An Edge Computing-Based Preventive Framework With Machine Learning- Integration for Anomaly Detection and Risk Management in Maritime Wireless Communications

Abstract

The safety of maritime environments in context with effective and secure wireless communication networks is required for ships, coastal stations, and maritime authorities. The dynamic nature of marine environments, where ships traverse vast and unpredictable expanses of oceans and seas, presents big challenges to safety and risk management. Wireless communication technology is widely employed in maritime activities for communication via ocean networks and underwater wireless sensor networks (UWSNs). Maintaining the safety of the maritime environment, effective anomaly detection, prompt risk mitigation, and real-time communication becomes more difficult due to its dynamic nature. International trade and transportation are facilitated by the maritime industry. In addition to protecting lives and averting environmental disasters, maritime safety is important for maintaining the effectiveness and dependability of shipping routes. To handle the intricacies of maritime safety, this work proposes a novel preventive framework for anomaly detection and risk management in Maritime Wireless Communications (MWC). The proposed framework is based on edge computing and machine learning models. The framework makes use of edge computing technology to process data locally, lowering latency and enabling real-time communication in maritime environments. A proactive safety approach has been adopted to ensure the well-being of seafarers, safeguard vessels, and protect the marine environment. As maritime cybersecurity threats continue to evolve, the proposed research aims to enhance the cybersecurity posture of MWC. The framework will incorporate measures to detect and respond to potential cyber threats, ensuring the integrity and security of communication channels under international maritime cybersecurity standards. The proposed anomaly detection framework incorporates machine learning models such as Long Short-Term Memory (LSTM) and Isolation Forests (IF). The proposed framework also places a strong emphasis on preventative safety measures, including cybersecurity safeguards to protect communication channels in the constantly changing digital marine operations environment. To demonstrate the effectiveness of the proposed framework, the experiments were performed based on a publicly available dataset and implemented in the context of marine communications. The results show significant accuracy as well as high precision, recall, and F1-score metrics generated by the LSTM and IF models. The results highlight that the proposed framework can detect anomalies and potential threats in real-time marine communications.

Full text

VOLUME XX, 2017 1
Date of publication xxxx 00, 0000, date of current version xxxx 00, 0000.
Digital Object Identifier 10.1109/ACCESS.2017.Doi Number
An Edge Computing-based Preventive Framework with
Machine Learning-Integration for Anomaly Detection and
Risk Management in Maritime Wireless Communications
Abdulmohsen Algarni 1, Tayfun Acarer 2, and Zulfiqar Ahmad 3, *
1 Department of Computer Science, King Khalid University, Abha 61421, Saudi Arabia; a.algarni@kku.edu.sa
2 Maritime Transportation and Management Vocational School of Higher Education, Piri Reis University, Tuzla, ˙Istanbul 34940, Türkiye;
tacarer@hotmail.com
3 Department of Computer Science and Information Technology, Hazara University, Mansehra, KPK, Pakistan; zulfiqarahmad@hu.edu.pk
* Corresponding Author: Zulfiqar Ahmad (zulfiqarahmad@hu.edu.pk)
Acknowledgment: This research was financially supported by the Deanship of Scientific Research at King Khalid University under
research grant number (R.G.P.2/93/45).
ABSTRACT The safety of maritime environments in context with effective and secure wireless
communication networks is required for ships, coastal stations, and maritime authorities. The dynamic nature
of marine environments, where ships traverse vast and unpredictable expanses of oceans and seas, presents
big challenges to safety and risk management. Wireless communication technology is widely employed in
maritime activities for communication via ocean networks and underwater wireless sensor networks
(UWSNs). Maintaining the safety of the maritime environment, effective anomaly detection, prompt risk
mitigation, and real-time communication becomes more difficult due to its dynamic nature. International
trade and transportation are facilitated by the maritime industry. In addition to protecting lives and averting
environmental disasters, maritime safety is important for maintaining the effectiveness and dependability of
shipping routes. To handle the intricacies of maritime safety, this work proposes a novel preventive
framework for anomaly detection and risk management in Maritime Wireless Communications (MWC). The
proposed framework is based on edge computing and machine learning models. The framework makes use
of edge computing technology to process data locally, lowering latency and enabling real-time
communication in maritime environments. A proactive safety approach has been adopted to ensure the well-
being of seafarers, safeguard vessels, and protect the marine environment. As maritime cybersecurity threats
continue to evolve, the proposed research aims to enhance the cybersecurity posture of MWC. The framework
will incorporate measures to detect and respond to potential cyber threats, ensuring the integrity and security
of communication channels under international maritime cybersecurity standards. The proposed anomaly
detection framework incorporates machine learning models such as Long Short-Term Memory (LSTM) and
Isolation Forests (IF). The proposed framework also places a strong emphasis on preventative safety
measures, including cybersecurity safeguards to protect communication channels in the constantly changing
digital marine operations environment. To demonstrate the effectiveness of the proposed framework, the
experiments were performed based on a publicly available dataset and implemented in the context of marine
communications. The results show significant accuracy as well as high precision, recall, and F1-score metrics
generated by the LSTM and IF models. The results highlight that the proposed framework can detect
anomalies and potential threats in real-time marine communications.
INDEX TERMS Machine learning; intelligent systems; sustainable navigation, autonomous vessels, ship
safety management systems; maritime shipping and satellite technology; sensing and communication in
maritime; automatic identification system; edge computing, prevention of ship accidents.
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3387529
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/

