Paper 39-Machine Learning Based Secure 5G Network Slicing

(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 14, No. 12, 2023
Machine Learning-based Secure 5G Network Slicing:

A Systematic Literature Review
Meshari Huwaytim Alanazi
Department of Computer Science, Northern Border University, Arar, Saudi Arabia
Abstract—As the fifth-generation (5G) wireless networks physical infrastructure. This technology empowers the
continue to advance, the concept of network slicing has gained customization of each network slice to cater to the unique
significant attention for enabling the provisioning of diverse demands of diverse applications and services. However, this
services tailored to specific application requirements. However, also introduces new security challenges, including
the security concerns associated with network slicing pose unauthorized access, data breaches, and denial of service
significant challenges that demand comprehensive exploration attacks [2]. Conventional security measures such as firewalls,
and analysis. In this paper, we present a systematic literature intrusion detection systems (IDS), and access control
review that critically examines the existing body of research on mechanisms may prove inadequate in effectively mitigating
machine learning techniques for securing 5G network slicing.
these threats.
Through an extensive analysis of a wide range of scholarly
articles selected from specific search databases, we identify and Given the critical importance of 5G networks and the
classify the key machine learning approaches proposed for growing security risks, it is imperative to develop effective
enhancing the security of network slicing in the 5G environment. security mechanisms for network slicing. Machine learning-
We investigate these techniques based on their effectiveness in based intrusion detection systems (IDS) have emerged as a
addressing various security threats and vulnerabilities while promising approach to enhance the security of 5G networks.
considering factors such as accuracy, scalability, and efficiency. These systems analyze network traffic, behavior patterns, and
Our review reveals that machine learning techniques, including
anomalies in real-time to detect and respond to security threats
deep learning algorithms, have been proposed for anomaly
promptly [3]. Nevertheless, despite the potential that machine
detection, intrusion detection, and authentication in 5G network
slicing. However, we observe that these techniques face
learning-based solutions hold for enhancing the security of 5G
challenges related to accuracy under dynamic and heterogeneous network slicing, there remains a dearth of comprehensive and
network conditions, scalability when dealing with a large number methodical research in this particular domain. Therefore, the
of network slices, and efficiency in terms of computational objective of this study is to conduct a thorough literature
complexity and resource utilization. To overcome these review of the current state-of-the-art in this field. The findings
challenges, our experimentation shows that the integration of of this study will provide valuable insights to researchers and
reinforcement learning techniques with CNNs, multi-agent practitioners, helping them understand the current research
reinforcement learning, and distributed SVM frameworks trends, identify research gaps, and develop effective security
emerged as potential solutions with improved accuracy and solutions for 5G networks. Ultimately, this research endeavor
scalability in network slicing. Furthermore, we identify aims to contribute to the development of secure and reliable
promising research directions, including the exploration of 5G networks, which are crucial for the success of various
hybrid machine learning models, the adoption of explainable AI industries and the digital economy.
techniques, and the investigation of privacy-preserving
mechanisms. This SLR is a significant contribution in the area of
machine learning-based security mechanisms for 5G network
Keywords—5G; accuracy; deep learning; efficiency; security; slicing. The focus of this review is to analyze the security
machine learning; network slicing; scalability challenges through advanced machine learning and deep
learning techniques, evaluating their effectiveness in the
I. INTRODUCTION context of 5G networks slicing security, and identifying
5G networks represent the fifth generation of mobile potential solutions for scalability and efficiency issues. This
communication networks and offer advanced features such as study analyzes various machine learning techniques for
high-speed connectivity, ultra-low latency, and extensive securing 5G network slicing in the subsequent sections
machine-type communication capabilities. These networks beginning with the background in Section II. Section III
have become crucial infrastructure for various industries due details the concept of network slicing in 5G networks,
to the increasing demand for high-speed data transmission and explaining its implementation, business model, and
real-time applications like autonomous vehicles, smart cities, significance. Section IV focuses on the security aspects of
and Industry 4.0. However, ensuring the security of 5G network slicing in 5G networks, addressing various security
networks is a significant concern due to the utilization of new issues and challenges. Section V describes the methodology
technologies and protocols, the complexity of the network used for the systematic literature review, including data
architecture, and the potential for new attack vectors [1]. sources, search strategies, and article selection processes.
Network slicing is a crucial element of 5G networks, enabling Section VI presents the analysis of the findings from the
the establishment of multiple virtual networks on a shared literature review, discussing the advanced machine learning
377 | P a g e
www.ijacsa.thesai.org
Vol. 14, No. 12, 2023
techniques employed for security in 5G network slicing, and efforts related to 5G [4] [5]. Notably, AT&T, headquartered in
highlighting the challenges and potential solutions. Section Dallas, covers approximately 16% of the United States, while
VII concludes and summarizes the key findings of the study Verizon, based in New Jersey, has successfully expanded its
and suggests directions for future research. ultra wideband network to 31 states [6]. Qualcomm predicts
that by 2035, 5G will generate a staggering USD 13 trillion in
II. BACKGROUND value for goods and service industries worldwide [7]. Given
The swift progression of technological innovations such as the connectivity capabilities of 5G, which facilitate faster data
the Internet of Things (IoT), augmented reality (AR), and rates and connect millions of devices, transitioning to 5G is
communication systems such as vehicle-to-vehicle (V2V) and imperative to meet market demands and expectations.
vehicle-to-everything (V2X) have created a pressing demand Numerous ongoing initiatives are harnessing the potential of
for substantial enhancements in network and communication 5G technology to transform various industries including
infrastructure [1][4]. As a response, 5G networks have gained robotics, healthcare, automotive, agriculture, mining, media,
prominence in meeting the growing consumer demands. The and fashion. These projects encompass applications such as
introduction of 5G technology has not only opened up untethered industrial robots, robotic systems for agricultural
opportunities for innovation but has also provided enhanced purposes, AI-assisted medical diagnosis, virtual reality (VR)
reliability for both service providers and consumers, resulting for palliative care, telesurgery enabling virtual patient
in a shift towards virtualization and widespread adoption of operations, and augmented reality (AR) smart glasses for
5G. The benefits of 5G technology include exceptional data enhanced safety [8].
rates that are 10 to 100 times faster, widespread coverage, Delivering the aforementioned services on conventional
heightened reliability, minimal latency, enhanced quality of 4G or other legacy networks presents significant challenges.
service (QoS), and cost-efficient service offerings. With the To overcome these challenges and provide network services
continuous expansion of these services and opportunities, efficiently with limited resources and minimal costs for
service providers and network operators are engaged in fierce network service providers, network slicing has emerged as a
competition to deploy 5G networks and implement network promising solution. Network slicing is a key feature of 5G
slicing within the physical network. technology that involves partitioning the physical network into
The 5G technology consists of three distinct services, multiple logical networks, each capable of delivering
namely enhanced mobile broadband (eMBB), massive customized services based on specific applications and their
machine type communication (mMTC), and ultra-reliable low- requirements [9]. By leveraging the progress of virtualization
latency communication (URLLC).eMBB offers peak data in cloud computing, the physical network resources are
rates ranging from 10 to 100 Gbps and achieves high mobility partitioned into numerous logical or virtual networks known
support up to 500 km/h while ensuring reduced power as "slices" in the 5G context. Each network slice functions as
consumption through the utilization of both macro and small an autonomous virtual network with dedicated resources,
cells. mMTC offers long-range connectivity with minimal data traffic flows, security measures, topology, and clearly defined
rates spanning from 1 to 100 Kbps, facilitating cost-effective quality of service (QoS) parameters. These slices are isolated
machine-to-machine (M2M) communication. Conversely, from each other and serve the distinct service requirements of
URLLC provides highly responsive connections across individual subscribers. [10]. Network slicing offers flexibility
multiple devices, achieving less than 1 ms latency and an end- and scalability by allowing various services to coexist on a
to-end latency of 5 ms between mobile devices and base shared physical network. It could adapt to evolving subscriber
stations. URLLC also ensures moderate data rates ranging needs, facilitate seamless end-to-end communication, support
from approximately 50 Kbps to 10 Mbps, accompanied by an a multi-service environment, provide on-demand network
exceptionally high service availability of 99.9999%, services, and incorporate multi-tenancy capabilities within the
establishing it as an exceptionally dependable service. [2]. The 5G ecosystem. [11][12].
deployment of 5G networks is poised to act as a catalyst for
market expansion. As of 2020, there were already 92 III. NETWORK SLICING
commercial networks operating across 38 countries, with The advancement and swift development of wireless
China accounting for 150 million 5G subscribers and South communication systems have created a need for a wide array
Korea having eight million. Projections from Ericsson suggest of services, applications, and scenarios tailored to meet the
that the United States alone will witness a subscriber base of specific demands of enhanced mobile broadband (eMBB),
320 million by 2025 [11]. The emergence of 5G technology ultra-reliable and low-latency communication (uRLLC), and
compels communication service providers (CSPs) to go massive machine type communication (mMTC).For instance,
beyond their conventional subscriber-centric business models eMBB applications, such as virtual reality and video
and position themselves as digital service providers (DSPs). streaming, demand high throughput, while uRLLC services,
This transformation enables them to fuel innovation, enhance including autonomous driving, require low latency and
safety, and drive productivity on a global level. Recognizing minimal errors. mMTC services, catering to sensing and
the transformative potential of 5G, the World Economic monitoring applications, call for high connectivity. However,
Forum identifies it as a driving force behind the fourth the existing network architecture is inadequate to meet the
industrial revolution. Numerous multinational companies diverse needs of these services.
including Huawei, Samsung, Qualcomm, LG, Ericsson, ZTE
Corp, Nokia, AT&T, NEC Corp, Cisco Systems, Verizon, and In response to this obstacle, 5G networks employ network
slicing, an approach that enables the provisioning of
Orange are actively engaged in research and development
378 | P a g e
Vol. 14, No. 12, 2023
customized services with distinct requirements over a unified slicing tenant, network slicing repository, slice border control,
network infrastructure. Network slicing includes the slice selection function, infrastructure owner, infrastructure
partitioning of the network into distinct slices, including slice, infrastructure slice provider, and infrastructure slice
access, transport, and core network slices. The network slicing tenant [20][21]. Each of these elements has specific roles and
framework is depicted in Fig. 1, illustrating this concept [14]. capabilities in managing the life cycle of a network slice. As
The core network slice consists of both the control plane and an example, the NSI denotes a group of comprehensive logical
user plane, supporting shared or dedicated functions like networks that provide various services tailored to specific
session management, mobility management, user plane, and requirements, encompassing multiple sub-slice instances.
policy control for various slices. Notably, industry players like Conversely, the NSSI represents the localized logical network
Ericsson and Nokia have developed their own network slicing within a network slice, which can be shared among multiple
systems tailored to their respective sectors [19]. In the current NSIs. Logical networks are virtual instances of network
wireless communication environment, a wide range of functions established on a single physical network.
services with different requirements in terms of security,
reliability, data rate, latency, resources, and cost have
emerged. Network slicing has emerged as a solution to enable
resource sharing among services, customers, and providers. It
requires the establishment of multiple logical networks on a
shared physical infrastructure, enabling the provision of
services with distinct characteristics that can concurrently
support multiple technologies. Each service is allocated
dedicated resources aligned with its specific requirements,
