RNN" approach outperforms the other approaches. A comprehensive evaluation of network traffic and new safety risks allow for the immediate modification of these policies. This work is interesting because it combines hybrid GAN-RNN algorithms to strengthen security protocols with adaptive access control while also optimizing network efficiency through realistic traffic modeling.

Full text

(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 14, No. 12, 2023
596 | P a g e
www.ijacsa.thesai.org
Optimizing Network Security and Performance
Through the Integration of Hybrid GAN-RNN
Models in SDN-based Access Control and Traffic
Engineering
Ganesh Khekare1, Dr.K.Pavan Kumar2, Kundeti Naga Prasanthi3, Dr. Sanjiv Rao Godla4,
Venubabu Rachapudi5, Dr. Mohammed Saleh Al Ansari6, Prof. Ts. Dr. Yousef A.Baker El-Ebiary7
Associate Professor, School of Computer Science and Engineering, Vellore Institute of Technology, Vellore, Tamil Nadu, India1
Sr Asst.Prof, Dept.of IT, Prasad V Potluri Siddhartha Institute of Technology, Kanuru, Vijayawada -072
Dept.of CSE, Lakireddy Balireddy College of Engineering, Mylavaram3
Professor, Department of CSE (Artificial Intelligence & Machine Learning), Aditya College of Engineering & Technology,
Surampalem, Andhra Pradesh, India4
Department of Computer Science and Engineering, Koneru Lakshmaiah Education Foundation, Vaddeswaram,
Guntur, Andhra Pradesh, India-5223025
Associate Professor, College of Engineering-Department of Chemical Engineering, University of Bahrain, Bahrain6
Faculty of Informatics and Computing, UniSZA University, Malaysia7
Abstract—By offering flexible and adaptable infrastructures
Software-Defined Networking (SDN) has emerged as a disruptive
technology that has completely changed network provisioning
and administration. By seamlessly integrating Hybrid Generative
Adversarial Network-Recurrent Neural Network (GAN-RNN)
modeling into the foundation of SDN-based traffic engineering
and accessibility control methods, this work presents a novel and
comprehensive method to improve network efficiency and
security. The proposed Hybrid GAN-RNN models address two
important aspects of network management: traffic optimization
and access control. They combine the benefits of Generative
Adversarial Networks (GANs) and Recurrent Neural Networks
(RNNs). Traditional traffic engineering techniques frequently
find it difficult to quickly adjust to situations that are changing
quickly within today's dynamic networking environments. The
models' capacity to generate synthetic traffic patterns that nearly
perfectly replicate the complexity of real network traffic
demonstrates the power of GANs. Network administrators can
now allocate resources and routing methods more dynamically,
as well as in responding to real-time network inconsistencies, due
to this state-of-the-art technology. The technique known as
Hybrid GAN-RNN addresses the enduring problem of network
security. With their reputation for continuous learning and by
utilizing Python software, recurrent neural networks (RNNs) are
at the forefront of developing flexible management of access
rules. With an incredible 99.4% accuracy rate, the "Proposed
GAN-RNN" approach outperforms the other approaches. A
comprehensive evaluation of network traffic and new safety risks
allow for the immediate modification of these policies. This work
is interesting because it combines hybrid GAN-RNN algorithms
to strengthen security protocols with adaptive access control
while also optimizing network efficiency through realistic traffic
modeling.
Keywords—Software-defined networking; generative
adversarial networks; recurrent neural networks; traffic
engineering
I. INTRODUCTION
Network performance directly impacts the efficiency of
operations within an organization. Faster data transfer and
lower latency lead to increased productivity, reduced
downtime, and better user experiences, all of which are critical
in today's fast-paced digital world [1]. Slow or unreliable
networks result in poor user experiences. This can frustrate
customers, employees, and partners, leading to dissatisfaction
and potentially driving them away. Ensuring a high-
performing network enhances user satisfaction and loyalty. In
an era where data is a valuable asset, efficient network
performance is crucial for transferring large volumes of data
quickly and securely [2]. This is especially important for
industries like healthcare, finance, and media, where sensitive
information needs to be transmitted reliably. Cyber security
threats are on the rise, and networks are prime targets for
attacks. A secure network helps protect sensitive data,
prevents unauthorized access, and mitigates risks associated
with data breaches, which can have severe legal, financial, and
reputational consequences. Many industries and organizations
must adhere to strict regulatory compliance standards
regarding data security and privacy. Maintaining a secure
network is essential to meeting these requirements and
avoiding legal penalties. A well-optimized network ensures
that network resources, such as bandwidth and hardware, are
used efficiently. This reduces costs associated with network
maintenance and upgrades while maximizing resource
availability [3]. Network failures or security breaches can
disrupt business operations, leading to downtime and financial
losses. Ensuring network resilience and security is crucial for
business continuity and disaster recovery planning. As
businesses grow, their network needs often grow too. A well-
optimized network can scale to accommodate increased
traffic, new devices, and expanding operations without