VOLUME XX, 2017 9
I. INTRODUCTION
Over time, both the volume and value of the cargo carried
have increased. For this reason, the damages caused by
accidents in maritime transportation are also increasing. It is
not possible to define the cost of loss of life in monetary
terms during these accidents [1], [2], [3]. The dynamic and
ever-evolving field of maritime safety deals with the
particular difficulties presented by the enormous and
frequently inaccessible regions of the world's oceans and
seas [4]. Wireless communication networks are important
components for ships, coastal stations, and maritime
authorities to communicate effectively and securely [5]. In
this case, ensuring maritime safety entails harnessing
technology improvements, putting in place efficient
processes, and addressing particular wireless communication
issues. Reliable and instantaneous information exchange
between ships and land-based stations is made possible by
maritime wireless communication (MWC) [1]. It is worth
consideration in emergency response, coordination, and
navigation. To avoid mishaps and guarantee the general
safety of maritime operations, communication systems, such
as satellite communication and Very High Frequency/Ultra
High Frequency (VHF/UHF) radio are required to be reliable
[6]. The Automatic Identification System (AIS) greatly
enhances maritime safety by giving vessels real-time
information on the positions, courses, and speeds of other
adjacent ships [7], [8], [9], [10]. This helps with navigation
and collision avoidance, enabling ships to modify their paths
to avoid mishaps and guarantee safe travel. MWC has been
used in emergencies for search and rescue operations.
Distress signals guarantee a prompt reaction from maritime
authorities and other vessels. They are typically sent by
Digital Selective Calling (DSC) on VHF radios [11], [12].
Navigators can plan routes that avoid bad weather conditions
since wireless connectivity makes it easier to receive weather
reports on time. With the ability to provide precise and
current meteorological information, this capability is
required for minimizing weather-related mishaps and
maximizing maritime safety [13], [14]. Maintaining
communication system cybersecurity has significant
importance because maritime communication is growing
more digital and depends on wireless technologies. The
integrity and safety of marine activities are preserved by
preventing illegal access, manipulation, and disruptions to
communication channels through the use of cybersecurity
preventions [10], [15], [16], [17]. Integrating Internet of
Things (IoT) devices and sensor networks with MWC
improves safety by facilitating the real-time data gathering
and transfer of environmental factors, equipment status, and
vessel conditions. Preventive and proactive risk management
are also important factors to enhance maritime safety [3], [6],
[18]. The International Maritime Orgazation (IMO) have
declared the “urgent need to raise awareness on cyber-risk
threats and vulnerabilities, to support safe and secure
shipping, which is operationally resilient to cyber-risks
encouraging administrations to ensure that cyber-risks are
appropriately addressed in safety management systems”
[19], [20]. New developments in wireless communication
technology offer better connectivity, lower latency, and more
capacity. One such development is the implementation of 5G
networks in maritime environments. Because they enable
increasingly complex applications like remote monitoring,
augmented reality navigation, intelligent systems and
autonomous vessel operations [21], [22].
By moving computation and data storage closer to the point
of data generation, edge computing has the ability to
significantly improve MWC [15], [23]. This lowers latency,
enhances real-time processing, and facilitates effective
communication in maritime situations. By processing data at
or close to the network's edge, edge computing reduces the
time it takes to send information to a centralized cloud server
and get a response. In marine environments, lower latency is
required to maintain communication system efficiency
through quick decision-making. The common features of
MWC include real-time data sharing such as the interchange
of navigational data, weather reports, and vessel positions
[12]. With the help of edge computing, such type of data can
be processed locally, facilitating speedy analysis and
decision-making without being exclusively dependent on
distant cloud servers. Edge computing allows information to
be filtered and aggregated locally at the edge before being
sent to the central cloud as given in Figure 1. Marine wireless
communication systems are required to be benefited from the
redundancy and robustness provided by the edge computing.
Edge nodes can function independently even in the event of
a disruption in the connection to the central cloud,
guaranteeing the dependability of essential communication
tasks. IoT devices and sensors are frequently deployed in
maritime environments to gather data. By offering a
distributed computing architecture, edge computing has the
ability to enhance the environment of marine communication
and makes it more extensive and decentralized [24], [25],
[26], [27].
In MWC, machine learning models can be implemented for
risk management and anomaly detection [28], [29], [30].
Isolation Forests (IF) [1], One-Class Support Vector
Machine (SVM) [31], Autoencoders [32], and LSTM [33]
are some of the examples of machine learning methods that
can be used for anomaly detection and risk management in
MWC. Isolation forests perform better for finding
abnormalities and outliers in data and it can be used to detect
unusual network behaviors or anomalous communication
patterns. One-Class Support Vector Machine is specifically
designed for datasets with few abnormalities, and it can be
used to create a model of typical communication patterns.
Autoencoders can learn a condensed representation of
communication patterns. They perform better for
encapsulating intricate connections in communication data.
LSTMs can be used in marine wireless communications to
evaluate transmission sequences over time and spot patterns
or variations that point to abnormalities or possible threats
[34], [35].
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3387529
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/

VOLUME XX, 2017 9
Figure 1: Edge computing-based data processing in maritime communication environment
A. RESEARCH MOTIVATION
Safety and risk management in maritime environments are
always challenging due to its changing situations as ships are
traveling over huge, unpredictably sized stretches of oceans
and seas. For communication over ocean networks and
underwater wireless sensor networks (UWSN), wireless
communication technology is extensively used in maritime
activities [23], [29]. However, it is challenging to maintain
the safety, efficient anomaly detection, timely risk
mitigation, and real-time communication in maritime
environments. The maritime industry supports international
trade and transportation. Maritime safety plays a significant
role in preserving the efficiency and dependability of
shipping routes, saving lives, and preventing environmental
disasters [3], [6]. Due to the potential for wireless
communication anomalies and hazards to endanger public
safety, it is imperative to develop a preventive framework.
With the increasing digitization of maritime communication,
new cybersecurity challenges arise. Cyberattacks have the
potential to compromise the integrity of communications and
endanger maritime safety [30], [36], [37], [38], [39]. A
preventive framework combining edge computing and
machine learning methods would be the better is solution for
real-time cybersecurity risk detection and mitigation. The
application of edge computing completely transforms marine
safety. Relocating computation closer to the data source
allows edge computing to speed up reaction times, reduce
latency, and enable real-time analytics [27]. Machine
learning algorithms have demonstrated efficacy in
interpreting complex datasets and identifying patterns that
indicate anomalies. Applying machine learning to the
maritime sector provide a proactive approach to safety
through risk prediction and prevention. Bandwidth
limitations are common in wireless communications for
marine applications [21], [26], [40]. The proposed
preventive system maximizes bandwidth consumption with
local data processing and filtering at the edge. This is driven
by the need to minimize data transfer to central servers to
maintain bandwidth and ensure efficient communication,
particularly in distant marine areas. The proposed preventive
framework is in line with the increasing focus on utilizing
cutting-edge technologies to both fulfill and surpass safety
requirements of international marine organizations.
B. MAIN CONTRIBUTIONS
Main contributions of this paper are briefed as follows:
 Development of a novel edge computing-based
preventive framework for MWC.
 Integration of machine learning methods to the
proposed preventive framework for providing a
robust and efficient solution for anomaly detection
and risk management in maritime environments.
 Implementation of a proactive safety approach,
ensuring the well-being of seafarers, safeguarding
vessels, and protecting the marine environment.
 Implementation of customized machine learning
methods to detect real-time anomalies in MWC.
 To acknowledge and adopt the unique challenges of
the maritime environment including its dynamic
conditions, limited connectivity, and the need for
resilience.
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3387529
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/