thereby enhancing overall network performance. The
management of network slicing involves coordinating virtual
and physical resource management components, including
Network Function Virtualization (NFV), Software-Defined
Networking (SDN) controllers, and orchestrators [15]. NFVs Fig. 1. Network slicing for 5g networks, adopted from [13].
(Network Function Virtualization), which refer to cloud-based
functions, specify the requirements and attributes of network The network slicing manager plays a pivotal role in
slices. Meanwhile, SDN (Software-Defined Networking) overseeing the complete lifecycle of each slice or sub-slice by
controllers establish instances of network slices by connecting performing various management functions. These include the
virtual functions through SDN networks. [16][17]. communication service management function (CSMF),
Orchestrators automate the management and configuration of network slice management function (NSMF), and network
resources across different domains and slices, simplifying the slice subnet management function (NSSMF). CSMFs are
process of creating, deploying, and monitoring services in an responsible for managing, communicating, and updating the
automated manner. Fundamentally, network slicing entails the requirements of the slice to support service requests through
establishment of logical networks on top of shared physical the communication service manager. NSMFs handle the
infrastructures, partitioning them into multiple virtual management of NSIs based on the notifications received from
networks with independent control and management CSMFs, while NSSMFs manage the NSSIs according to the
capabilities. Two types of orchestrators, Service Orchestrators requirements specified by NSMFs. The network slicing
(SOs) and Resource Orchestrators (ROs), play a crucial role in requirements encompass various aspects such as network type,
creating and managing multiple services. Several open-source network capacity, quality of service (QoS), latency, security
network slicing orchestrators have been developed, such as level, device count, and throughput. The resource slice refers
openMANO, OSM, openNFV, openFV, openBaton, ZooM, to the combination of physical and virtual resources necessary
ONAP, SliMANO, OpenBaton, cloudNFV, JOX, FlexRAN for the functioning of network slices. The network slicing
and Cloudify, to efficiently manage resources in access, core, provider is the entity responsible for owning the physical
and transport networks [15][18]. Notably, Huawei has infrastructure where multiple slices are created, whereas the
developed an end-to-end (E2E) network slicing orchestrator network slicing tenant refers to the users of the NSI who
that effectively allocates resources across the three network deliver specific services requested by customers. The
domains. Its performance has been validated through real infrastructure owner denotes the entity that owns the physical
hardware implementation, demonstrating reliable resource infrastructure, On the other hand, the infrastructure slice
isolation per slice [19]. provider is the entity that owns the infrastructure and leases it
to host a variety of services via network slicing. The
The business model related to network slicing comprises infrastructure slice tenants are the users of the infrastructure
several integral components and entities vital to its slice itself [13].
functioning. These include the network slicing instance (NSI),
network slicing subnet instance (NSSI), logical network, IV. SECURITY IN NETWORK SLICING
network subslice (NSS), network slicing template (NST), Ensuring security is of utmost importance during the
network segment, Network Function Virtualization (NFV), implementation of network slicing, which enables the support
Software-Defined Networking (SDN), network slicing of diverse services with varying security requirements [22].
manager, communication service manager, resource slice, The utilization of network slicing in multi-domain
network slicing provider, network slicing terminal, network infrastructures, serving multiple customers, can introduce
379 | P a g e
Vol. 14, No. 12, 2023
complex security challenges. This is especially evident when The significance of identifying isolation characteristics,
resources are shared among slices that adhere to distinct implementing an abstraction layer for achieving end-to-end
security policies established by various verticals and isolation at an appropriate level, and establishing appropriate
operators. To effectively address security issues both within security policies was emphasized in a study [28]. This survey
and among slices, it is crucial to consider security highlighted the lack of a standardized description for isolation
coordination and protocols during the resource allocation and capabilities that can be employed for automated deployment.
design stages. Neglecting these aspects may result in the Hence, it is crucial to define the desired initial level of
emergence of new and advanced security vulnerabilities in 5G isolation, especially concerning security, and devise dynamic
systems and beyond [23]. Each slice is created with isolation isolation mechanisms capable of enforcing the required level
constraints in order to prevent the propagation of attack of isolation for each specific service. To tackle the security
impacts across slices and allow for independent security concerns associated with network slicing in the context of 5G,
solutions [24]. It is essential to ensure adherence to several organizations, such as the Next Generation Mobile
fundamental security principles such as confidentiality, Networks (NGMN), have released guidelines [29]. These
authentication, availability, integrity, and authorization within recommendations assist in identifying potential security risks
each slice [25]. Confidentiality safeguards against in the general packet core. Additionally, technology
unauthorized data disclosure, authentication verifies the guidelines, such as ETSI's recommendations for Network
identities of parties involved in interactions, availability Function Virtualization (NFV), provide further guidance in
ensures the accessibility of slices and applications, integrity addressing security concerns. [30], have been taken into
guarantees that slice owners maintain control over account. ETSI's guidelines encompass security considerations
functionalities and configurations, and authorization throughout the lifecycle of virtual network functions.
determines the permissible capabilities for each network
element. Availability refers to the ability of the system to meet B. More Security Challenges to Network Slicing
the requirements of service level agreements by ensuring that Network slicing implementation in the
slices and applications can be accessed as needed, while telecommunications industry presents challenges, particularly
Network Slice Manager (NSM) and Network Functions (NFs) concerning security and the deployment of the radio access
remain consistently accessible. On the other hand, integrity network (RAN). These challenges have been extensively
ensures that only slice owners possess the authority to modify discussed by researchers like Kotulski et al. [32]. Security
or replace the functionality and configuration of their concerns arise with the introduction of network slicing. One
respective slices [26]. Authorization defines the permitted significant risk is the potential for denial-of-service (DoS)
functionalities for each network element, where slice owners attacks and resource depletion [32], which can disrupt network
are responsible for managing and controlling their respective availability and performance. Other threats include
slices, end-users engage exclusively with authorized slices, monitoring, traffic injection, and impersonation attacks [27],
infrastructure providers oversee the network slice management jeopardizing data integrity and network efficiency. Moreover,
(NSM), NSM governs the network slice instances (NSIs) and the use of diverse systems from different vendors in network
network functions, and network functions exercise control slicing creates a security vulnerability due to the absence of
over resources. These elements consist of slice owners, end- standardized security measures [32]. Exploiting vulnerabilities
users, service providers, infrastructure providers, network in these systems, attackers can successfully target network
slice management (NSM), and Network Function slices. Additionally, specific weaknesses in the isolation and
Virtualization (NFV). The security of network slicing requires protection mechanisms designed for network slices may
each slice and its owners to independently fulfill these enable unauthorized access to resources and sensitive data. To
requirements in order to mitigate the potential exploitation of address these security threats and vulnerabilities, robust
network slicing features by attackers, which could lead to security measures are necessary. These can involve
system failures [13]. implementing intrusion detection and prevention systems,
enforcing access control policies, and establishing
A. Security Issues Introduced by Network Slicing comprehensive security monitoring. Furthermore, integrating
Network Slicing is distinguished by its crucial attribute of security considerations into the design of network slicing
isolation, which has a direct impact on the dependability of the systems and protocols ensures a secure-by-default approach.
slicing solution. Attaining a high level of isolation is essential Standardized security practices enhance overall network
to achieve optimal outcomes. Merely supporting a single slice slicing security and promote interoperability among different
in a slicing system would essentially replicate a conventional vendor systems [31] [32]. The physical realization of the RAN
non-sliced network, which is already extensively studied. poses challenges in implementing network slicing. Ensuring
Hence, the coexistence of multiple slices becomes a necessary resource, traffic, and user isolation within radio network
prerequisite for network slicing, wherein these slices share the elements is a significant obstacle. One suggested approach
same underlying infrastructure. The ability to coexist without involves the utilization of millimeter waves for covering small
interference hinges on establishing the minimum requirements cells, leveraging the cell size to achieve isolation.
for each slice. By meeting these requirements, interference can Nevertheless, current technologies like cognitive radio and
be avoided, thereby ensuring effective isolation. Ensuring non-orthogonal multiple access fall short in delivering the
security in network slicing involves precisely defining the necessary levels of isolation and slicing required to meet the
boundaries of interference for each slice, specifying the desired standards. Thus, further research and development are
minimum requirements, and enforcing compliance with these essential to overcome the physical implementation challenges
requirements [27]. of network slicing in the RAN [31].
380 | P a g e
Vol. 14, No. 12, 2023
The challenges pertaining to security and access TABLE I. RESEARCH QUESTIONS AND MOTIVATION
management in network slicing originate from the varying
security and privacy requirements [27] of different network Question Motivation
slices in 5G networks. Network slicing is dependent on the What are the most
It aims to identify and analyze existing
implementation of software-defined networking (SDN) and advanced machine
machine learning-based techniques for
network function virtualization (NFV). [33], which virtualize learning techniques
RQ1 securing 5G network slicing to develop
network functions as software components. However, currently employed for
more effective and efficient security
ensuring the security of
ensuring secure inter-slice access and end-to-end network 5G network slicing?
solutions.
slicing security becomes more complex in 5G due to its What are the main
comprehensive slicing approach. Securing the management of It aims to identify the potential issues
obstacles and
that may arise when implementing
network slicing encompasses various tasks, including constraints faced by
machine learning-based techniques for
establishing secure access between the radio access network RQ2 these techniques in
securing 5G network slicing, and to
terms of accuracy,
and core network resources, ensuring secure connections scalability, and
develop strategies to overcome these
between user equipment (UE) and network slice instances, and challenges and limitations.
efficiency?
effectively managing shared resources among different It aims to identify and propose possible
network slices. Neglecting these challenges can lead to What are the potential
solutions/recommendations to overcome
solutions to address
security risks, including compromised controllers or RQ3
these challenges and
the challenges and limitations of
orchestrators, isolation failures, insider threats, compromised machine learning-based techniques for
limitations?
NFV instances, and unauthorized data access [34][35]. The securing 5G network slicing.
centralized slice manager also presents security It aims to explore the integration of
considerations, including the security of network slice explainable AI and federated learning,
What are the future
templates, concerns regarding unauthorized access, and issues investigate the transferability of models
research directions and
RQ4 across different network architectures
related to trust [36]. Additionally, in multi-domain opportunities in this
and scenarios, and design new evaluation
infrastructures or multi-tenant environments, network slicing area?
metrics and methodologies for machine
may encounter additional security and privacy challenges. To learning-based solutions.
mitigate these concerns, it is essential to adhere to established
security principles, including but not limited to confidentiality, A. Data Sources
integrity, authenticity, availability, and authorization [37][38]. Table II displays the reputable publishers such as IEEE,
In the context of network function virtualization (NFV), Science Direct, Springer, ACM Digital Library, Wiley, Sage,
maintaining confidentiality is vital to mitigate potential MDPI and Google Scholar, from which the articles were
threats. For example, a compromised slice manager can lead selected for review.
to unauthorized monitoring of traffic through both northbound
and southbound interfaces, potentially exposing sensitive slice TABLE II. DATABASE SOURCES