(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 14, No. 12, 2023
597 | P a g e
www.ijacsa.thesai.org
sacrificing performance or security. Organizations with high-
performing, secure networks can gain a competitive advantage
[4]. They can offer faster services, better customer
experiences, and innovative solutions that competitors with
subpar networks may struggle to match. Improved network
performance and security enable the adoption of advanced
technologies like IoT (Internet of Things), cloud computing,
and AI, which can drive innovation and digital transformation
within an organization. Innovative solutions are required due
to the constantly changing network infrastructure landscape in
order to improve security and performance. SDN, or software-
defined networking, has become a game-changing network
management technology that gives network administrators the
freedom to customize access control and traffic engineering
[5]. In order to enhance network performance and security in
SDN settings, this study investigates a unique method that
combines hybrid generative adversarial networks (GANs) with
recurrent neural networks (RNNs). The project intends to
address important issues in traffic engineering and access
control by smoothly integrating these cutting-edge machine
learning approaches, leading to ultimately more effective and
secure network operations [6]. The article digs into the
principles of Software-Defined Networking (SDN) and
discusses how it applies to contemporary network architecture.
It gives an overview of how SDN can provide centralized
network administration and dynamic traffic engineering by
separating the control and data planes. As a prelude to the
suggested remedy, the section also illustrates the difficulties in
optimizing traffic flows in SDN settings [7].
The use of techniques based on machine learning has a lot
of potential to enhance network performance in several ways.
ML algorithms have the capacity to analyses network data,
adjust to changing circumstances, and make choices in real
time, which can improve network operations' efficiency,
dependability, and security. ML models can spot unusual
network activity that may indicate security risks or
performance problems. Network assaults may be swiftly
detected and responded to by intrusion detection systems
(IDS) and intrusion prevention systems (IPS) driven by ML,
improving security while minimizing service interruption [8].
Applications will perform better and use resources more
effectively as a result of getting the resources they require
when they need them. To maintain equitable server utilization,
ML models may track server load and distribute requests that
arrive around servers. Based on past data and user behavior,
ML may help forecast future network requirements [9]. This
can assist network administrators in making plans for
infrastructure or capacity modifications so that the network is
ready to meet growing demand. For instance, during periods
of high demand, they might give priority to particular sorts of
traffic. GANs use a loss function that guides the training
process. The generator's loss depends on how well the
discriminator is fooled, while the discriminator's loss is based
on its ability to distinguish real from fake data. The training
aim to find a balance where the generator generates highly
convincing data and the discriminator becomes uncertain
about its classifications. Access control in the context of SDN
is covered in the second part. It explains the idea of dynamic
access control lists and discusses the significance of access
control for network security [10]. It examines the difficulties
with access control in SDN, highlighting the requirement for
more sophisticated and flexible security mechanisms.
The merging of Hybrid GANs and RNNs, the research's
main novelty, is presented in the publication. It describes how
RNNs may examine this data to find patterns and
abnormalities using synthetic network traffic data produced by
GANs. This hybrid strategy tries to simultaneously improve
network security and performance. The practical use of the
hybrid GAN-RNN models for traffic engineering is covered in
this section. It offers information on how dynamic network
traffic flow optimization may be accomplished using synthetic
traffic data produced by GANs. The advantages of this
strategy are explored in terms of decreased congestion,
enhanced Quality of Service (QoS), and effective resource
utilization. The research examines the use of hybrid GAN-
RNN models in SDN access control. It explains how RNNs
may inspect network traffic data for irregularities and security
risks. The flexibility of this strategy to changing security
threats is discussed, as is how it enhances access control by
dynamically updating access lists in response to threat
detection in real time. The article covers the overall effects of
incorporating Hybrid GAN-RNN models into SDN settings,
highlighting the enhancements to network security and
performance. It also describes possible future avenues for
study and application in the area of SDN-based traffic
engineering and access management. The current limitation of
these investigations is the lack of focus on scalability and real-
world implementation, which makes it difficult to actually
apply suggested security solutions in intricate network
systems. Furthermore, there isn't much talk about possible
interoperability issues, resource limitations, and how well
these solutions may change to meet new threats in the
cyberspace. A more thorough examination of these factors
might improve the research findings' relevance and efficacy in
real-world contexts.
Key contributions of the research include:
 Using traffic trends produced by GANs, SDN
controllers can optimize the allocation of network
resources, reducing latency and enhancing QoS.RNN-
based authorization rules continuously identify trends
in network activity, minimising potential
vulnerabilities and adapting to new threats.
 By automating both access control and traffic design,
the method lessens the operational load on the
network's management and promotes more effective
resource utilization.
 Assessing the efficacy of the hybrid GAN-RNN
models in SDN scenarios through multiple simulations
and real-world experiments. The results show notable
gains in network security and effectiveness,
highlighting the approach's potential for contemporary
network management.
 This study offers a ground-breaking framework for
utilizing the powers of hybrid GAN-RNN models to
optimize software-defined networking. The next phase
of network administration is anticipated with the