VOLUME XX, 2017 9
 As maritime cybersecurity threats continue to
evolve, the research aims to enhance the
cybersecurity posture of MWC. The framework
will incorporate measures to detect and respond to
potential cyber threats, ensuring the integrity and
security of communication channels in accordance
with international maritime cybersecurity
standards.
C. ORGANIZATION OF THE PAPER
The rest of the paper is organized as follows. Section 2
presents the related work in context with maritime safety,
edge computing technology and implementation of machine
learning models in situations similar with maritime
environments. Section 3 presents the system design and
model. Section 4 provides ‘an edge computing-based
preventive framework for anomaly detection and risk
management in MWC’. Section 5 presents the performance
evaluation method. Section 6 provides experiments, results
and discussions. Section 7 concludes the article with future
directions.
II. RELATED WORK
This section reviews the related work by categorizing it into
the techniques used for maritime safety, edge computing
technology and the machine learning approaches used for
anomaly detection and risk management in maritime
intelligent systems.
The ability of future Maritime Autonomous Surface Ships to
recognize possible threats and respond appropriately are
important for their safety. The study in [4] set out to
determine the research paths for the most important safety
indicators in the three operational safety-sensitive areas of
Maritime Autonomous Surface Ships: communication, intact
stability, and collision avoidance. The findings show that
many academics agree that operational leading safety
indicators are necessary, and occasionally they even offer
recommendations for the indicators' specific composition
[4]. Several prominent safety markers for self-navigating
boats are easily recognized in scholarly works and applied in
contemporary operations.
Spaceborne Synthetic Aperture Radar (SSAR) and
Automatic Identification System (AIS) are being used in a
variety of research projects for applications that support
marine security and safety. However, it is necessary to open
a separate parenthesis for AIS among these systems. Because
AIS devices automatically detect other AIS devices within
the coverage area and receive their navigation information
such as position, speed, route, etc. It also sends information
about itself to surrounding ships and Coast Radio Stations
(CRS) [41]. In densely populated shipping areas, the data
association becomes further challenging as ships seen using
SSAR imaging may be mistakenly linked to AIS
observations. This frequently leads to an inaccurate or
erroneous impression of the marine environment. In order to
categorize ship kinds in SSAR imagery, a classification-
aided data association strategy is developed [42] that makes
use of a transfer learning mechanism. In particular, AIS data
is used to train a ship categorization model, which is
subsequently applied to forecast SSAR ship detections.
These predictions are then applied to the data association
process, which creates a strong match between the data by
using a rank-ordered assignment technique. Based on the
types of SSAR products utilized for maritime surveillance,
two case studies in the UK are used to assess the
effectiveness of the classification-aided data association
technique: targeted data association in the Solent and wide-
area and large-scale data association in the English Channel
[42]. The results demonstrate a high degree of correlation
between the data that is resistant to heavy shipping or traffic,
and the use of class (i.e., ship type) information increases the
trust in the data linkage.
The creation of a virtual training tool for marine safety
education is discussed in [43]. A diverse team made up of
business developers, VR experts, computer scientists, and
maritime specialists built this solution. The technology is a
portable, reasonably priced maritime training system that
may be utilized at home, in training facilities, or even on
board. When an officer has time for training, using VR-
training programs to improve situation awareness in
navigation is a simple and effective approach to practice.
This can be accomplished in an enjoyable and efficient
manner, providing quantifiable training progress indices.
The necessity of VR training for the shipping sector, its
obstacles, and the proof-of-concept using MarSEVR
(Maritime Safety Education with VR) technology are all
highlighted in [43]. This primary goal of the study is to
demonstrate a technology prototype that can be used to
provide immersive training scenarios for experts and
learners.
In [44], the authors examine possible solutions to the
problems related to the implementation of edge computing
for autonomous vehicles. The ultimate challenge of
designing an edge computing ecosystem for autonomous
vehicles is to provide sufficient computing capacity,
redundancy, and security to ensure the safety of autonomous
vehicles including maritime intelligent systems. In
particular, autonomous driving systems are highly
sophisticated, tightly integrating a wide range of
technologies, such as sensing, localization, perception, and
decision-making, in addition to seamless cloud platform
interactions for the creation of high-definition (HD) maps
and data storage [44]. Autonomous driving edge computing
systems must process massive amounts of data in real time,
with extremely diverse incoming input from many sensors.
Edge computing systems with autonomous driving
capabilities frequently have very stringent energy
consumption limitations since they are mobile. Therefore, it
is essential to provide adequate processing power while
maintaining a reasonable energy consumption to ensure the
safety of autonomous vehicles—even when they are
traveling at high speeds. Second, vehicle-to-everything
(V2X) relieves severe performance and energy constraints
on the edge side and offers redundancy for autonomous
driving workloads besides the edge system design [44].
By offering a unified platform for networking, processing,
and storage resources, edge computing makes it possible to
analyze data quickly and effectively close to its source [45].
As a result, the industrial Internet of things now uses it as its
foundational platform (IIoT). But the special qualities of
computing have also brought up new security issues. In [45],
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3387529
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/