configurations. Similarly, vulnerabilities in the application
programming interface (API) used during configuration can Publisher URL
enable malicious actors to interfere with slice installation, IEEE https://www.ieee.org
configuration, or activation. Thus, ensuring the confidentiality Science Direct https://www.sciencedirect.com
of inter-slice communications is crucial for establishing a Springer https://link.springer.com
secure environment [39][40][41]. Proactive measures like ACM Digital Library https://www.acm.org
implementing secure communication protocols, enforcing Wiley https://onlinelibrary.wiley.com
access controls, and conducting regular security policy Sage https://journals.sagepub.com
reviews are necessary to mitigate these threats effectively. MDPI https://www.mdpi.com
Google Scholar https://scholar.google.com/
V. METHODOLOGY
This section provides an overview of the methodology B. Search Strategy
employed to conduct a Systematic Literature Review (SLR) Due to limited research in this area initially, articles
on the subject of Machine Learning-based secure 5G network considered for review in this study were limited to those
slicing, following the recommended guidelines outlined in published from the year 2018 onwards. The initial stage in
[42]. The process of formulating research questions is constructing the search query involved identifying appropriate
discussed, along with the motivating factors behind these keywords aligned with the theme and the research questions
questions. Various data sources were used to select relevant put forth. Primary keywords including "network slicing,"
articles, and a specific search strategy was employed to obtain "5G," "security," and "machine learning" were identified, and
articles in the domain. Inclusion and exclusion criteria were logical operators such as "AND" and "OR" were used to link
applied to select articles for review. In order to present a these keywords appropriately. After conducting several tests,
comprehensive overview of the current state-of-the-art in the researchers arrived at a search string that produced a
Machine Learning-based 5G network slicing, Table I sufficient number of related research articles. The keywords
illustrates the research questions along with their respective used in the search string are listed in Table III.
motivations.
381 | P a g e
Vol. 14, No. 12, 2023
TABLE III. SEARCH STRING the article selection process [43] to ensure a systematic and
transparent approach to article selection, as illustrated in Fig.
String Batch1 Batch2 Batch3 Batch4 2. Once the pertinent literature was identified, a
Network Machine comprehensive review of load balancing techniques in
String1 5G Security
Slicing Learning software-defined networks was performed. The review
Network- Machine- process concluded with a meticulous categorization of the load
String2
Slicing Learning balancing techniques to ensure thoroughness. Many articles
String3 ML were excluded during the screening process due to their titles
not meeting the inclusion criteria or abstracts not being
C. Article Selection Process relevant for the survey.
The methodology employed for selecting articles
commenced with formulating research questions that guided Search String: (([Batch1, String1] OR [Batch1, String2])
the creation of a search query for article retrieval. Only AND ([Batch2, String1]) AND ([Batch3, String1]) AND
articles published in the English language were taken into ([Batch4, String1] OR [Batch4, String2] OR [Batch4,
account. The PRISMA flow diagram was utilized to visualize String3])).
Fig. 2. PRISMA flow diagram.
382 | P a g e
Vol. 14, No. 12, 2023
D. Data Extraction findings, and discussions presented. The aim of this evaluation
The research methodology utilized in this study was to identify articles that provide valuable insights and
encompassed an extensive exploration to collect relevant contribute directly to the research topic. Following this
literature pertaining to the subject of secure 5G network rigorous evaluation process, a final set of 18 articles was
slicing utilizing machine learning. The primary objective was identified as the most relevant to the research topic. The
to identify and select articles that provide valuable insights selection of these 18 essential research articles was conducted
into this field of study. The research process consisted of meticulously, taking into account the alignment between the
several distinct phases, including a targeted search, application titles, abstracts, and comprehensive content of the articles. The
of inclusion and exclusion criteria, and comprehensive selected articles were recognized as valuable sources of
evaluation of selected articles. The initial step in the research insights into machine learning-based secure 5G network
process involved performing a targeted search using specific slicing, and their inclusion in the study was deemed essential
keywords. This search was conducted across reputable for the progression of the current research endeavor.
publishers and covered the period between 2018 and 2023.
TABLE IV. INCLUSION AND EXCLUSION CRITERIA
Fig. 3 shows the year-wise number of articles that were
published by these publishers. The list of sources used to Inclusion Exclusion
perform this survey is provided in Table II. The aim was to
The study focuses on Machine The study that focuses on areas other
gather a comprehensive collection of articles related to the Learning-based secure 5G network than Machine Learning-based secure
subject matter. As a result of this search, a total of 241 articles slicing 5G network slicing
were identified and retrieved.
Only the articles written in English Articles written in non-English
language are considered language are not considered
Unpublished articles and those that
Articles published by the publishers
are not peer-reviewed, are not
listed in Table II
considered
Articles published in well-reputed White papers, editorials, keynote
and high impact factor journals are speeches, and articles from predatory
considered journals are not considered
VI. DISCUSSION
In this systematic literature review, we investigated the
current state of research on machine learning-based solutions
for secure 5G network slicing. Our review identified a total of
18 relevant articles that met our inclusion criteria. From our
analysis of these articles, several key themes emerged. Firstly,
machine learning techniques are widely used for various
security-related tasks in network slicing, including anomaly
Fig. 3. Year-wise categorization of articles. detection, intrusion detection, and malware detection.
Secondly, while there is a broad consensus on the potential
E. Inclusion and Exclusion Criteria benefits of machine learning-based solutions for securing 5G
To refine the selection and focus on significant research, network slicing, there is also a lack of standardization and
inclusion and exclusion criteria were established. These interoperability among different solutions. Finally, the use of
criteria were specifically designed to guarantee the inclusion machine learning in network slicing introduces several
of only articles that directly relate to the research topic. Table challenges related to data privacy, explainability, and
IV presents a comprehensive breakdown of the inclusion and scalability. In this discussion section, we will elaborate on
exclusion criteria implemented during this process. By these themes while answering the research questions and
applying these criteria, the initial pool of 241 articles was provide recommendations for future research in this area.
reduced to 170, thereby eliminating articles that did not meet Q1. What is the most advanced machine learning
the predefined criteria. The subsequent phase of the research techniques currently employed for ensuring the security of 5G
involved a detailed examination of the remaining 170 articles. network slicing?
During this phase, the titles and abstracts of the articles
underwent a thorough examination to further refine the Network slicing is a vital component of 5G and future
selection process. The purpose of this review was to identify generation networks since it enables the creation of
articles that aligned closely with the research focus on customized logical networks with diverse functionalities,
machine learning-based secure 5G network slicing. As a result dependability, and security properties. Scholars have proposed
of this review, 32 articles were selected for further evaluation. multiple types of Machine Learning (ML) and Deep Learning
(DL) approaches to increase the security and reliability of
The selected 32 articles underwent a comprehensive network slicing. Table V summarizes the state-of-the-art
evaluation based on their content to ensure a close match with ML/DL algorithms used in the scientific literature for secure
the objectives and scope of the current research. This 5G network slicing. This discussion examines the strategies
evaluation involved an in-depth analysis of the full texts of the for secure 5G network slicing presented in the selected
articles, including an examination of the methodology, articles. Fig. 4 illustrates the performance of various machine
383 | P a g e
Vol. 14, No. 12, 2023
learning algorithms concerning the security of 5G network chart clearly distinguishes the strengths and weaknesses of
slicing. Each algorithm is evaluated based on three key different algorithms. For instance, algorithms like RL
criteria: accuracy, scalability, and efficiency. Accuracy (Reinforcement Learning) and DQN (Deep Q-Network)
reflects the algorithm's ability to provide precise and reliable demonstrate high scores in all three categories, making them
security measures. Scalability measures its capacity to handle robust choices for secure 5G network slicing. On the other
an increasing number of network slices efficiently, accounting hand, algorithms like LSTM (Long Short-Term Memory) and
for complex dependencies. Efficiency assesses how well the DBN (Deep Belief Networks) show comparatively lower
algorithm utilizes resources during security processes. The scores, suggesting room for improvement in their application.
TABLE V. STATE-OF-THE-ART ML TECHNIQUES, CHALLENGES AND POTENTIAL SOLUTIONS
Challenges
Ref. Algorithm Potential Solutions
Accuracy Scalability Efficiency
Lack of accuracy in Integrating reinforcement learning techniques with
network slice Scalability CNNs can address accuracy, scalability, and
CNN
orchestration and challenges in Inefficient resource resource utilization challenges in secure 5G
[45][49][51][52 (Convolutional
optimization and deep securing multiple utilization during network slicing, including network slice
][61] Neural
learning-based DDoS network slices attack mitigation orchestration, DDoS attack detection, and
Networks)
attack detection for simultaneously simultaneous protection of multiple network
network slicing. slices.
Difficulty in
Multi-agent RL for accurate, scalable, and
RL Limited accuracy in scaling up Inefficient resource
[44][55][56][62 efficient secure 5G network slicing for intelligent
(Reinforcement intelligent resource network slices due utilization during IP
] resource allocation, complex dependencies, and IP
Learning) allocation to complex shuffling processes
shuffling processes.
dependencies
Scalability
Challenges in Distributed and scalable SVM frameworks can
SVM (Support challenges in Inefficient
[46][52][54][66 achieving accurate address accuracy and scalability challenges in
Vector managing security utilization of
] end-to-end security in securing network slices, optimizing resource
Machine) across numerous security resources
network slices utilization in 5G network slicing.
network slices
Scalability
Application of advanced deep reinforcement
Accuracy challenges concerns in Inefficient resource
learning techniques, such as hierarchical or multi-
in context-aware managing allocation for
[50][57][60][68 DQN (Deep Q- agent DQN, to improve accuracy, scalability, and
authentication authentication authentication
] Network) resource allocation for context-aware
handover for secure handover across handover processes
authentication handover in secure 5G network
network slicing numerous network
slicing.
slices
Lack of accuracy in Scalability issues Inefficient resource Advanced LSTM-based deep learning with
LSTM (Long
network slice in managing a allocation and RL/attention improves accuracy, scalability, and
[45][49][69] Short-Term
orchestration and large number of utilization resource allocation in secure 5G network slice
Memory)
optimization network slices orchestration.
Limited
computational Sub-optimal
Limited local data in Improve model performance with data
resources in resource allocation
network slices augmentation and transfer learning, optimize
FL (Federated network slices and usage can
[58][59] hampers model resource allocation for scalability, and employ
Learning) hinder scalability, impact system
performance and efficient scheduling techniques to enhance system
impacting efficiency and
accuracy. efficiency in secure 5G network slicing.
Federated performance
Learning.
Scalability
K-means Accuracy issues in concerns in Inefficient Improving accuracy of automated machine
clustering, automated machine managing and utilization of learning models; Scalable orchestration
[48]
Naive Bayes learning for network orchestrating a automated processes frameworks; Efficient utilization of automated
classifier slice automation large number of processes through optimization
network slices
Scalability issues
Accuracy challenges
GAN in detecting and Inefficient resource Improving accuracy of adversarial machine
in adversarial machine
(Generative mitigating utilization during learning models; Scalable flooding attack
[53] learning for flooding
Adversarial flooding attacks attack mitigation detection mechanisms; Efficient utilization of
attacks in network
Networks) across multiple resources during attack mitigation
slicing
network slices
Scalability issues
Accuracy issues due in managing Inefficient resource
to complex network numerous slices utilization, Data augmentation, transfer learning, efficient
DBN (Deep
slices, limited labeled efficiently and communication parallelization, optimized resource allocation, and
[54] Belief
data, and difficulty handling increased overhead, and adaptive strategies to improve accuracy,
Networks)
capturing intricate computational and synchronization scalability, and resource utilization.
patterns. communication processes.
overhead.
384 | P a g e
Vol. 14, No. 12, 2023
mitigating such attacks. A hybrid DL-based approach for