(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 14, No. 12, 2023
598 | P a g e
www.ijacsa.thesai.org
integration of flexible access controls and realistic
traffic generation.
The Section I provides an overview of the paper. The
Section II reviews existing literature and emphasizes the gap
in addressing techniques for network security enhancement.
Section III defines the central research problem concerning
driver drowsiness detection complexities. Section IV outlines
data collection, preprocessing, feature extraction, and the
integration of Hybrid GAN-RNN. Section V presents
empirical findings, compares classifier performance, and
explores implications and future research directions is in
Section VI, which is solidifying the research's significance in
Network security.
II. RELATED WORKS
Ramprasath and Seethalakshmi [11] examines a crucial
element of SDN, focusing on the requirement for improved
security controls in SDN systems. By separating the
information plane from the control plane, SDN enables on-
demand services and the ability to configure networks
dynamically. The study correctly highlights the fact that,
despite the fact that SDN controls traffic flows and flow labels
based on Open Flow virtual switches successfully, it lacks
built-in security safeguards to combat malicious traffic, such
as Denial-of-Service (DoS) assaults, which can significantly
lower service availability. It is laudable that the paper's main
emphasis is on identifying and reducing DoS threats by
dynamically setting firewalls within SDN setups. The study
makes an effort to close this security gap by using dynamic
access control lists. This study's noteworthy feature is the use
of Mininet to simulate SDN with dynamic access control list
attributes. This enables real-world testing and
experimentation. The practical relevance of the findings is
given more weight by this empirical confirmation. The work
would benefit from a more thorough examination of the exact
methods and tools employed for DoS attack detection and
mitigation inside the SDN environment to further strengthen
its contribution. Readers would have a better grasp of the
suggested strategy if more information was provided about the
dynamic access control list implementation and the standards
for differentiating malicious from normal traffic. In order to
reduce DoS attacks, the article tackles a critical security issue
in SDN systems and offers a potential solution using dynamic
access control lists. The research offers important insights
towards strengthening the security of SDN systems by fusing
theoretical understanding with real-world testing. The study
would be even more helpful and effective if it elaborated on
the technical specifics of the suggested strategy.
Vimal et al. [12] provides a fascinating and current
investigation into how integrating Internet of Things (IoT)
devices might improve the security of Software-Defined
Networks (SDN), with an emphasis on boosting information
access control using encryption. The research's emphasis on
creating a strong infrastructure for IoT devices is well-placed
given the quick spread of IoT devices. The notion of a stability
routing protocol, which evaluates the reliability of devices and
packet flows, is introduced in this work. To create dependable
SDN routes, this method makes use of the mutual trust
between network components, Quality of Service (QoS), and
energy circumstances. An important advancement is the
incorporation of SDN architecture into the Cognitive Protocol
Network (CPN) technology platform to improve energy
efficiency. A novel strategy for tackling security issues is the
use of stochastic neural networks (SNNs) for decentralized
decision-making based on data gleaned from perceptual
packets. It is a praiseworthy effort to include these
components into SerIoT approaches to provide IoT encryption
for information access control. The implementation of various
techniques and technologies, particularly the precise
approaches utilized for IoT encryption and access control,
might need more explicit explanations in the study. The
complexity of the study would also be increased by providing
more detail on how the suggested network infrastructure
solves issues like erratic connectivity, constrained
cryptographic capacity, and energy restrictions. The study
emphasizes the significance of tackling cluster instability for
platform efficiency as well as the necessity of collaboration. A
deeper grasp of the research's practical relevance might be
provided by providing additional information on the
difficulties and potential solutions linked to these issues.
Shin et al. [13] addresses the potential of Software-
Defined Networking to improve network security as it goes
into a crucial and modern junction of technology. Because it
can separate control logic from conventional network
hardware, SDN has attracted a lot of interest as a
transformational technology that can improve network
administration and innovation. The authors note that despite
its capabilities, SDN is still largely disregarded by the security
community, highlighting the underutilized potential of SDN in
the area of network security. The article presents a thorough
overview of the prospects offered by this technology by
meticulously evaluating how the distinctive features and
capabilities of SDN may strengthen network security and the
larger information security process. This in-depth analysis of
SDN's potential to advance network security research creates
fresh directions for future study in this crucial area. The article
does a good job of outlining the main ideas and goals, but it
might make a bigger impact if it went into more detail with
examples or case studies of how SDN has been used to
successfully handle security issues. Giving readers specific
examples of how SDN is used to enhance network security
can help readers understand the real-world applications of the
technology and may encourage more research projects. The
report effectively discusses the important role that SDN may
play in strengthening network security and sheds light on an
exciting yet underappreciated field of study. It is a useful tool
for academics and professionals who want to strengthen
network security in the context of a changing technological
environment by utilizing SDN's capabilities. Extending on
actual use cases and useful implementations might increase
the paper's impact and usefulness.
Ahmad et al. [14] focuses on the use of machine learning
(ML) approaches to thwart Denial of Service (DoS) and
Distributed DoS (DDoS) attacks inside the SDN framework. It
provides a timely analysis of the essential confluence between
Software Defined Networking (SDN) and security. With its
logically centralized control plane, SDN offers enhanced
network administration as a viable response to a number of

(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 14, No. 12, 2023
599 | P a g e
www.ijacsa.thesai.org
issues in conventional networks. But because of the security
flaws introduced by this centralization, SDN control systems
are becoming tempting targets for malicious attacks. Given
ML's shown efficacy in finding security vulnerabilities, the
paper made a sound decision to use ML approaches for
recognizing and mitigating DoS and DDoS attacks. It is a
useful addition to test these ML approaches in practice in an
SDN system, especially by subjecting the SDN controller to
DDoS attacks. It offers useful perceptions on the applicability
and constraints of ML-based security methods for upcoming
communication networks. The work may benefit from a more
in-depth examination of the various ML approaches used and
the standards for judging their efficacy. Readers would
comprehend the use of ML models or algorithms to SDN
security more clearly if examples or case studies of these
applications were given. This article discusses security flaws
resulting from centralized control, a serious issue in the SDN
space. The work offers a significant addition to the area by
outlining and assessing ML strategies to defend against DoS
and DDoS assaults within SDN. It highlights the value of ML-
based solutions in securing upcoming communication
networks and provides a viable path for boosting network
security. The paper's usefulness and effect would be increased
with additional clarification of the ML approaches employed.
Pérez-Díaz et al. [15] provides a significant and pertinent
addition to the continuing problem of LR-DDoS attack
mitigation in the context of Software-Defined Networks
(SDN). Due to the notoriously difficult-to-detect nature of LR-
DDoS assaults and the potential harm they pose in SDN
environments, a flexible modular architecture for their
detection and mitigation has been developed. The study
utilizes Machine Learning models such as J48, Random Tree,
REP Tree, Random Forest, Multi-Layer Perceptron and
Support Vector Machines to train an Intrusion Detection
System (IDS). Despite the inherent difficulties presented by
LR-DDoS assaults, the assessment of these ML models using
the Canadian Institute of Cyber security (CIC) DoS dataset
showed a remarkable detection rate of 95%. One of the key
advantages of this study is the practical implementation of the
open network operating system (ONOS) controller within a
Mini-net virtual machine, which tries to faithfully imitate real
production network circumstances. This strategy strengthens
the paper's credibility and explains how it may be used in real-
world network security settings. The article also emphasizes
that the intrusion prevention detection system successfully
mitigates all assaults identified by the IDS, highlighting the
usefulness of the suggested architecture in LR-DDoS attack
detection and mitigation. This study presents a novel approach
that combines ML approaches with a flexible modular
architecture to solve the ongoing problem of LR-DDoS
assaults in SDN systems. Its potential as a formidable tool for
network security is highlighted by its easy deployment and
remarkable detection results. The impact of the work would be
increased and new insights would be provided for security
practitioners and academics with a more thorough
investigation of the difficulties and model choices.
Latif et al. [16] discusses security, a key concern in the
context of the Industrial Internet of Things environment.
Smart cities, agriculture, and healthcare are just a few areas
where IIoT is crucial due to its integration of sensors, devices,
and databases. This study acknowledges the distinct security
risks that the IIoT presents as a result of its integration into
more complex operational systems. In this research, a unique
method for anticipating and detecting several cybersecurity
attacks—including denial of service, malicious operation,
malicious control, data type probing, espionage, scan, and
incorrect setup—that are frequently seen in IIoT contexts is
presented. It undertakes a comparison analysis with
conventional machine learning methods including artificial
neural networks, support vector machines, and decision trees,
and proposes a lightweight random neural network (RaNN) as
the foundation for its prediction model. The study's main
conclusions show that the suggested RaNN-based model
performs admirably, with accuracy rates of 99.20%, precision,
recall, and the F1 score all above 99%. The model also has
short prediction duration of 34.51 milliseconds. These
findings show how well the RaNN model predicts and
recognises IIoT cybersecurity threats. The work makes an
important addition to the area since it tackles the urgent
demand for reliable security solutions in IIoT. It is an
intriguing option for boosting IIoT security since it uses a
lightweight RaNN model and performs better than
conventional approaches.
The claimed accuracy gains of 5.65% for IoT security over
cutting-edge machine learning algorithms is notable and
demonstrates the practical applicability of this study.
Nevertheless, when using this approach in complex IIoT
contexts, it's critical to take into account potential limits, such
as the range and variety of attack scenarios, and scalability,
including real-world deployment issues. Despite this, the
article offers a useful framework for more study and
advancement in the field of IIoT security, including the
potential for real-world use in securing vital industrial
systems. While the previously discussed works provide
important insights into various aspects of protecting SDN and
addressing cybersecurity concerns within the IIoT framework,
a common shortcoming of these studies is the lack of
comprehensive real-world implementation and evaluation.
Although some studies employ simulation techniques, little is
known regarding whether the proposed solutions can be scaled
and applied in complex, real-world large-scale network
environments. A more thorough analysis of the potential
challenges and setbacks that came across throughout the
development of their safety processes, such as issues with
interoperability, resource constraints, and adaptability to
evolving attack strategies, would also greatly increase the
study's practical significance and utility.
III. PROBLEM STATEMENT
Although SDN presents the possibility of dynamic and
adaptable network management, efficiency optimization and
security remain major obstacles. Lack of integrated security
measures to thwart malicious traffic, particularly Denial-of-
Service (DoS) attacks, is one of the major problems that can
seriously impair service availability. Many times, existing
SDN solutions are unable to adequately handle these security
issues. As such, the development of a comprehensive strategy
that enhances security protocols while simultaneously
optimizing network performance is imperative. The purpose of