VOLUME XX, 2017 9
an edge computing-based blockchain-based identity
management and access control method is devised to address
the issue. To achieve network entity registration and
authentication, self-certified cryptography is used. The
authors created a blockchain-based identity and certificate
management system and connect the implicit certificate that
is generated to its identity. Second, a Bloom filter-based
access control system is created and linked with identity
management.
The majority of container transportation from the Republic
of Korea, Republic of China, Japan and Singapore to other
nations or continents, both by land and sea, takes more than
a week. Cargoes in reefer containers need to be maintained
at the proper temperatures in such an environment. These
containers must be inspected daily to prevent cargo spoiling
throughout the lengthy navigation period. However, because
they are dispersed throughout the distinct cargo holding area
of a ship, it can be challenging to frequently check and
maintain them with the limited crew on board. A reefer
container monitoring system that can gather data from the
device sensors and makes use of the current onboard power
line(s) is suggested as a solution to such a scenario in [46].
One benefit of this system is that it doesn't require more
wiring, which means shipping companies may put it on their
ships at a minimal expense.
Ship groundings frequently result in damages, such as oil
spills, flooding, and eventual capsizing of the ship. Risks can
be assessed objectively by analyzing statistics on maritime
traffic or qualitatively by consulting experts. In [34], the
authors propose a big data analytics approach to assess
grounding risk in actual environmental settings. The
technique uses nowcast data, bottom depth data from the
General Bathymetric Chart of the Oceans (GEBCO), and
massive data streams from the Automatic Identification
System (AIS). In shallow water, the evasive maneuvers of
passenger ships functioning in the role of Ro-Pax are
modeled in a variety of traffic patterns that correspond to
side- or forward-grounding scenarios. As a result, to detect
possible grounding scenarios, an Avoidance Behavior-based
Grounding Detection Model (ABGD-M) is presented, and
the grounding probability risk is measured at observation
stations along ship routes in different voyages. The technique
is used aboard a Ro-Pax ship that sails the Gulf of Finland
during a 2.5-year ice-free season. The findings show that
depending on the operational conditions, observation
stations, and trip routes, grounding probabilistic risk
estimation can take many different forms.
In [16], a deep reinforcement learning (DRL) system for
autonomous ship collision avoidance in continuous action
spaces is presented. With the help of dynamic ship data, the
obstacle zone by target (OZT) algorithm calculates the
potential collision region. Agents of DRL use a virtual sensor
known as the grid sensor to identify the approach of many
ships. Agents used the Imazu problem, a series of
hypothetical ship contact scenarios, to learn collision
avoidance movements. In [16], the authors provided a novel
DRL-based collision avoidance strategy with a greater safe
passing distance. The authors created a brand-new technique
called inside OZT, which extends OZT and boosts learning
consistency. Using the long short-term memory (LSTM)
cell, the authors redesigned the network and trained in
continuous action spaces to create a model that has a longer
safe distance than the one investigated previously.
The growing globalization of navigation and the
dehumanization of ships have led to a mismatch between the
growing need for oversight of ship behavior and the scarce
resources of traffic services. This has resulted in a high
frequency of maritime accidents [1]. An essential component
of marine transportation is the observation of unusual ship
behavior. The automatic identification system (AIS) is
widely utilized in the management of ship static information
and the real-time transmission of dynamic information due
to the growing popularity of the system and increased marine
research. The authors identified abnormal ship behavior
from the perspective of spatial information and thematic
information based on moving ship trajectory data, taking into
account the state of abnormal ship behavior research at the
time. For this reason, the cognition of aberrant ship behavior
was first modeled in [1]. The authors identified and
explained the anomalous behavior indicated by ship thematic
data using the isolation forest algorithm. The experimental
findings demonstrate the effectiveness of the methodology
this research proposes for identifying anomalous ship
behavior.
Reliable vessel trajectory forecasting is necessary for
managing and controlling maritime traffic. Precise vessel
trajectory prediction not only helps prevent collisions but
also shortens sailing distances, improves navigation
efficiency, and helps design navigation routes. In the
maritime industry, vessel trajectory prediction using
automated identification system (AIS) data has so garnered
significant attention [7]. Because original AIS data may
contain noise, its use in actual maritime traffic management
is limited. This paper [7] suggests a vascular trajectory
prediction technique that combines a deep learning
prediction model with data denoising in order to solve this
issue. Three phases are involved in this process to achieve
data denoising: trajectory separation, data denoising, and
standardization. The moving average approach is used to
further clean up the data once outliers from the initial AIS
data samples are eliminated. Eventually, the denoised data
are standardized into uniformly distributed time-series data.
Next, the use of bidirectional long short-term memory (Bi-
LSTM) is used to forecast vessel trajectory.
III. SYSTEM DESIGN AND MODEL
In this research work, we propose an edge computing-based
preventive framework for anomaly detection and risk
management in MWC through machine learning methods to
enhance maritime safety as shown by Figure 2. The
framework unifies cutting-edge technologies i.e., edge
computing and machine learning to tackle the ever-changing
problems brought up by the maritime environments. By
integrating edge computing, data processing has been
localized, which lowers latency and makes real-time analysis
at the information source possible. This is especially used for
making quick decisions during emergencies or avoiding
collisions. By filtering and aggregating data locally, the
framework maximizes bandwidth use and provides an
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3387529
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/