wireless network slicing is proposed in [52]. This approach
combines deep neural networks for network traffic prediction
and a reinforcement learning algorithm for resource allocation
optimization. It aims to improve the reliability and accuracy of
network slicing while ensuring end-to-end security. In [53], an
adversarial ML approach is proposed to detect and mitigate
flooding attacks on 5G radio access network slicing. It
enhances security by effectively identifying and mitigating
such attacks. Benzaid et al. [54] propose an AI-based
autonomic security management architecture for secure
network slicing in B5G networks, focusing on effective
security management leveraging AI techniques. Another
approach proposed in [56] introduces a learning augmented
optimization approach to safeguard network slicing in 5G. It
presents a mathematical model and optimization algorithm
Fig. 4. Algorithm performance for secure 5G network slicing. that utilizes machine learning to adapt to network environment
changes, minimizing network cost while ensuring secure slices
Jiang et al. [44] proposed a "Intelligence Slicing" meeting quality of service requirements. [57] presents
framework combining network slicing and AI for improved SliceBlock, a secure network slicing scheme utilizing a DAG-
intelligence, security, and flexibility. To maximize the blockchain to authenticate handover requests between network
efficiency of network slicing and to ensure end-to-end slices in edge-assisted SDN/NFV-6G environments. It
security, the framework employs AI techniques such as integrates DAG-blockchain technology with SDN/NFV and
reinforcement learning, transfer learning, and deep learning. edge computing, providing a secure and reliable network
Deep neural networks are used in [45] to predict resource slicing service. Federated learning, proposed in [59], aims to
requirements for different slices, and a clustering technique is improve security in network slicing by aggregating training
used to group analogous slices. The goal is to eliminate data from multiple slices to detect anomalies and security
wastage of resources while maintaining network stability. Liu threats across the entire network. Simulation experiments
et al. [46] suggested a learning-assisted secure end-to-end demonstrate its effectiveness in enhancing security threat
network slicing technique for cyber-physical systems (CPS). detection accuracy. Lastly, [60] proposes a reinforcement
For efficient resource allocation among similar CPSs, ML learning approach for attacking and defending NextG radio
approaches such as k-means clustering and k-nearest access network slicing. It utilizes multi-agent reinforcement
neighbors are used. To maintain slice confidentiality and learning to train agents capable of launching attacks and
integrity, a secure key exchange mechanism is used. defending against them in network slicing scenarios.
Sedjelmaci [47] presented a cooperative attack detection Q2. What are the main obstacles and constraints faced by
system based on AI, applying machine learning techniques these techniques in terms of accuracy, scalability, and
such as random forest and decision trees to identify any type efficiency?
of malicious activity and differentiate between legitimate and
malicious traffic. This method improves network slicing Network slicing has become a pivotal concept within 5G
security by detecting and mitigating threats more effectively. networks, enabling operators to divide their networks into
Kafle et al. [48] developed an ML-based network slicing virtualized networks that are customized to meet the distinct
automation strategy based on a multi-agent reinforcement needs of various applications and services. Integrating
learning technique. To provide end-to-end security, the machine learning (ML) and deep learning (DL) techniques
approach learns from previous slicing events to make accurate presents a promising avenue for enhancing the security of
slicing decisions while optimizing resource allocation and network slicing. This integration facilitates the detection and
restricting the attack surface. Secure5G is a DL-based prevention of security threats in real-time, offering an
architecture for secure network slicing that was introduced in effective approach to safeguarding network slicing. However,
[49]. It makes use of deep neural networks to estimate slice leveraging ML and DL in secure network slicing poses
resource requirements and efficient resource allocation. The various challenges, including the need for labeled data,
framework consists of a security module that detects and complexity of models, and ensuring data confidentiality and
mitigates network attacks using a deep belief network. A privacy. Subsequent sections will delve into a comprehensive
reinforcement learning-based technique for attacking and examination of these challenges, providing a detailed analysis,
defending 5G radio access network slicing is proposed in [50]. and presenting potential solutions to address them.
It learns optimal attack and defense techniques through deep
Jiang et al. [44] propose a comprehensive framework
reinforcement learning, enabling network administrators to
named "Intelligence slicing" that combines artificial
identify and mitigate vulnerabilities. The DeepSecure
intelligence (AI) with 5G networks. However, they do not
technique introduced in [51] analyzes traffic patterns and
discuss the limitations in terms of accuracy and scalability
detects distributed denial-of-service (DDOS) attacks on 5G
associated with their framework. Thantharate et al. [45]
network slicing, using convolutional neural networks. It
propose "DeepSlice," an approach based on deep learning
enhances reliability and security by effectively detecting and
(DL) for achieving efficient and dependable network slicing in
385 | P a g e
Vol. 14, No. 12, 2023
5G networks. The authors recognize the challenges associated their paper [58], Bandara et al. introduces a federated learning
with handling large volumes of training data and the need for platform for network slicing that incorporates blockchain and
substantial computational resources, which could potentially zero trust security mechanisms. However, the proposed
impede scalability and efficiency. Liu et al. [46] introduce the platform encounters challenges in terms of ensuring high
concept of "Learning-assisted secure end-to-end network accuracy and scalability. Similarly, Wijethilaka and Liyanage
slicing" for cyber-physical systems. They emphasize the issue [59] suggest a federated learning approach to enhance security
of explainability in machine learning-based security in network slicing. However, the approach faces challenges in
approaches, underscoring its potential impact on trust and the terms of scalability and efficiency. On the other hand, Shi et
acceptance of such methods. Sedjelmaci [47] puts forward a al. [60] utilize reinforcement learning for attacking and
cooperative attack detection system based on artificial defending NextG radio access network slicing. However, their
intelligence (AI) for 5G networks. The authors highlight the proposed approach encounters limitations in terms of
drawbacks of conventional rule-based systems and emphasize scalability and accuracy. Lastly, Chowdhury et al. [61]
the potential advantages of AI-based systems, but they do not introduce AUTODEEPSLICE, a data-driven technique for
explicitly address the specific limitations associated with such network slicing that employs automatic deep learning.
approaches. Kafle et al. [48] suggest the automation of 5G However, they face challenges in achieving accurate results
network slicing through the utilization of machine learning due to the inherent complexity of network slicing.
techniques. The authors recognize the challenges brought
about by the complexity of the network and the requirement Q3. What are the potential solutions to address these
for substantial amounts of training data. Thantharate et al. [49] challenges and limitations?
introduce "Secure5G," a DL framework for secure network 1) Solutions presented in the literature: One possible
slicing in 5G and future networks. However, they recognize solution to tackle the challenges and limitations discussed in
the challenge of developing DL models capable of detecting the papers involves developing and implementing intelligent
and defending against new and unknown attacks. The research
slicing frameworks that integrate artificial intelligence (AI)
in [50] presents a reinforcement learning approach for
attacking and defending 5G radio access network slicing, but into 5G networks [44]. This strategy capitalizes on machine
does not delve into the accuracy or scalability limitations of learning algorithms to optimize and automate network slicing
the approach. Kuadey et al. [51] introduce "DeepSecure," processes [48], enhancing the reliability of network slicing
which is a deep learning-based approach for detecting [52], and fortifying the security of 5G networks against cyber-
distributed denial-of-service (DDoS) attacks in 5G network attacks [46][49][51]. An alternative strategy is to utilize
slicing. The authors acknowledge the difficulty of developing reinforcement learning to protect against flooding attacks in
accurate models while minimizing false positives. The study 5G radio access network slicing [53][60][63][64][65].
presented in [52] proposes a hybrid deep learning approach for Furthermore, a context-aware authentication handover and
achieving high accuracy and reliability in wireless network secure network slicing that incorporates a DAG-blockchain in
slicing within the context of 5G. However, the study does not
edge-assisted SDN/NFV-6G environments can enhance the
delve into the limitations of this approach. Shi and Sagduyu
the approach proposed in [53] is based on adversarial machine security of network slicing [57]. Federated learning can also
learning and aims to address flooding attacks on 5G radio be deployed to bolster the security of network slicing [59]. By
access network slicing. However, the study does not explicitly incorporating the Skunk-A Blockchain and Zero Trust
discuss the challenge of developing robust models that are Security Enabled Federated Learning Platform, the security of
resilient to adversarial attacks. Benzaid et al, [54] introduces 5G/6G network slicing is significantly enhanced. [58]. Finally,
an AI-based autonomic and scalable security management data-driven network slicing techniques utilizing automatic
architecture for secure network slicing in B5G. However, they deep learning can be employed to enhance network slicing in
do not specifically discuss the limitations related to 5G networks [61].
developing models capable of handling the dynamic and 2) Recommended solutions: The integration of
heterogeneous nature of B5G networks.. Yoon et al. [55] Convolutional Neural Networks (CNNs) with reinforcement
presents a technique called "Moving target defense for in-
vehicle software-defined networking" that utilizes IP shuffling learning techniques offers a promising solution to tackle
in network slicing, combined with multi-agent deep challenges related to accuracy, scalability, and resource
reinforcement learning. They address the challenges utilization in secure 5G network slicing effectively. This
associated with accurate attack detection and mitigation in integration can be applied to various aspects, including
dynamic and mobile environments. Cheng et al. [56] suggests network slice orchestration, DDoS attack detection, and
a method called "Safeguard network slicing in 5G" that simultaneous protection of multiple network slices.
employs a learning-augmented optimization approach. They a) Markov Decision Process (MDP): Consider the
acknowledge the challenge of managing the complexity and problem of network slice orchestration as a Markov Decision
dynamics inherent in 5G networks. Abdulqadder and Zhou Process (MDP) [63], defined by the tuple ( ). Here,
[57] present "SliceBlock," a solution that combines context-
is the state space, is the action space, represents the
aware authentication handover and secure network slicing
transition probabilities, is the reward function, and is the
using DAG-blockchain in edge-assisted SDN/NFV-6G
environments. However, the study does not delve into the discount factor. The objective is to find an optimal policy
specific limitations associated with their proposed models. In that maximizes the expected cumulative reward:
[∑ ( )] (1)
386 | P a g e
Vol. 14, No. 12, 2023
Where, represents the optimal policy that provides the achievement of optimal resource utilization and scalability in
best actions in each state , taking into account the long- 5G network slicing.
term cumulative reward. The MDP formulation considers We define the ‗NetworkSlicingEnv‘ environment and then
network slice orchestration as a sequential decision-making train a PPO model to optimize resource allocation within the
problem. By optimizing the policy , it ensures that actions environment. After training, we evaluate the model's
taken in each state lead to the maximum cumulative reward, performance by calculating the mean reward over a specified
which is vital for efficient network resource allocation. number of episodes, denoted by ‗n_episodes‘. This mean
b) Bayesian reinforcement learning: To improve DDoS reward serves as an indicator of the model's resource
attack detection, we can model it as a Bayesian Reinforcement allocation capabilities in the secure 5G network slicing
scenario. The algorithm presented below depicts the scenario:
Learning problem [64], where we need to infer the optimal
policy given a sequence of observations and actions . Algorithm:
We seek to maximize the posterior probability of the policy
Inputs:
given the data:
( ) ( ) ( ) (2) ‗n_episodes‘: Number of episodes for evaluation
Where, ( ) is the posterior policy probability, Outputs:

( ) is the likelihood of observations given the policy,
‗mean_reward‘: Mean reward over the evaluation
and ( ) is the prior policy probability. The Bayesian RL
episodes
formulation models DDoS attack detection as a probabilistic
learning problem. It estimates the posterior policy given Initialize:
observed data, enabling improved detection accuracy through
probabilistic reasoning. 1. Define the environment class ‗NetworkSlicingEnv‘:
a. Initialize the action space with 10 discrete resource
c) Multi-objective optimization: For simultaneous allocation options.
protection of multiple network slices, we can formulate it as a b. Initialize the observation space with a 5-dimensional
Multi-Objective Optimization problem [65]. Let state observation.
[ ( ) ( ) ( )] represent a vector of objective
functions, and [ ] denote the vector of Procedure ‗TrainModel‘ (Total Timesteps: 10,000):
decision variables. Additionally, 1. Create a multi-agent environment instance ‗env‘.
[ ( ) ( ) ( )] represents a vector of 2. Define a PPO model with a multi-layer perceptron policy
inequality constraints. The goal is to find a vector of decision (‗MlpPolicy‘) for policy optimization.
variables that optimizes multiple objectives while 3. Train the model with the ‗learn‘ method using
satisfying constraints: ‗total_timesteps=10,000‘.
( ) [ ( ) ( ) ( )] (3)
Procedure ‗EvaluateModel‘ (Number of Episodes:
Subject to: ‗n_episodes‘):
( ) 1. Initialize ‗mean_reward‘ to 0.
2. For each episode in the range ‗n_episodes‘, do the
The multi-objective optimization approach provides a
following:
rigorous framework to balance multiple objectives while
considering constraints. It ensures that resources are a. Reset the environment (‗env‘) and obtain the initial
efficiently allocated to protect multiple network slices. state.
b. For each time step:
d) Multi-agent reinforcement learning: In the context of
secure 5G network slicing, multi-agent reinforcement learning i. Predict an action using the trained model (‗model‘)
[66] demonstrates potential for addressing accuracy, based on the current state.
scalability, and resource utilization challenges. Specifically, it ii. Simulate the environment's response and receive
can enable intelligent resource allocation by considering the reward.
complex dependencies and facilitating IP shuffling processes.
We performed an experiment using python to demonstrate a iii. Accumulate the reward.
custom reinforcement learning environment, termed iv. Check if the episode is done. If done, break the
NetworkSlicingEnv, designed to emulate a 5G network slicing loop.
scenario, where an agent makes resource allocation decisions
with the goal of optimizing the allocation of resources based 3. Calculate the ‗mean_reward‘ as the sum of rewards over
on dependencies among actions. The experiment utilizes the ‗n_episodes‘ divided by ‗n_episodes‘.
Proximal Policy Optimization (PPO) algorithm, a cutting-edge Output ‗mean_reward‘.
reinforcement learning method. The agent's training process
includes learning to optimize resource allocation while End
considering dependencies among actions, contributing to the
387 | P a g e
Vol. 14, No. 12, 2023
Algorithm:
Input:
● Training data: ( )
● Testing data: ( )
Initialization:
a. Initialize SVM model: .
Data Preparation:
a. Split the data into training and testing sets:
● for training.
● for testing.
Accuracy Optimization:
Fig. 5. Multi-Agent reinforcement learning training rewards over time. a. Train the SVM model on the training data:
● ( )
The experimental results reveal the agent's learning b. Make predictions on the test data:
progress, represented through a 3D plot showcasing the ● ( ).
cumulative reward over training steps as shown in Fig. 5. The c. Calculate accuracy (A) using the ground truth and
positive rewards indicate that the agent is learning to make predictions:
resource allocation decisions that result in outcomes deemed
favorable. This implies that the agent is successfully adapting ●
its behavior over time to maximize the cumulative rewards.
The learning dynamics suggest that the agent is gradually Scalability Optimization:
improving its understanding of the environment and making a. In SVM, the margin is automatically maximized
better choices in allocating resources. As the training during training to improve scalability (S).
progresses, the agent becomes more proficient in resource
allocation. The positive rewards show that the agent is making Resource Utilization:
decisions that lead to efficient resource utilization. This can be a. SVMs automatically select a subset of support
seen as an indication of improved performance in managing vectors for decision function, which reduces resource
network resources for 5G network slicing. The 3D plot utilization (R).
visualizes how the agent's rewards change over the course of
training. Gradually, the agent should learn to make resource Output:
allocation decisions that result in higher cumulative rewards, a. Return:
which signifies an improved understanding of the environment ● Trained SVM model: .
and more effective decision-making. The mean reward, ● Accuracy (A).
calculated at the end of training, provides a summary of the ● Scalability (S).
agent's performance. A higher mean reward suggests that the ● Resource Utilization (R).
agent has improved its decision-making skills regarding
resource allocation, which is a critical aspect of network End
slicing in 5G environments. The positive rewards demonstrate
The data is divided into training and testing sets to
that the agent's learning dynamics are effective in improving
evaluate the SVM model's performance. The SVM model
its performance in resource allocation decision-making. The
employs a linear kernel for classification and is trained on the
agent gradually learns to allocate resources more efficiently,
training data. The decision boundary is plotted along with the
resulting in higher cumulative rewards and, by extension, data points, with a color scale indicating the accuracy level of
better performance for secure 5G network slicing. 0.88 in this case. The visualization demonstrates how SVM, as
e) Support Vector Machine (SVM): For securing a scalable framework, can effectively classify data while
network slices in a scalable manner, distributed and scalable maintaining a high level of accuracy as depicted in Fig. 6.
Support Vector Machine (SVM) [67] frameworks can be This technique proves to be an optimal solution for ensuring
employed. These frameworks contribute to improving both network slice security and efficient resource allocation within
accuracy and scalability while optimizing resource utilization the complex 5G environment, making this experiment a
within the 5G network slicing environment. In the experiment valuable illustration of SVM's capabilities in a network slicing
conducted, a synthetic dataset was generated for binary context.
classification, simulating a scenario in which SVM-based
classification is employed. The code utilizes the Scikit-Learn
library to create a two-dimensional dataset with 1000 samples.
The algorithm presented below depicts the scenario:
388 | P a g e
Vol. 14, No. 12, 2023
● For each agent (‗i = 1‘ to ‗num_agents‘):