(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 14, No. 12, 2023
600 | P a g e
www.ijacsa.thesai.org
this research is to determine how well a hybrid GAN-RNN
approach, which sets firewalls dynamically using dynamic
access control lists, can handle this dual challenge.
Furthermore, it aims to provide a new solution that improves
network safety and efficiency in SDN environments.
The ability of the proposed Hybrid GAN-RNN method to
focus on both security concerns and network efficiency
optimizing in SDN systems is what makes it effective. By
using Generative Adversarial Networks (GANs) to simulate
feasible traffic patterns and Recurrent Neural Networks
(RNNs) for adaptable controls on access, this method offers a
multidimensional solution. GANs can be used to model
various traffic scenarios, and RNNs can be used to flexibly set
access control rules based on real-time threat detection to
achieve optimal network designs. This hybrid paradigm
effectively adapts to changing network conditions and security
threats, significantly enhancing the overall resilience of SDN
systems. Moreover, employing the Mini net for real-world
assessment boosts the practical value of the results and
increases the possibility of their successful implementation in
operational networks. More information about the specific
methods and tools employed for DoS attack detection and
avoidance in an SDN environment is required in order to
improve the strategy's effectiveness.
IV. PROPOSED HYBRID GAN-RNN FOR NETWORK
SECURITY
The proposed methodology represented in Fig. 1 starts
with the collection of network traffic data, which is then
meticulously pre-processed to cleanse, format, and extract
relevant features. A tailored Generative Adversarial Network
(GAN) architecture is crafted to generate synthetic traffic
patterns that closely resemble real network data. These
synthetic patterns are crucial for enhancing security analysis.
Simultaneously, a Recurrent Neural Network (RNN) is
employed to predict network attacks based on the generated
traffic patterns. The RNN learns to recognize temporal
patterns and anomalies in the data, aiding in the proactive
identification of potential security threats. Following the
hybrid GAN-RNN approach, the system's performance is
thoroughly analyzed, assessing its ability to generate realistic
traffic and predict attacks accurately. Additionally, a
comparative evaluation is conducted to benchmark the
proposed methodology against existing approaches, providing
insights into its effectiveness in bolstering network security.
A. Data Collection
The CICIDS2017 dataset provides a valuable resource for
improving network performance and security through the
utilization of Hybrid GAN-RNN models within Software-
Defined Networking (SDN) environments. This dataset
incorporates both benign network traffic and a wide range of
common attacks, making it a suitable foundation for our
research and development in the field of networks security and
optimization. The CICIDS2017 dataset offers a
comprehensive view of network traffic, encompassing benign
background traffic and real-world attack scenarios. The
dataset is constructed with meticulous attention to realism,
featuring the following key components: Generated using the
B-Profile system, the dataset simulates the naturalistic
behaviors of 25 users engaging in various protocols such as
email, HTTP, FTP, HTTPS, and SSH. This component
mimics real-world user interactions, contributing to the
authenticity of the dataset. The dataset represents a complete
network infrastructure, including components like Modem,
Firewall, Switches, Routers, and a diverse array of operating
systems (e.g., Ubuntu, Windows, and Mac OS X). This
realistic topology ensures that the dataset mirrors complex
network environments. The CICIDS2017 dataset incorporates
real attacks from the Attack-Network, enabling researchers to
analyze and develop security measures against a wide range of
threats. This includes the most up-to-date common attacks,
adding relevance to the research context [17].
B. Data Pre-processing using Handling Missing Values
The time series dataset representing network-wide traffic
states, which is denoted as. This dataset encompasses
observations collected over time, each of which corresponds
to a specific time step. The dimensionality of the dataset is
determined by the amount of sensor stations in the network,
denoted as. Mathematical representation of this time series
dataset  as follows in Eq. (1):
 󰇝  󰢹󰇞 (1)
Fig. 1. Proposed workflow.