VOLUME XX, 2017 9
affordable solution for maritime areas where bandwidth is
scarce.
Figure 2: An Optimized Edge Computing-based Preventive Framework with Machine Learning-Integration for Anomaly
Detection and Risk Management in Maritime Wireless Communications
The system gains anomaly detection capabilities from the
integration of customized machine learning models i.e.,
LSTM and Isolation Forests. By guaranteeing the prompt
detection of anomalous patterns and possible hazards, this
proactive safety framework enhances maritime safety. The
framework makes cybersecurity a top priority by processing
sensitive data locally via edge computing, which reduces the
possibility of tampering or unwanted access during data
transfer. The main components of the proposed framework
are given as follows:
A. MARITIME ENVIRONMENT
The maritime environment presents intricate challenges in
the field of maritime safety including poor connectivity,
unpredictable weather, and the requirement for robust
communication systems [47]. Let S represents the maritime
environment characterized by vastness and unpredictability
oceans and seas then challenges in maritime safety (MS)
attributed by expansive nature of S is given in equation 1.
𝑀𝑆
= 𝐶𝑜𝑚𝑝𝑙𝑒𝑥𝑖𝑡𝑦 (𝑆)
× 𝑈𝑛𝑝𝑟𝑒𝑑𝑖𝑐𝑡𝑎𝑏𝑖𝑙𝑖𝑡𝑦 (𝑆) 𝐸𝑞. 1
Communication infrastructure (CI) required to cover
maximum distance efficiently within S to ensure reliable and
immediate information sharing and it is represented by
equation 2.
𝐶𝐼=𝐸𝑓𝑓𝑖𝑐𝑖𝑒𝑛𝑡(𝑆)×𝑃𝐶+𝑈𝑊
+ 𝑅𝐶𝑆 𝐸𝑞. 2
Where, PC represents poor connectivity, UW represents
unpredictable weather and RCS represents robust
communication system.
The foundation of maritime communication is provided by
conventional techniques like satellite communication and
Very High Frequency/Ultra High Frequency (VHF/UHF)
radio [47]. Maritime communication (MC) is based on the
shipping and satellite communication (SC) and VHF/UHF
and is given by equation 3.
𝑀𝐶= 𝑆𝐶+ 𝑉𝐻𝐹/𝑈𝐻𝐹 𝐸𝑞. 3
The framework acknowledges that innovation is necessary to
address enduring problems like cybersecurity, bandwidth
optimization, and dependability. The proposed framework
uses edge computing to recognize the need to process data
closer to its source. It will achieve low-latency necessary for
navigation, emergency response, and collision avoidance.
The proposed framework with incorporation of edge
computing to address the cybersecurity, bandwidth
optimization and dependability is represented by Eq. 4.
𝑃𝐹𝑊= 𝐸𝐶× 𝐼𝑛𝑛𝑜𝑣𝑎𝑡𝑖𝑜𝑛 (𝑆)+𝐶𝑆+𝐵𝑂
+ 𝐷 𝐸𝑞. 4
Where, PFW represents the proposed framework, Innovation
(S) is innovation in maritime environment, CS is
cybersecurity, BO is bandwidth optimization and D is
dependability. The framework presents a proactive safety
paradigm through machine learning methods that enables the
real-time detection and remediation of anomalies and
possible hazards in marine wireless communications.
B. MARITIME WIRELESS COMMUNICATIONS (MWC)
Oceans and seas are large and challenging environments in
which MWC are utilized to enable safe, effective, and
coordinated activities. These communication systems are
essential to ships, coastal stations, and marine authorities
because they enable sensitive operations like emergency
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3387529
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/

VOLUME XX, 2017 9
response, navigation, coordination, and information
exchange. A variety of technologies and protocols are
available in the MWC space, with the goal of addressing the
unique challenges posed by the constantly shifting maritime
environment. Mathematically the challenging environment
of maritime is employed by the equation 5.
𝑆: 𝑀𝑊𝐶
→{𝑆𝑎𝑓𝑒, 𝐸𝑓𝑓𝑒𝑐𝑡𝑖𝑣𝑒, 𝐶𝑜𝑜𝑟𝑖𝑛𝑑𝑎𝑡𝑒𝑑 𝐴𝑐𝑡𝑖𝑣𝑖𝑡𝑖𝑒𝑠} 𝐸𝑞. 5
Where, S represents the unpredictable and vast oceans and
seas and MWC represents maritime wireless
communication. One of the fundamental components of
wireless communication in the maritime environment is
satellite communication. Satellites in orbit around the Earth
give global coverage, allowing ships to communicate
anywhere they are, even over very long distances. This
technology is necessary for weather reporting, maritime
safety, coordination, and reliable long-range
communication. Because UHF and VHF radio
communication technologies are so effective at
communicating over short to medium distances, they are
widely used in maritime settings [6]. In particular, VHF radio
is a widely used tool for ship-to-ship and ship-to-shore
communication. Applications such as navigation, distress
signals, and maritime coordination make extensive use of it.
The AIS aims to enhance situational awareness and prevent
collisions. Real-time data transmission, including position,
and speed, is facilitated by AIS transponder-equipped ships.
Ships are used to be interact with one another and reduce the
likelihood of collisions through such type of data exchange.
In maritime environments, the usage of Wireless Sensor
Networks (WSNs) for data collection and real-time
monitoring is increasing exponentially. These networks
consist of sensors placed on boats, buoys, or even the ocean
floor to collect data. Wireless transmission of such data
facilitates timely decision-making. MWC required to remain
connected even in inclement weather due to bandwidth
limitations, possible cyberattacks, and other issues. The
integration of edge computing and machine learning to
address such issues is represented by equation 6.
𝑀𝑊𝐶: 𝐸𝐶 + 𝑀𝐿→ 𝐶𝑆 + 𝐵𝑂+ 𝑅𝐴𝐷 𝐸𝑞. 6
Where, EC represents edge computing, ML represents
machine learning, CS represent cybersecurity, BO represent
bandwidth optimization and RAD represents real-time
anomaly detection. Innovations like the integration of edge
computing and machine learning are meant to help solve
these problems by improving cybersecurity, optimizing
bandwidth, and facilitating real-time anomaly detection. In
MWC, edge computing improves real-time processing by
processing data closer to its source, lowering latency. Edge
computing can play a key role in supporting time-sensitive
applications in MWC, such as navigation and emergency
response.
C. CYBER-PHYSICAL ATTACKS
MWC are used for the functioning of various systems within
the maritime industry, including navigation, control,
monitoring, and communication. As per Maritime Cyber
Attack Database (MCAD), Figure 3 shows the numbers and
regions of known cyber-attacks in Maritime [48]. Cyber-
physical attacks on MWC have serious consequences,
potentially leading to navigation errors, communication
breakdowns, and compromised safety. Following are the
major types of cyber-physical attacks that are used to target
MWC:
 GPS Spoofing: GPS spoofing involves sending
false signals to a Global Positioning System (GPS)
receiver, making it believe it is located at a different
location. Ships relying on GPS for navigation are
misdirected, leading to collisions, grounding, or
navigation into restricted areas. Implementation of
signal authentication and deploying secure GNSS
receivers with anti-spoofing capabilities mitigate
the risk of GPS spoofing.
 Jamming: Jamming attacks involve the intentional
interference with radio frequency signals,
disrupting normal communication. Maritime
communication systems, including distress signals
and navigation communication, are rendered
ineffective, affecting safety and operational
efficiency. Employing frequency-hopping
techniques, utilizing spread spectrum modulation,
and deploying anti-jamming antennas ensure
reliability in martime communication systems.
 Eavesdropping: The unlawful interception of
wireless communications in order to get private
data. There can be security hazards if confidential
information, like cargo data or navigation plans, is
compromised. Implementation of end-to-end
encryption protocols, use of secure communication
channels, and protection of cargo data and
navigation plans from unauthorized access are the
major preventive measures against eavesdropping.
 Man-in-the-Middle (MitM) Attacks: In order to
intercept and maybe modify a message, attackers
place themselves in the middle of the
communication channel. Attackers tamper with
communications sent between a ship and a shore
station, spreading false information or gaining
unapproved authority. Strong cryptographic
protocols, like Transport Layer Security (TLS) and
mutual authentication between the ship and the
shore station, protect the integrity of
communication and stop people from getting to
send data without permission.
 Rogue Access Points: Unauthorized wireless access
points installed with the intention of gaining access
to communication networks. Attackers breach the
communication network to take over important
systems or pilfer private data. Implementation of a
wireless intrusion detection system to monitor for
unauthorized access points, as well as regular
network scans to detect and remove rogue devices,
can prevent unauthorized access or data theft.
 Malware and Software Exploitation: Introduction
of harmful software or taking advantage of holes in
communication software are examples of malware
and software exploitation. The integrity of
communication systems can be compromised by
malware, which might result in illegal access or
interfere with regular operations. Regularly
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3387529
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/