a. Create a DQNAgent (‗agent[i]‘) with
‗state_size‘ and ‗action_size‘.
b. Build a neural network model for ‗agent[i]‘ with
three dense layers.
Training Loop:
● For each episode (‗e = 1‘ to ‗episodes‘):
a. Generate random initial states for each agent and
store in states.
b. For each step in the episode:
i. Query each agent for actions based on their current
state.
ii. Implement the logic to take actions and receive new
states, rewards, and 'done' flags.
Fig. 6. Secure network slice classification: SVM decision boundary with
88% accuracy. iii. Update each agent's Q-network based on their
experiences.
f) Deep learning: By leveraging sophisticated deep
iv. If all agents are done (all 'done' flags are True), exit
reinforcement learning methods like hierarchical or Multi-
the loop.
Agent Deep Q-Networks (MADQN) [68], it is possible to
improve accuracy, scalability, and resource allocation in the Output:
context of context-aware authentication handover within ● Trained MADQN agents.
secure 5G network slicing. We used TensorFlow library to
demonstrate the concept of rewards over a series of episodes, End
environments, which is a critical aspect of evaluating the
performance of Multi-Agent Deep Q-Networks (MADQN) in
context-aware authentication handover within 5G network
slicing environments. These rewards are indicative of the
efficiency and learning progress of a group of agents working
collaboratively to make decisions. The rewards represent
various factors, including the success of authentication
handover processes, the minimization of resource utilization,
and the overall optimization of network slicing. By examining
the rewards over episodes, we gain valuable insights into how
well the MADQN is adapting and learning within this
environment as shown in Fig. 7. Consistent, positive rewards
indicate that the MADQN is effectively enhancing the
authentication handover process and resource allocation,
contributing to scalability and the overall security of 5G Fig. 7. Multi-Agent Deep Q-Networks (MADQN) episode rewards over
network slicing. time.
Algorithm:
Furthermore, the utilization of advanced Long Short-Term
Input: Memory (LSTM)-based deep learning architectures [69],
● State size (‗state_size‘) integrated with reinforcement learning or attention
● Action size (‗action_size‘) mechanisms, shows promise in improving accuracy,
● Number of agents (‗num_agents‘) scalability, and resource allocation for secure 5G network
● Total training episodes (‗episodes‘) slice orchestration.
Initialization: g) Other potential solutions: To further improve the