(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 14, No. 12, 2023
601 | P a g e
www.ijacsa.thesai.org
 Signifies the total number of time steps in our dataset. 
signifies the number of sensor stations distributed across the
network. Each vector󰣐, associated with time step, belongs to
the real numbers󰢩. This vector encapsulates the traffic state
information of the  sensor stations at that specific time [18].
Within this vector, each element 
corresponds to the
traffic speed observed at the d-th sensor station. This research
discusses "traffic state," It specifically focuses on traffic
speed. This definition aligns with the characteristics of the
datasets we employ, particularly those used in our
experimental section. Traffic sensors, such as inductive
looping detector, may encounter failures due to various
reasons, including wire insulations breakdown, damage from
building activities, or electronic unit failures. These sensors
failures result in missing values within our collected data. To
address the issue of missing values, Research employs a
masking vector, which is binary and takes values from the
set {0, 1}, to indicate whether traffic states are missing at a
specific time step t. The masking vectors for  is defined as
follows in Eq. (2):



 (2)
Consequently, for a given traffic state data sample 
in, derivation of a corresponding masking data sample
M, is represented in Eq. (3):
 󰇝  󰢹󰇞 󰢹  (3)
The traffic state prediction problem revolves around the
objective of learning a function󰇛󰇜. This function is designed
to map T historical traffic state datas observations to the
subsequent traffic state data at the next time step. This
problem can be formally described as in Eq. (4):
󰇛󰇟󰢹󰇠󰇟󰢹󰇠󰇜󰇟󰇠
(4)
where,  aims to predict the traffic state at time step  
depend on the historical traffic state data up to time step,
taking into account the masking information to handle missing
values.
C. Hybrid GAN-RNN Architecture for Generating Traffic
Patterns and Access Control
The Hybrid GAN-RNN architecture is designed to
enhance network security by generating realistic network
traffic patterns and making access control decisions based on
those patterns. This architecture comprises two main
components: a Generative Adversarial Network (GAN) for
traffic pattern generation and a Recurrent Neural Network
(RNN) for access control. The hybrid GAN-RNN architecture
is shown in Fig. 2. In the GAN component, the generator (G)
takes random noise (z) as input and generates synthetic traffic
patterns (X_synthetic). The discriminator (D) then evaluates
these synthetic patterns and real traffic patterns (X_real),
aiming to distinguish between them. The objective is to train
the generator to produce traffic patterns that are
indistinguishable from real ones, while the discriminator
becomes more adept at differentiating real from synthetic
patterns. This adversarial training process is guided by a GAN
loss function that encourages the generator to improve its
pattern generation capabilities. The discriminator's formal
objective is to acquire characteristics that maximize the
likelihood of properly categorizing both training and produced
data; the generator's objective is to discover settings  that
minimize  󰇛󰇛󰇜󰇜 the following two-player
minimax game with value functions 󰇛󰇜is therefore
played by the two neural networks.
󰇛󰇜 󰇝󰇟󰇛󰇜󰇠 󰇟󰇛 
󰇛󰇛󰇜󰇜󰇜󰇠󰇞 (5)
where, G(z) is the created false data provided by the noise
vector z, D(G(z)) is the estimated chance of a fake instance
being honest, and D(x) is the estimated likelihood of an actual
model being real generated by the discriminating neural
networks. The generator theoretically learns to produce
genuine samples when it reaches equilibrium.
Fig. 2. Hybrid GAN-RNN architecture.

(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 14, No. 12, 2023
602 | P a g e
www.ijacsa.thesai.org
Algorithm 1: Hybrid GAN-RNN Algorithm for Network Security Enhancement
Input: Network Traffic data
Output: Predicting the attack in Network traffic data
Load input data
  󰇝   󰢹󰇞
// data acquisition
Preprocess network traffic data
Cleanse and Handling missing traffic data’s, and normalize.
//handling missing values
Split the data into training and testing sets.
Generation of Traffic Patterns
// GAN Training
Initialize the GAN model with a generator (G) and discriminator (D).
Train the GAN by iteratively optimizing G and D
Generate synthetic traffic patterns using G
Calculate the GAN loss based on D's ability to distinguish real from synthetic patterns.
Back propagate the loss and update the G and D weights.
Repeat until convergence or a predefined number of epochs
Attack Prediction
//RNN Training
Initialize the RNN model for attack prediction.
Define the RNN architecture, loss function, and optimizer.
Train the RNN using the generated synthetic traffic patterns
Input the synthetic traffic data sequence to the RNN
Calculate the loss based on the predicted attacks and actual labels (ground truth).
Back propagate the loss and update the RNN weights
Repeat until convergence or a predefined number of epochs
Prediction of Attack in Network
//RNN
Evaluate the hybrid GAN-RNN system's performance using testing data
//Performance Evaluation
Measure the accuracy of attack predictions
Calculate other relevant metrics such as precision, recall, and F1-score
Assess the quality of generated traffic patterns
The RNN component processes the generated traffic
patterns (X_synthetic) and performs access control. For this
purpose, the RNN can employ a LSTM architecture, which
allows it to consider temporal dependencies in the traffic data.
The RNN's internal state (ht) evolves as it processes the traffic
patterns, and at each time step, it produces access control
decisions (yt) through a Softmax layer. These decisions can
take various forms, such as binary access control (allows or
deny) or multiclass access policies based on the traffic content
and context.
The key innovation of this architecture lies in its
combination of GAN and RNN components. The GAN
generates synthetic traffic patterns that are realistic and
diverse, reflecting various network activities. The RNN, in
turn, leverages these patterns to make access control decisions
in real-time. This approach enables a more dynamic and
adaptable access control system that can respond effectively to
evolving network conditions and potential security threats.
During training, the entire hybrid architecture is optimized
through a joint loss function that balances the GAN loss and
the access control loss. This ensures that the generated traffic
patterns are not only realistic but also suitable for access
control decision making. The RNN's parameters are fine-tuned
to make accurate access control decisions based on the
generated patterns, thereby enhancing network security.
V. RESULTS AND DISCUSSION
The result section provides a comprehensive evaluation of
the proposed network security enhancement method,
employing various evaluation metrics such as accuracy,
precision, recall, and F1-score. The analysis begins with a
comparison of the method's accuracy on different datasets,
highlighting the notably high accuracy of the "Proposed GAN-
RNN" approach on the CICIDS2017 dataset. A comparative
assessment with existing methods further underscores the
method's superiority, showcasing exceptional precision, recall,
and F1-score. Graphs depict the performance trends,
demonstrating the model's convergence and its ability to
generalize to unseen information. The training and testing
graphs illustrate the model's progression, while the loss graph
reveals its capacity to avoid overfitting. The results validate
the effectiveness of the proposed methods in network intrusion
detection, emphasizing its potential to enhance network
security with impressive accuracy and robustness. The
practical usefulness of the Hybrid GAN-RNN technique
extends to dynamic business networks, allowing for adaptive
routing and resource allocation. Its 99.4% accuracy in
cybersecurity guarantees quick access policy changes, which
are essential for sectors like banking and healthcare and
improve overall network security and efficiency.
A. Evaluation Metrics
Four assessment measures were used in the study to
evaluate the designs: F1-score, accuracy, precision, and recall.
Such specific variables are described as in Eq. (6), (7), (8) and
(9):
  