VOLUME XX, 2017 9
updating software and firmware to patch
vulnerabilities and implementing robust
antimalware safeguards the integrity of
communication systems and prevents unauthorized
access or disruption of operations.
 Denial of Service (DoS) Attacks: Overloading
networks or systems to prevent them from
functioning normally. Limiting the availability of
vital communication services, which has an impact
on safety, control, and navigation systems.
Implementation of network traffic monitoring and
filtering mechanisms to detect and mitigate
abnormal traffic patterns ensures the reliability of
vital communication services for safety, control,
and navigation systems.
 Firmware Manipulation: Tampering with the
firmware of communication devices or systems.
Manipulating firmware leads to unauthorized
control over communication systems, allowing
attackers to interfere with maritime operations.
Using secure boot mechanisms to verify the
firmware integrity during startup and only allowing
authorized sources to update the firmware prevents
unauthorized users from taking control of
communication systems and ensures the safety of
maritime operations.
Figure 3: The numbers and regions of known cyber-attacks in Maritime [48]
D. PROBLEM FORMULATION
The dynamic and expansive features of maritime
environments, characterized by large and frequently
unreachable oceans and seas, present distinct problems for
the maritime safety area. Wireless communication networks
are the core components of coastal stations, and maritime
authorities to communicate securely and effectively. MWC
system is used for emergency response, coordination, and
navigation; however, several issues including bandwidth
optimization, cybersecurity, and dependability still exist.
Marine organizations rely on communication methods, such
as satellite communication and VHF/UHF radio, there is an
increased requirement for reliable and fast information
sharing. By giving real-time information on vessel positions,
courses, and speeds, technologies like the AIS improve
safety by making navigation and collision avoidance easier.
In order to prevent accidents and guarantee the general safety
of maritime activities, the dynamic maritime environment
necessitates constant innovation in communication systems.
Mathematically, maritime communication system at time t is
represented by Equation 7.
𝐶 (𝑡)=󰇯
𝐵 (𝑡)
𝐶𝑆 (𝑡)
𝑅 (𝑡)
󰇰 Eq. 7
Where, C represents maritime communication system, B
represents bandwidth optimization, CS represents
cybersecurity and R represents risk management. Maritime
communication system in context with risk management at
time t is represented by Equation 8
𝑅 (𝑡)=󰇯
𝐸𝑅 (𝑡)
𝐶𝑂 (𝑡)
𝑁 (𝑡)
󰇰 Eq. 8
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3387529
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/

VOLUME XX, 2017 9
Where, ER represents emergency response, CO represents
coordination and N represents Navigation of marine
communication network at time t. Let MWC (t) be the matrix
representing maritime wireless communication methods at
time t then mathematically it is represented by Equation 9.
Each row represents the communication method at time t and
each column represents the risk management aspects
including reliability, anomaly detection and bandwidth
optimization, in maritime communication networks.
𝑀𝑊𝐶 (𝑡)=
⎣
⎢
⎢
⎢
⎢
⎢
⎡
𝑀𝑊𝐶,(𝑡)
𝑀𝑊𝐶,(𝑡)
.
..
𝑀𝑊𝐶,(𝑡)
𝑀𝑊𝐶,(𝑡)
𝑀𝑊𝐶,(𝑡)
.
..
𝑀𝑊𝐶,(𝑡)
…
…
…
…
…
…
𝑀𝑊𝐶,(𝑡)
𝑀𝑊𝐶,(𝑡)
.
..
𝑀𝑊𝐶,(𝑡)
⎦
⎥
⎥
⎥
⎥
⎥
⎤
Eq. 9
The objective function of safety in maritime activities is
represented by 𝑆 (𝑡) and it is a function “f” of sum of C(t),
R(t) and MWC (t) as given in Equation 10.
𝑆 (𝑡)=𝑓 {𝐶 ( 𝑡)× 𝑅 ( 𝑡)× 𝑀𝑊𝐶 ( 𝑡)}