performance of secure 5G network slicing, data augmentation
● Initialize ‗state_size‘, ‗action_size‘, ‗num_agents‘, and and transfer learning techniques can enhance model
‗episodes‘. performance, while optimized resource allocation ensures
● Initialize the multi-agent environment (‗env‘) with scalability. Efficient scheduling techniques contribute to
given parameters.
enhancing system efficiency within the secure 5G network
Agent Creation: slicing context. Moreover, potential solutions for improving
389 | P a g e
Vol. 14, No. 12, 2023
the accuracy of automated machine learning models, scalable promise in enhancing network slice orchestration, enabling
orchestration frameworks, and efficient utilization of DDoS attack detection, and facilitating simultaneous
automated processes through optimization are essential protection of multiple network slices [44]. Another area of
considerations. Additionally, addressing the accuracy of interest is multi-agent reinforcement learning, which can be
adversarial machine learning models, deploying scalable explored to tackle accuracy, scalability, and resource
flooding attack detection mechanisms, and efficiently utilizing utilization challenges in secure 5G network slicing. By
resources during attack mitigation are crucial aspects to leveraging intelligent resource allocation and effectively
enhance the security and performance of secure 5G network managing complex dependencies, multi-agent reinforcement
slicing. Overall, employing techniques such as data learning offers the potential for more efficient and effective
augmentation, transfer learning, efficient parallelization,
network slicing [45].
optimized resource allocation, and adaptive strategies holds
promise for enhancing accuracy, scalability, and resource Furthermore, the literature suggests the exploration of
utilization in the secure 5G network slicing domain. distributed and scalable frameworks such as support vector
machines (SVM) to address accuracy and scalability
challenges in securing network slices and optimizing resource
utilization in 5G network slicing [46]. Additionally, advanced
deep reinforcement learning techniques, including hierarchical
or multi-agent deep Q-networks (DQN), can be applied to
improve accuracy, scalability, and resource allocation in
context-aware authentication handover for secure 5G network
slicing [52][68]. The use of advanced LSTM-based deep
learning architectures, coupled with reinforcement learning or
attention mechanisms, also holds promise for enhancing
accuracy, scalability, and resource allocation in secure 5G
network slice orchestration [69].
Fig. 8. Word cloud visualization of potential solutions for secure 5G
network slicing
2) Recommended future directions
a) Ensemble learning techniques: Ensemble learning
The word cloud depicted in Fig. 8 is a visual (stacking or boosting for accuracy and robustness
representation of potential solutions for enhancing secure 5G enhancement) involves combining multiple individual models
network slicing, as discussed in the literature. The above word to create a stronger and more accurate predictive model.
cloud was generated using python. This visualization method Stacking and boosting are two common ensemble techniques.
succinctly conveys the key strategies and concepts described Stacking combines the predictions of multiple models using
in the literature by assigning word sizes based on their another model, often referred to as a meta-learner. Boosting,
frequency of occurrence. Larger and bolder words indicate the on the other hand, assigns more weight to instances that are
prominence and importance of specific terms within the text. misclassified by the previous models in the ensemble, thereby
Notably, terms such as "security," "reinforcement learning," focusing on improving the accuracy of these instances. The
"slicing frameworks," "resource utilization," and "scalability"
application of ensemble techniques in secure 5G network
are the most prominent in the word cloud. These words
slicing aims to enhance prediction accuracy and overall
represent the primary solutions put forward in the literature for
addressing the challenges and limitations related to secure 5G robustness by leveraging the strengths of multiple models.
network slicing. The word cloud effectively distills the b) Integration of explainable AI: Explainable AI
essence of the text, providing an at-a-glance overview of the techniques (for transparency and trust) aim to provide
critical ideas and strategies to enhance accuracy, scalability, understandable and interpretable explanations for the
and resource utilization in the secure 5G network slicing decisions made by machine learning models. In the context of
domain. Scientists and researchers can quickly grasp the core secure 5G network slicing, incorporating explainable AI
themes and solutions discussed in the paper, making it a methods allows users to understand the rationale behind
valuable addition to the scientific discourse for understanding decisions made by the network slicing system. This
complex topics with enhanced clarity. transparency enhances trust and accountability, particularly in
Q4. What are the future research directions and critical applications where the ability to explain decisions is
opportunities in this area? essential. By enabling stakeholders to comprehend how
decisions are reached, explainable AI techniques contribute to
1) Future directions as discussed in the literature: The increased confidence and acceptance of the system's
papers surveyed in this study present several potential future outcomes.
research directions in the domain of secure 5G network c) Federated learning: Federated learning (for data
slicing. One promising avenue is the integration of privacy and security) is a decentralized machine learning
reinforcement learning techniques with convolutional neural approach that enables model training across multiple devices
networks (CNNs) to address challenges related to accuracy, or nodes while keeping data localized. This approach
scalability, and resource utilization. This approach has shown prioritizes data privacy and security by not requiring data to be
390 | P a g e
Vol. 14, No. 12, 2023
centralized for training. In the context of secure 5G network importance of securing network slices against cyber threats,
slicing, federated learning offers a solution to challenges with studies exploring the detection and mitigation of DDoS
related to sharing sensitive data across different slices. It attacks, cooperative attacks, and adversarial machine learning
allows models to be trained collaboratively without sharing models. Additionally, considerations for privacy,
raw data, thus ensuring privacy while still improving the explainability, edge computing, energy efficiency, and the
performance of network slicing algorithms. application of emerging techniques like Ensemble Learning,
Federated learning, GANs, etc. were highlighted as potential
d) Edge computing: Utilization of edge computing and
future research directions. By exploring these avenues, this
edge intelligence for localized decision-making involves review has laid a foundation for researchers to contribute to
processing data closer to the data source, reducing the advancement of secure 5G network slicing, addressing the
communication overhead and latency. Incorporating edge challenges and ensuring the reliability, efficiency, and security
intelligence in network slicing enables localized decision- of future network infrastructures.
making at the edge nodes, which can lead to quicker responses
and efficient resource utilization. By processing data and This review explored the significant advancements and
making decisions closer to where it is generated, edge challenges in the integration of 5G networks with security,
computing optimizes the performance of network slicing and network slicing, and machine learning techniques. The
findings reveal that 5G networks offer immense potential for
enhances its responsiveness.
delivering high accuracy, scalability, and efficiency in various
e) Energy efficiency: Energy efficiency through applications. Network slicing emerges as a crucial mechanism
algorithmic development and resource allocation is a critical for resource allocation and management in 5G networks,
concern in any network infrastructure. In the context of secure enabling efficient utilization of network resources for different
5G network slicing, developing energy-efficient algorithms services. Moreover, machine learning and deep learning
and resource allocation strategies is essential to minimize algorithms demonstrate promising capabilities in enhancing
power consumption while maintaining optimal performance. network performance and security by enabling intelligent
By optimizing the allocation of resources based on workload decision-making and anomaly detection. However, the
and demand, energy-efficient network slicing can contribute to integration of these technologies also presents notable
reducing operational costs and improving overall challenges, such as ensuring robust security measures,
sustainability. optimizing network slicing algorithms, and addressing
f) Generative adversarial networks: Generative scalability issues. Further research and development efforts
Adversarial Networks (GANs) are a type of machine learning are required to overcome these challenges and fully exploit the
potential of 5G networks with security, network slicing, and
model used in tasks such as image generation and data
machine learning for various domains and applications.
synthesis. In the context of secure 5G network slicing, GANs
can be employed to generate realistic adversarial examples ACKNOWLEDGMENT
that simulate potential attacks on the network slicing
This project was funded by Deanship of Scientific
algorithms. By testing the robustness of these algorithms Research, Northern Border University for their financial
against such simulated attacks, GANs can help identify support under grant no. (NR-NBU-2023-12-2125). The
vulnerabilities and weaknesses in the system's security authors, therefore, acknowledge with thanks DSR technical
measures, thereby enhancing its overall security posture. and financial support.
VII. CONCLUSION REFERENCES
This literature survey presented a comprehensive overview [1] X. Li et al., ―Network slicing for 5G: Challenges and opportunities,‖
of the research conducted in the domain of secure 5G network IEEE Internet Computing, pp. 1–1, 2018.
slicing. The reviewed papers highlighted the challenges doi:10.1109/mic.2018.326150452.
related to accuracy, scalability, and resource utilization in [2] H. Yu, H. Lee, and H. Jeon, ―What is 5G? emerging 5G Mobile Services
network slicing and proposed various techniques to address and network requirements,‖ Sustainability, vol. 9, no. 10, p. 1848, 2017.
doi:10.3390/su9101848.
these issues. The experimental results show that the
[3] R. Dangi et al., ―ML-based 5G network slicing security: A
integration of reinforcement learning techniques with CNNs, comprehensive survey,‖ Future Internet, vol. 14, no. 4, p. 116, 2022.
multi-agent reinforcement learning, and distributed SVM doi:10.3390/fi14040116.
frameworks emerged as potential solutions to improve [4] T. Chhabra, ―5G in India: The journey is about to begin - ET telecom,‖
accuracy and scalability in network slicing. Advanced deep ETTelecom.com, https://telecom.economictimes.indiatimes.com/news/
reinforcement learning architectures, such as hierarchical or 5g-in-india-the-journey-is-about-to-begin/81671088 (accessed Feb. 12,
multi-agent DQN, and LSTM-based models with 2023).
reinforcement learning or attention mechanisms were [5] T. GreyB, ―5G companies: 12 players are leading the research,‖ GreyB,
https://www.greyb.com/blog/5g-companies/ (accessed Feb. 9, 2023).
identified as effective approaches for enhancing accuracy,
[6] G. Narcisi, ―These are the 5G trends to watch in 2021,‖ CRN,
scalability, and resource allocation in network slice https://www.crn.com/news/networking/these-are-the-5g-trends-to-
orchestration. Furthermore, data augmentation, transfer watch-in-2021 (accessed Feb. 10, 2023).
learning, efficient parallelization, optimized resource [7] ―What is 5g: Everything you need to know About 5G: 5G FAQ:
allocation, and adaptive strategies were suggested as methods Qualcomm,‖ Wireless Technology & Innovation,
to improve model performance, scalability, and resource https://www.qualcomm.com/5g/what-is-5g# (accessed Feb. 9, 2023).
utilization. The surveyed literature also shed light on the
391 | P a g e
Vol. 14, No. 12, 2023
[8] N. Brittain, ―18 5G projects providing a vision for the future,‖ 5Gradar, [30] GROUP SPECIFICATION, ―ETSI GS NFV-SEC 001 V1.1.1 (2014-