 (6)
  
 (7)
  
 (8)
  
 (9)
TP is the number of information that, irrespective of every
one of the types of information which was genuinely positive,
was precisely identified as positive. TN is the number of
information that, irrespective of all the results which was truly
negative, were properly identified as negatives. The number of
variables that the equation incorrectly categorized as negative
despite the fact they had been positive in the input data is
represented by the letter FN. The number of values that the
algorithm incorrectly categorized as positive when they had
been negative in the source data is known as false positives, or

(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 14, No. 12, 2023
603 | P a g e
www.ijacsa.thesai.org
FP. The percentage of the number of information that the
algorithm identified as being positive to the number of
positive results that were really present in the collection of
data is known as recall. Precision can be defined as the
proportion of the entire amount of information that the model
properly identified as positive to the number of data which the
algorithm categorized as positive. Lastly, as mentioned in, the
F1-score represents the harmonious average of recall and
precision Top of Form [19].
This Table I present the performance results of different
intrusion detection methods on these datasets, with the
"Proposed GAN-RNN" method achieving notably high
accuracy on the CICIDS2017 dataset at 99.4%. It's important
to note that the choice of dataset and the specific evaluation
metrics used can significantly impact the reported accuracy,
and the effectiveness of a method may vary depending on the
dataset's characteristics and the complexity of the network
security task.
The graph in Fig. 3 depicts, the methods consistently
outperform others across multiple datasets and which ones
may excel in specific contexts. This comparison aids in the
selection of the most robust intrusion detection method for
diverse network security environments, contributing to
informed decision-making in network security strategy.
Table II presents a comparative overview of different
methods applied to network intrusion detection, showcasing
their performance across multiple evaluation metrics. It
includes the methods GRU, CNN, B-GRU, and the "Proposed
GAN-RNN." the "Proposed GAN-RNN" method exhibits
exceptional accuracy, achieving an impressive 99.4%,
surpassing the other methods in the accuracy metric.
Furthermore, it excels in precision with a score of 99.25%,
ensuring a high proportion of correctly classified positive
predictions. It demonstrates remarkable recall at 99.6%,
effectively capturing a significant portion of actual positive
instances. The F1-score, a balanced measure of precision and
recall, remains strong at 99.4%, further affirming the method's
robustness in network intrusion detection, making it a highly
promising approach for bolstering network security.
The Graph represents in Fig. 4 shows the performance
comparison of different intrusion detection methods on
various metrics, including accuracy, precision, recall, and F1-
score.
Fig. 5 represents the training and testing graph for the
proposed network security enhancement method illustrates the
model's performance throughout the training process. During
the training phase, the metrics are plotted as they evolve with
each epoch, showing how the model learns and improves its
performance over time. The testing phase is also depicted on
the same graph, showcasing how the model generalizes to
unseen data. This graph provides a clear visualization of the
model's convergence and its ability to avoid overfitting or
underfitting, thus assisting in the evaluation and refinement of
our network security enhancement approach. The testing
accuracy attained is 99.4%.
Fig. 6 represents the training and testing loss graph is a
graphical representation that illustrates the changes in the loss
function values of a machine learning or deep learning model
during both the training and testing phases. The testing loss
curve reveals the model generalizes to unseen data, and
ideally, it should exhibit a similar decreasing trend, indicating
that the model is not overfitting.
TABLE I. ACCURACY COMPARISON OF DATASET
Dataset
Methods
Accuracy
KDD99 [20]
DT
92.3
UNSW-NB15 [20]
LR
85.56
CICIDS2017
Proposed GAN-RNN
99.4
TABLE II. PERFORMANCE COMPARISON WITH EXISTING METHODS
Methods
Accuracy
Precision
Recall
F1-score
GRU [21]
99
99
99
99
CNN [21]
97.7
97.4
98.2
99
B-GRU [21]
98.74
98.9
99
99
Proposed GAN-RNN
99.4
99.25
99.6
99.4
Fig. 3. Dataset comparison.
75 80 85 90 95 100 105
DT
LR
Proposed GAN-RNN
KDD99 UNSW-
NB15 CICIDS
2017
Percentages
Accuracy