  𝐸𝑞. 10
The components C(t), R(t), and MWC(t) are the independent
factors collectively contributing to the overall safety function
for maritime environments. The function ‘f’ represents the
overall performance in terms of weight obtained by adding
the products of all the three components for ‘n’ number of
maritime environments. The multiplicative effect represents
that all the factors are equally important, and deficiencies in
any one factor significantly impact the overall safety
performance.
Cybersecurity challenges are raised by the growing
digitization of maritime communication, which calls for
precautions against unauthorized access, channel
manipulation, and disruption. Real-time data collection on
environmental variables, equipment status, and vessel
conditions is made easier by the integration of IoT devices
and sensor networks with marine wireless communication.
Enhancing maritime safety requires both proactive and
preventive risk management, especially considering recent
advancements in wireless communication technologies, such
as the possible deployment of 5G networks in maritime
contexts. These developments enable the adoption of
sophisticated applications like remote monitoring and
autonomous vessel operations by providing better
connectivity, reduced latency, and expanded capacity.
The proposed research aims to create a novel preventive
framework based on edge computing to address the above-
mentioned issues. The goal of this framework is to apply
machine learning techniques to risk management and
anomaly detection in MWC. Edge computing improves real-
time processing, reduces latency, and maximizes bandwidth
use by relocating computation closer to the data source.
Machine learning models i.e., LSTM and IF are used for real-
time anomaly identification and risk management.
IV. AN EDGE COMPUTING-BASED PREVENTIVE
FRAMEWORK WITH MACHINE LEARNING-
INTEGRATION FOR ANOMALY DETECTION AND
RISK MANAGEMENT IN MARITIME WIRELESS
COMMUNICATIONS
This study proposes a novel approach to improve the safety
MWC by using the edge computing in combination with
machine learning models for anomaly detection and risk
management. Edge computing processes and analyses MWC
data in close proximity to its source, which includes
communication nodes or sensors accountable for data
transmission and collection. The integration of edge
computing into MWC for the purpose of anomaly detection
and risk management presents a multitude of benefits. These
include the ability to conduct analyses in real time, reduced
latency, and effective allocation of resources. This is
accomplished by means of the implementation of
interconnected smart devices. Edge computing enables the
prompt assessment of risk management and, consequently,
expedites the detection of anomalies. It also prevents the
transmission of sensitive data across networks and
guarantees its localization, thus mitigating the potential for
data breaches and safeguarding user privacy. The data
analysis process is further optimized by the edge devices
performing preprocessing operations directly. Machine
learning models are hosted on high-performance edge
devices, providing immediate feedback regarding anomaly
detection and risk management. IF and LSTM machine
learning models have been customized in context with
maritime environments and are used to find abnormalities
and possible hazards in the data. This makes it possible to
proactively mitigate or stop adverse situations.
Algorithm 1 shows the process of the proposed
framework in which an edge computing
technology has been integrated with machine
learning methods for anomaly detection and risk
management to enhance maritime safety. The
algorithm presents a novel preventive approach for risk
management and anomaly. It implements machine learning
techniques integrated with edge computing technology. The
initial parameters are the number of edge computing nodes,
the duration of the data processing window, and the total
number of features in the dataset. The system collected data
from sources of marine wireless communication. The edge
computing nodes process the data locally and make it ready
for anomolies detection. The machine learning models are
integrated for risk management and anomaly detection. The
processed data is used to train LSTM and IF models, which
give the system the capacity to identify abnormalities and
evaluate the risks associated with them. Each edge
computing node uses the learned IF and LSTM models to
analyze its processed data in real-time anomaly detection,
alerting users or initiating preventive measures when
abnormalities are detected. Risk evaluations are carried out
according to the degree of irregularities found. The
processed data, anomaly alarms, and risk evaluations are the
outcomes that are shared with central monitoring stations or
other appropriate authorities.
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3387529
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/

VOLUME XX, 2017 9
Algorithm 1: An Edge Computing-Based Preventive Framework with Machine Learning-Integration for
Anomaly Detection and Risk Management in Maritime Wireless Communications
Input: MWC: Maritime Wireless Communications
Output: AR-MWC: Anomaly detection and Risk management in MWC
Procedure: Anomaly Detection & Risk Management (MWC)
1. Parameter Initialization
i. N: Number of edge computing nodes
ii. T: Time stamp for data processing
iii. K: Number of features in the dataset
2. Data Collection
i. MWSN: Data collection from maritime WSN
ii. 𝐷𝑖: Store the collected data in a local buffer
3. Edge Computing Processing
i. for each computing node 𝑖
𝑃𝑖 = Local_Processing (𝐷𝑖)
4. Machine Learning Model Integration
i. Train and deploy machine learning methods
ii. 𝑀 = 𝑇𝑟𝑎𝑖𝑛𝐼𝑠𝑜𝑙𝑎𝑡𝑖𝑜𝑛𝐹𝑜𝑟𝑒𝑠𝑡 (𝑃,𝑃,…,𝑃)
iii. 𝑀 = 𝑇𝑟𝑎𝑖𝑛𝐿𝑆𝑇𝑀 (𝑃,𝑃,…,𝑃)
5. Real-time anomaly detection
i. For each edge computing node 𝑖
𝐴() =𝐴𝑝𝑝𝑙𝑦𝐼𝑠𝑜𝑙𝑎𝑡𝑖𝑜𝑛𝐹𝑜𝑟𝑒𝑠𝑡 (𝑃,𝑀)
𝐴() =𝐴𝑝𝑝𝑙𝑦𝐿𝑆𝑇𝑀 (𝑃,𝑀)
ii. If anomalies detected on node 𝑖
𝑀𝑎𝑟𝑖𝑡𝑖𝑚𝑒_𝑆𝑎𝑓𝑒𝑡𝑦=𝑇𝑟𝑖𝑔𝑔𝑒𝑟 (𝑃𝑟𝑒𝑣𝑒𝑛𝑡𝑖𝑣𝑒𝐴𝑐𝑡𝑖𝑜𝑛𝑠, 𝐴𝑙𝑒𝑟𝑡, 𝑆𝑎𝑓𝑒𝑡𝑦𝑀𝑒𝑎𝑠𝑢𝑟𝑒𝑠)
6. Risk Management
i. Assess the severity of detected anomalies and determine risk management
𝑅() =𝐴𝑠𝑠𝑒𝑠𝑠𝑅𝑖𝑠𝑘 (𝑅())
𝑅() =𝐴𝑠𝑠𝑒𝑠𝑠𝑅𝑖𝑠𝑘 (𝑅())
7. Result Generation
i. Assess the severity of detected anomalies and determine risk management
𝐴𝑅 − 𝑀𝑊𝐶 = 𝐷𝑒𝑡𝑒𝑐𝑡𝑒𝑑 𝑎𝑛𝑜𝑚𝑎𝑙𝑖𝑒𝑠 𝑎𝑛𝑑 𝑠𝑢𝑏𝑠𝑒𝑞𝑢𝑒𝑛𝑡 𝑟𝑖𝑠𝑘 𝑚𝑎𝑛𝑎𝑔𝑒𝑚𝑒𝑛𝑡
8. Return 𝑨𝑹−𝑴𝑾𝑪
V. EXPERIMENTS, RESULTS & DISCUSSIONS
We perform the simulations and evaluate the performance of
proposed framework in context with the safety of MWC.
A. EVALUATION METRICS
We use the evaluation metrics of Accuracy, Precision, Recall
and F1 score for evaluating the performance of the proposed
framework [49]. These values are calculated based on the
following terms [38].
 True Positives (TP): The number of tuples that are
really found to be intrusive at the end of the process.
In the proposed framework, TP has been used for
correctly identification of risky situation in
anomaly detection and risk management.
 True Negatives (TN): The number of valid tuples
that are found at the end of the detecting process. In
the proposed framework, TN has been used for
correctly identification of the situation as not being
risk in anomaly detection and risk management.
 False Positives (FP): The number of safe tuples that,
at the conclusion of the detection process, are
identified as intrusions. In the proposed framework,
FP has been used for identification of unnecessary
alerts or action, potentially causing disruptions in
anomaly detection and risk management.
 False Negatives (FN): The quantity of dangerous
tuples that, at the conclusion of the detection
process, are found normally. In the proposed
framework, FN has been used for identification of
situation where a system fails to detect an actual
risk in anomaly detection and risk management.
Accuracy is a frequently employed metric for evaluating the
performance of classification models, especially in binary
classification tasks. By calculating the percentage of
accurately predicted instances among all the instances in the
dataset, it evaluates the overall accuracy of the model's
predictions [38]. Mathematically, it is calculated with the
help of Equation 11.
𝐴 = 𝑇𝑃+𝑇𝑁
𝑇𝑃+𝑇𝑁+𝐹𝑃+𝐹𝑁 𝐸𝑞. 11
Precision is a metric used to assess the efficacy of a
classification model. It measures the percentage of true
positive predictions among all positive predictions, or true
positives plus false positives, in order to assess the model's
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3387529
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/