https://www.5gradar.com/features/5g-projects-that-will-blow-your-mind 10),‖ Etsi.org. [Online]. Available:
(accessed Feb. 14, 2023). https://www.etsi.org/deliver/etsi_gs/nfv-
[9] H. Zhang et al., ―Network slicing based 5G and future mobile networks: sec/001_099/001/01.01.01_60/gs_nfv-sec001v010101p.pdf. [Accessed:
Mobility, Resource Management, and challenges,‖ IEEE 10-April-2023].
Communications Magazine, vol. 55, no. 8, pp. 138–145, 2017. [31] A. Mathew, ―Network slicing in 5G and the security concerns,‖ in 2020
doi:10.1109/mcom.2017.1600940. Fourth International Conference on Computing Methodologies and
[10] C. Campolo, A. Molinaro, A. Iera, and F. Menichella, ―5G network Communication (ICCMC), 2020.
slicing for vehicle-to-everything services,‖ IEEE Wireless [32] Z. Kotulski et al., ―On end-to-end approach for slice isolation in 5G
Communications, vol. 24, no. 6, pp. 38–45, 2017. networks. Fundamental challenges,‖ in Proceedings of the 2017
doi:10.1109/mwc.2017.1600408. Federated Conference on Computer Science and Information Systems,
[11] G. Werélius, What we know: A look at current 5G market trends - 2017.
ericsson, https://www.ericsson.com/en/blog/2020/10/what-we-know-a- [33] I. Afolabi, T. Taleb, K. Samdanis, A. Ksentini, and H. Flinck, ―Network
look-at-current-5g-market-trends (accessed Feb. 9, 2023). slicing and softwarization: A survey on principles, enabling
[12] R. F. Olimid and G. Nencioni, ―5G network slicing: A security technologies, and solutions,‖ IEEE Commun. Surv. Tutor., vol. 20, no.
overview,‖ IEEE Access, vol. 8, pp. 99999–100009, 2020. 3, pp. 2429–2453, 2018.
doi:10.1109/access.2020.2997702. [34] F. Reynaud et al., ―Attacks against network functions virtualization and
[13] F. Salahdine, Q. Liu, and T. Han, ―Towards secure and intelligent software-defined networking: state-of-the-art,‖ in Proceedings of
network slicing for 5G networks,‖ IEEE Open Journal of the Computer Workshop on Security in Virtualized Networks, Sec-Virtnet, 2016.
Society, vol. 3, pp. 23–38, 2022. doi:10.1109/ojcs.2022.3161933. [35] V. N. Sathi, M. Srinivasan, P. K. Thiruvasagam, and S. R. M.
[14] E. P. Neto, F. S. Silva, L. M. Schneider, A. V. Neto, and R. Immich, Chebiyyam, ―A novel protocol for securing network slice component
―Seamless Mano of multi-vendor SDN controllers architectures and association and slice isolation in 5G networks,‖ in Proceedings of the
future challenges,‖ Comput. Netw., vol. 167, no. 106984, p. 106984, 21st ACM International Conference on Modeling, Analysis and
2020. Simulation of Wireless and Mobile Systems, 2018.
[15] S. Chaabnia and A. Meddeb, ―Slicing aware QoS/QoE in Software [36] ―Security aspects of network capabilities exposure in 5G,‖ in Final
Defined Smart Home Network,‖ NOMS 2018 - 2018 IEEE/IFIP deliverable (approved-P Public), 2018.
Network Operations and Management Symposium, 2018. [37] R. Khan, P. Kumar, D. N. K. Jayakody, and M. Liyanage, ―A survey on
doi:10.1109/noms.2018.8406195. security and privacy of 5G technologies: Potential solutions, recent
[16] P. K. Chartsias et al., ―SDN/NFV-based end to end network slicing for advancements, and future directions,‖ IEEE Commun. Surv. Tutor., vol.
5G multi-tenant networks,‖ in 2017 European Conference on Networks 22, no. 1, pp. 196–248, 2020.
and Communications (EuCNC), 2017. [38] S. Zhou, L. Wu, and C. Jin, ―A privacy-based SLA violation detection
[17] C. Bektas, S. Monhof, F. Kurtz, and C. Wietfeld, ―Towards 5G: An model for the security of cloud computing,‖ China Commun., vol. 14,
empirical evaluation of software-defined end-to-end network slicing,‖ in no. 9, pp. 155–165, 2017.
2018 IEEE Globecom Workshops (GC Wkshps), 2018. [39] S. Y.-T. Fan C-I, ―Cross-network-slice authentication scheme for the 5th
[18] A. A. Barakabitze, A. Ahmad, R. Mijumbi, and A. Hines, ―5G network generation mobile communication system,‖ IEEE Transactions on
slicing using SDN and NFV: A survey of taxonomy, domains,‖ IEEE Network and Service Management, 2021.
Access, vol. 8, pp. 29525–29537, 2020. [40] ―Security,‖ 5G Second Phase Explained, pp. 235–258, 2021.
[19] X. Li, R. Ni, J. Chen, Y. Lyu, Z. Rong, and R. Du, ―End-to-end network doi:10.1002/9781119645566.ch7.
slicing in radio access network, transport network and core [41] S. Bao, Y. Liang, and H. Xu, ―Blockchain for network slicing in 5G and
networkacross Federated Multi-domains,‖ Computer Networks, vol. beyond: Survey and challenges,‖ J. Commun. Inf. Netw., vol. 7, no. 4,
186, p. 107752, 2021. doi:10.1016/j.comnet.2020.107752. pp. 349–359, 2022.
[20] S. D‘Oro, F. Restuccia, A. Talamonti, and T. Melodia, ―The slice is [42] S. K. Boell and D. Cecez-Kecmanovic, ‗‗On being ‗systematic‘ in
served: Enforcing radio access network slicing in virtualized 5G literature reviews,‘‘ in Formulating Research Methods for Information
systems,‖ in IEEE INFOCOM 2019 - IEEE Conference on Computer Systems. London, U.K.: Palgrave Macmillan, 2015, pp. 48–78.
Communications, 2019. [43] M. D. J. Peters, C. M. Godfrey, H. Khalil, P. McInerney, D. Parker, and
[21] A. Kaloxylos, ―A survey and an analysis of network slicing in 5G C. B. Soares, ‗‗Guidance for conducting systematic scoping reviews,‘‘
networks,‖ IEEE Commun. Stand. Mag., vol. 2, no. 1, pp. 60–65, 2018. Int. J. Evidence-Based Healthcare, vol. 13, no. 3, pp. 141–146, Sep.
[22] F. Salahdine and N. Kaabouch, ―Security threats, detection, and 2015.
countermeasures for physical layer in cognitive radio networks: A [44] W. Jiang, S. D. Anton, and H. Dieter Schotten, ―Intelligence slicing: A
survey,‖ Phys. Commun., vol. 39, no. 101001, p. 101001, 2020. unified framework to integrate artificial intelligence into 5G networks,‖
[23] M. A. Habibi, B. Han, and H. D. Schotten, ―Network slicing in 5G in 2019 12th IFIP Wireless and Mobile Networking Conference
mobile communication architecture, profit modeling, and challenges,‖ (WMNC), 2019.
arXiv [cs.NI], 2017. [45] A. Thantharate, R. Paropkari, V. Walunj, and C. Beard, ―DeepSlice: A
[24] N. Alliance, ―5G security recommendations package# 2: Network deep learning approach towards an efficient and reliable network slicing
slicing,‖ NGMN, pp. 1–12, 2016. in 5G networks,‖ in 2019 IEEE 10th Annual Ubiquitous Computing,
Electronics & Mobile Communication Conference (UEMCON), 2019.
[25] S. Sharma et al., ―Secure authentication protocol for 5G enabled IoT
network,‖ in 2018 Fifth International Conference on Parallel, Distributed [46] Q. Liu, T. Han, and N. Ansari, ―Learning-assisted secure end-to-end
and Grid Computing (PDGC), 2018. network slicing for cyber-physical systems,‖ IEEE Netw., vol. 34, no. 3,
pp. 37–43, 2020.
[26] N. Alliance, ―Description of network slicing concept,‖ NGMN 5G P,
vol. 1, no. 1, pp. 1–11, 2016. [47] H. Sedjelmaci, ―Cooperative attacks detection based on artificial
intelligence system for 5G networks,‖ Comput. Electr. Eng., vol. 91, no.
[27] V. A. Cunha et al., ―Network slicing security: Challenges and 107045, p. 107045, 2021.
directions,‖ Internet Technol. Lett., vol. 2, no. 5, p. e125, 2019.
[48] V. P. Kafle, Y. Fukushima, P. Martinez-Julia, and T. Miyazawa,
[28] Z. Kotulski et al., ―Towards constructive approach to end-to-end slice
―Consideration on automation of 5G network slicing with machine
isolation in 5G networks,‖ EURASIP J. Inf. Secur., vol. 2018, no. 1,
learning,‖ in 2018 ITU Kaleidoscope: Machine Learning for a 5G Future
2018.
(ITU K), 2018.
[29] X. Foukas, G. Patounas, A. Elmokashfi, and M. K. Marina, ―Network
[49] A. Thantharate, R. Paropkari, V. Walunj, C. Beard, and P. Kankariya,
slicing in 5G: Survey and challenges,‖ IEEE Commun. Mag., vol. 55,
―Secure5G: A deep learning framework towards a secure network
no. 5, pp. 94–100, 2017.
392 | P a g e
Vol. 14, No. 12, 2023
slicing in 5G and beyond,‖ in 2020 10th Annual Computing and [60] Y. Shi, Y. E. Sagduyu, T. Erpek, and M. C. Gursoy, ―How to attack and
Communication Workshop and Conference (CCWC), 2020. defend NextG radio access network slicing with reinforcement
[50] How to attack and defend 5G radio access network slicing with learning,‖ IEEE Open J. Veh. Technol., vol. 4, pp. 181–192, 2023.
reinforcement learning. arXiv 2021. [61] D. Chowdhury, R. Das, R. Rana, A. D. Dwivedi, P. Chatterjee, and R. R.
[51] N. A. E. Kuadey, G. T. Maale, T. Kwantwi, G. Sun, and G. Liu, Mukkamala, ―AUTODEEPSLICE: A Data Driven Network Slicing
―DeepSecure: Detection of distributed denial of service attacks on 5G Technique of 5G network using Automatic Deep Learning,‖ in 2022
network slicing—deep learning approach,‖ IEEE Wirel. Commun. Lett., IEEE Globecom Workshops (GC Wkshps), 2022.
vol. 11, no. 3, pp. 488–492, 2022. [62] J. Wang and J. Liu, ―Secure and reliable slicing in 5G and beyond
[52] Highly accurate and reliable wireless network slicing in 5th generation vehicular networks,‖ IEEE Wirel. Commun., vol. 29, no. 1, pp. 126–
networks: a hybrid deep learning approach, Journal of Network and 133, 2022.
Systems Management: Springer, 2022. [63] L. Tang, Q. Tan, Y. Shi, C. Wang, and Q. Chen, ―Adaptive virtual
[53] Y. Shi and Y. E. Sagduyu, ―Adversarial machine learning for flooding resource allocation in 5G network slicing using constrained Markov
attacks on 5G radio access network slicing,‖ in 2021 IEEE International decision process,‖ IEEE Access, vol. 6, pp. 61184–61195, 2018.
Conference on Communications Workshops (ICC Workshops), 2021. [64] V. Sciancalepore, X. Costa-Perez, and A. Banchs, ―RL-NSB:
[54] C. Benzaid, T. Taleb, and J. Song, ―AI-based autonomic and scalable Reinforcement learning-based 5G network slice broker,‖ IEEE ACM
security management architecture for secure network slicing in B5G,‖ Trans. Netw., vol. 27, no. 4, pp. 1543–1557, 2019.
IEEE Netw., vol. 36, no. 6, pp. 165–174, 2022. [65] G. Zhou, L. Zhao, G. Zheng, Z. Xie, S. Song, and K. C. Chen, ―Joint
[55] S. Yoon et al., ―Moving target defense for in-vehicle software-defined Multi-objective Optimization for Radio Access Network Slicing Using
networking: IP shuffling in network slicing with multiagent deep Multi-agent Deep Reinforcement Learning,‖ IEEE Transactions on
reinforcement learning,‖ in Artificial Intelligence and Machine Learning Vehicular Technology, 2023.
for Multi-Domain Operations Applications II, 2020. [66] Y. Kim and H. Lim, ―Multi-agent reinforcement learning-based resource
[56] X. Cheng, Y. Wu, G. Min, A. Y. Zomaya, and X. Fang, ―Safeguard management for end-to-end network slicing,‖ IEEE Access, vol. 9, pp.
network slicing in 5G: A learning augmented optimization approach,‖ 56178–56190, 2021.
IEEE J. Sel. Areas Commun., vol. 38, no. 7, pp. 1600–1613, 2020. [67] O. A. Latif, M. Amer, and A. Kwasinski, ―Classification of network
[57] I. H. Abdulqadder and S. Zhou, ―SliceBlock: Context-aware slicing requests using support vector machine,‖ in 2022 International
authentication handover and secure network slicing using DAG- Conference on Electrical and Computing Technologies and Applications
blockchain in edge-assisted SDN/NFV-6G environment,‖ IEEE Internet (ICECTA), 2022.
Things J., vol. 9, no. 18, pp. 18079–18097, 2022. [68] P. Tam, S. Math, A. Lee, and S. Kim, ―Multi-agent deep Q-networks for
[58] E. Bandara, X. Liang, S. Shetty, R. Mukkamala, A. Rahman, and N. W. efficient edge federated learning communications in software-defined
Keong, ―Skunk-A Blockchain and Zero Trust Security Enabled IoT,‖ Comput. Mater. Contin., vol. 71, no. 2, pp. 3319–3335, 2022.
Federated Learning Platform for 5G/6G Network Slicing,‖ in 2022 19th [69] R. Li, C. Wang, Z. Zhao, R. Guo, and H. Zhang, ―The LSTM-based
Annual IEEE International Conference on Sensing, Communication, and advantage actor-critic learning for resource management in network
Networking (SECON), IEEE, 2022, pp. 109–117. slicing with user mobility,‖ IEEE Commun. Lett., vol. 24, no. 9, pp.
[59] S. Wijethilaka and M. Liyanage, ―A federated learning approach for 2005–2009, 2020.
improving security in network slicing,‖ in GLOBECOM 2022 - 2022
IEEE Global Communications Conference, 2022.
393 | P a g e

Paper 39-Machine Learning Based Secure 5G Network Slicing

Uploaded by

Copyright:

Available Formats

Paper 39-Machine Learning Based Secure 5G Network Slicing

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Paper 39-Machine Learning Based Secure 5G Network Slicing

Uploaded by

Copyright:

Available Formats

(IJACSA) International Journal of Advanced Computer Science and Applications,

Vol. 14, No. 12, 2023

Machine Learning-based Secure 5G Network Slicing:

Fig. 2. PRISMA flow diagram.

mitigating such attacks. A hybrid DL-based approach for

Where, ( ) is the posterior policy probability, Outputs:

● For each agent (‗i = 1‘ to ‗num_agents‘):

Initialization: g) Other potential solutions: To further improve the

You might also like