(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 14, No. 12, 2023
604 | P a g e
www.ijacsa.thesai.org
Fig. 4. Evaluation of performance with existing approaches.
Fig. 5. Training and testing accuracy.
Fig. 6. Training and testing loss.
The Fig. 7 displays a Generative Adversarial Network with
Recurrent Neural Network (GAN-RNN) model's Receiver
Operating Characteristic (ROC) curve. The model's ability to
distinguish between classes, especially in issues with binary
classification, is represented graphically by the ROC curve.
The true positive rate (sensitivity) during each threshold,
which ranges from 0 to 0.7, is paired with a corresponding
threshold value in the table. The true positive rate tends to rise
in tandem with the threshold, suggesting that the model is
becoming more accurate at identifying positive instances. The
true positive rate increases gradually from 0.07 to 0.994 to be
the threshold increases, indicating that the GAN-RNN model
has high discriminatory power. This indicates that the model
performs well in classifying positive instances, demonstrating
its efficacy in achieving high sensitivity across a range of
threshold values.
Fig. 7. ROC of GAN-RNN model.
B. Discussion
The accuracy of several methods for identifying intrusions
on a range of datasets is assessed at the outset of the findings
section. The efficiency results for the three different
datasets—KDD99, UNSW-NB15, and CICIDS2017—are
shown in Table I. On the CICIDS2017 dataset, the "Proposed
GAN-RNN" approach notably obtains a very high accuracy
value of 99.4%. This extraordinary accuracy demonstrates
how well the approach works in a specific dataset to recognize
network security problems. However, given that network data
can differ greatly in complexity and feature sets, it is
imperative to recognize that the selection of dataset is a
critical factor in deciding accuracy. As a tool for comparing
datasets, the chart in Fig. 3 shows how the techniques
96
96.5
97
97.5
98
98.5
99
99.5
100
Accuracy Precision Recall F1-score
Percentages
Performance Comparison
GRU
CNN
B-GRU
Proposed GAN-RNN
0
0.2
0.4
0.6
0.8
1
1.2
020 40 60 80 100
Accuracy
Epoch
Train-Test Accuracy
Train
Test
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
010 20 30 40 50
loss
Epoch
Train-Test Loss
Train
Test
0
0.2
0.4
0.6
0.8
1
0 0.2 0.4 0.6 0.8
FPR
TPR
ROC of GAN-RNN Model

(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 14, No. 12, 2023
605 | P a g e
www.ijacsa.thesai.org
continually perform better than others on a variety of datasets.
It makes it possible to pinpoint the approaches that work best
in certain situations. The comparison is essential because it
helps choose the most reliable intrusion detection technique
for various network security scenarios. The graph presents the
overall efficacy of the "Proposed GAN-RNN" approach,
indicating a potential option for improving network security
on a variety of datasets.
Table II provides an extensive comparative analysis of
various intrusion detection techniques, such as GRU, CNN, B-
GRU, and the "Proposed GAN-RNN." The "Proposed GAN-
RNN" approach performs exceptionally well according to a
number of evaluation parameters. It is noteworthy for
achieving a 99.4% accuracy rate, which is higher than the
other approaches. The technique also performs exceptionally
well in terms of precision (99.25%), guaranteeing a large
percentage of accurately identified positive predictions.
Furthermore, it exhibits exceptional memory (99.6%),
successfully catching a substantial proportion of true positive
cases. At 99.4%, the F1-score—a measure that strikes a
compromise between recall and precision—remains robust.
All of these findings support the "Proposed GAN-RNN"
approach's robustness and dependability in detecting network
intrusions. Because of its exceptional performance, the
approach offers both accuracy and precision in spotting
security issues, making it a very viable alternative to
strengthen network security. The suggested network security
enhancement approach is thoroughly evaluated in the results
section, which highlights its remarkable accuracy, precision,
recall, and F1-score. It demonstrates how the approach can
continuously beat other approaches on various datasets, which
makes it a strong option for secure network application. These
results provide insightful information that may be used to
make well-informed decisions about network security
planning and technological implementation. The access
control as well as traffic engineering systems that are now
dependent on network security may not be scalable, may have
trouble adapting to changing threats in real-time, and may find
it difficult to properly handle growing cyber threats [8].
VI. CONCLUSION AND FUTURE WORK
In the framework of SDN-based traffic management and
access control, hybrid GAN-RNN models were proposed and
their efficacy was shown in the present investigation. The
results obtained suggest that this novel technique holds
significant potential for improving software-defined network
security and performance. The Hybrid GAN-RNN design has
demonstrated notable gains in network effectiveness and
threat reduction through the creation of realistic patterns of
traffic and accurate access control choices. In today's intricate
and constantly evolving network systems, the capacity to
detect abnormalities, adjust to changing conditions, and
optimize traffic flows is an essential skill. The suggested
technique's high recall, accuracy, and precision highlight its
potential as a vital resource for network managers and security
experts. This strategy has the ability to enable enterprises to
strengthen their safety posture, maximize resource efficiency,
and handle their networks more effectively as the network
environment changes. Through the integration of Hybrid
GAN-RNN models within SDN, this study considerably
improves knowledge while improving the efficiency and
security of networks. Results validate hypotheses and lay the
groundwork for more investigation into adaptive access
management and optimizing traffic in dynamic contexts in the
future.
Provide means by which access control policies can be
automatically modified in response to threat assessments and
network conditions in real time, enabling more flexible and
responsive security. Instead of depending only on historical
data, investigate real-time analysis abilities that allow the
system to identify and address security threats and operational
issues as they arise. To convert created traffic trends into
useful network configurations and rules, tighten the
connection with SDN controllers. In multi-domain or mixed-
cloud situations, when network intricacy and safety issues are
heightened, expand the Hybrid GAN-RNN technique to
optimize network security and performance. To ensure
resilience in the midst of sophisticated dangers, assess the
proposed method's resistance against adversarial assaults that
attempt to interfere with traffic patterns or evade control of
access. To handle increasing threats, future research might
focus on improving the hybrid GAN-RNN method's
scalability and flexibility. A more thorough and forward-
thinking approach would involve looking into effective ways
of managing larger networks and new security challenges.
REFERENCES
[1] P. Mishra, V. Varadharajan, U. Tupakula, and E. S. Pilli, ―A Detailed
Investigation and Analysis of Using Machine Learning Techniques for
Intrusion Detection,‖ IEEE Commun. Surv. Tutorials, vol. 21, no. 1, pp.
686–728, 2019, doi: 10.1109/COMST.2018.2847722.
[2] V. Kapoor and R. Yadav, ―A Hybrid Cryptography Technique for
Improving Network Security,‖ IJCA, vol. 141, no. 11, pp. 25–30, May
2016, doi: 10.5120/ijca2016909863.
[3] C. Yu, J. Lan, Z. Guo, and Y. Hu, ―DROM: Optimizing the Routing in
Software-Defined Networks With Deep Reinforcement Learning,‖ IEEE
Access, vol. 6, pp. 64533–64539, 2018, doi:
10.1109/ACCESS.2018.2877686.
[4] N. Awadallah Awad, ―Enhancing Network Intrusion Detection Model
Using Machine Learning Algorithms,‖ Computers, Materials &
Continua, vol. 67, no. 1, pp. 979–990, 2021, doi:
10.32604/cmc.2021.014307.
[5] Muthukumaran V., V. V. Kumar, R. B. Joseph, M. Munirathanam, and
B. Jeyakumar, ―Improving Network Security Based on Trust-Aware
Routing Protocols Using Long Short-Term Memory-Queuing Segment-
Routing Algorithms:,‖ International Journal of Information Technology
Project Management, vol. 12, no. 4, pp. 47–60, Oct. 2021, doi:
10.4018/IJITPM.2021100105.
[6] S. Akbar, J. A. Chandulal, K. N. Rao, and G. S. Kumar, ―Improving
network security using machine learning techniques,‖ in 2012 IEEE
International Conference on Computational Intelligence and Computing
Research, Coimbatore, India: IEEE, Dec. 2012, pp. 1–5. doi:
10.1109/ICCIC.2012.6510197.
[7] R. Ahmad, R. Wazirali, and T. Abu-Ain, ―Machine Learning for
Wireless Sensor Networks Security: An Overview of Challenges and
Issues,‖ Sensors, vol. 22, no. 13, p. 4730, Jun. 2022, doi:
10.3390/s22134730.
[8] S. Anbalagan et al., ―Machine-Learning-Based Efficient and Secure
RSU Placement Mechanism for Software-Defined-IoV,‖ IEEE Internet
Things J., vol. 8, no. 18, pp. 13950–13957, Sep. 2021, doi:
10.1109/JIOT.2021.3069642.
[9] S. Nanda, F. Zafari, C. DeCusatis, E. Wedaa, and B. Yang, ―Predicting
network attack patterns in SDN using machine learning approach,‖ in
2016 IEEE Conference on Network Function Virtualization and

(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 14, No. 12, 2023
606 | P a g e
www.ijacsa.thesai.org
Software Defined Networks (NFV-SDN), Palo Alto, CA: IEEE, Nov.
2016, pp. 167–172. doi: 10.1109/NFV-SDN.2016.7919493.
[10] M. A. Alsheikh, S. Lin, D. Niyato, and H.-P. Tan, ―Machine Learning in
Wireless Sensor Networks: Algorithms, Strategies, and Applications,‖
IEEE Commun. Surv. Tutorials, vol. 16, no. 4, pp. 1996–2018, 2014,
doi: 10.1109/COMST.2014.2320099.
[11] J. Ramprasath and V. Seethalakshmi, ―Secure access of resources in
software‐defined networks using dynamic access control list,‖ Int J
Communication, vol. 34, no. 1, p. e4607, Jan. 2021, doi:
10.1002/dac.4607.
[12] V. Vimal et al., ―Enhance Software-Defined Network Security with IoT
for Strengthen the Encryption of Information Access Control,‖
Computational Intelligence and Neuroscience, vol. 2022, pp. 1–10, Oct.
2022, doi: 10.1155/2022/4437507.
[13] S. Shin, L. Xu, S. Hong, and G. Gu, ―Enhancing Network Security
through Software Defined Networking (SDN)‖.
[14] A. Ahmad, E. Harjula, M. Ylianttila, and I. Ahmad, ―Evaluation of
Machine Learning Techniques for Security in SDN,‖ in 2020 IEEE
Globecom Workshops (GC Wkshps, Taipei, Taiwan: IEEE, 2020, pp. 1–
6. doi: 10.1109/GCWkshps50303.2020.9367477.
[15] J. A. Pérez-Díaz, I. A. Valdovinos, K.-K. R. Choo, and D. Zhu, ―A
Flexible SDN-Based Architecture for Identifying and Mitigating Low-
Rate DDoS Attacks Using Machine Learning,‖ IEEE Access, vol. 8, pp.
155859–155872, 2020, doi: 10.1109/ACCESS.2020.3019330.
[16] S. Latif, Z. Zou, Z. Idrees, and J. Ahmad, ―A Novel Attack Detection
Scheme for the Industrial Internet of Things Using a Lightweight
Random Neural Network,‖ IEEE Access, vol. 8, pp. 89337–89350,
2020, doi: 10.1109/ACCESS.2020.2994079.
[17] ―CICIDS2017.‖ https://www.kaggle.com/datasets/cicdataset/cicids2017
(accessed Sep. 18, 2023).
[18] Z. Cui, R. Ke, Z. Pu, and Y. Wang, ―Stacked Bidirectional and
Unidirectional LSTM Recurrent Neural Network for Forecasting
Network-wide Traffic State with Missing Values.‖ arXiv, May 23, 2020.
Accessed: Sep. 19, 2023. [Online]. Available:
http://arxiv.org/abs/2005.11627.
[19] B. Jang, M. Kim, G. Harerimana, S. Kang, and J. W. Kim, ―Bi-LSTM
Model to Increase Accuracy in Text Classification: Combining
Word2vec CNN and Attention Mechanism,‖ Applied Sciences, vol. 10,
no. 17, p. 5841, Aug. 2020, doi: 10.3390/app10175841.
[20] N. Moustafa and J. Slay, ―The evaluation of Network Anomaly
Detection Systems: Statistical analysis of the UNSW-NB15 data set and
the comparison with the KDD99 data set,‖ Information Security Journal:
A Global Perspective, vol. 25, no. 1–3, pp. 18–31, Apr. 2016, doi:
10.1080/19393555.2015.1125974.
[21] H. Wang and W. Li, ―DDosTC: A Transformer-Based Network Attack
Detection Hybrid Mechanism in SDN,‖ Sensors, vol. 21, no. 15, p. 5047,
Jul. 2021, doi: 10.3390/s21155047.