VOLUME XX, 2017 9
accuracy in making positive predictions [38].
Mathematically, it is represented by Equation 12.
𝑃𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛= 𝑇𝑃
𝑇𝑃+𝐹𝑃 𝐸𝑞. 12
Recall is a metric used to assess a classification model's
performance. It is sometimes referred to as Sensitivity or
True Positive Rate. The model's recall quantifies its capacity
to accurately identify each and every positive case in the
dataset [38]. Mathematically, it is given by Equation 13.
𝑅𝑒𝑐𝑎𝑙𝑙= 𝑇𝑃
𝑇𝑃+𝐹𝑁 𝐸𝑞. 13
The F1 score is a way to measure how well classification
models work, especially when they are asked to choose
between two options. When there is an imbalance between
precision and recall, the F1 score becomes helpful [38].
Mathematically, it is calculated with the help of Equation 14.
𝐹1 𝑆𝑐𝑜𝑟𝑒
= 2 × 𝑃𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛 × 𝑅𝑒𝑐𝑎𝑙𝑙
𝑃𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛+𝑅𝑒𝑐𝑎𝑙𝑙 𝐸𝑞. 14
B. DATASET
In order to evaluate the proposed framework, a publically
available dataset on a Kaggle website with title, “WSN-DS:
A dataset for intrusion detection systems in wireless sensor
networks” [50][51] has been used. The dataset replicates
many Denial-of-Service (DoS) attacks on WSN using the
Low Energy Adaptive Clustering Hierarchy (LEACH)
protocol. It includes Blackhole, Grayhole, Flooding, and
Scheduling attacks, which are four different categories of
attacks. In the Blackhole attack, at the beginning of a round,
an attacker assumes the identity of a Cluster Head (CH).
When nodes connect to this fake CH, they unintentionally
submit their data packets to it, which are then transmitted to
the Base Station (BS). As with the Blackhole attack,
attackers assume the identity of CHs in the Grayhole assault.
These attackers do this on the basis of the sensitivity of the
data included in the packets they drop or delete. The goal of
the flooding attack is to flood the network with too many
high-transmission-power advertising CH messages. The
scheduling attack takes place in the setup stage of the
LEACH protocol. Assuming the role of CHs, attackers
provide every node the same time slot for data transmission,
which causes packet collisions and eventual data loss.
The proposed study focuses on wireless communications in
maritime environments, and while there are similarities
between maritime and conventional wireless
communications, we understand the necessity of addressing
potential differences in attack vectors and environmental
factors. We have carefully aligned the attacks considered in
the proposed study dataset with those commonly
encountered in maritime scenarios. Despite the existence of
frequency range variations between maritime and
conventional wireless communications, we clarify that the
proposed framework remains independent of these
considerations.
C. EXPERIMENTAL DESIGN
The evaluation of the models integrated within the proposed
framework was conducted using "WSN-DS," a dataset
specifically designed for intrusion detection systems in
wireless sensor networks. There are two parts to the dataset:
the training set and the test set. Eighty percent of all the
records in the dataset were in the training set. While the test
set was 20% of all the records. The "cross_val_score"
function from scikit-learn was used to perform cross-
validation on LSTM. For IF, on the other hand, we used a
train-test split with an 80:20 ratio to make sure the model
was correct, since cross-validation does not work for
unsupervised learning. All of the tests are performed in
Python on a GPU-based system with a CPU speed of 1.8
GHz and 16 GB of RAM. The pre-configured machine
learning packages and libraries have been used: Numpy,
Seaborn, LabelEncoder, OneHoTencoding, Pandas, and
Matplotlib.
D. RESULTS AND EVALUATION
The experiments were performed by implementing two
machine learning methods i.e., LSTM and IF. The evaluation
results for each model is given below:
1) LSTM
The LSTM model is used to detect the anomalies with
Python libraries and modules including Pandas and Sklearn.
Table 1 shows the results in the form of classification report
generated by LSTM on a given dataset which is further
visualized in Figure 4. The results show that the framework
performs well in differentiating between "Normal" and
"Anomaly" cases. A variety of attacks that affect WSNs are
simulated by the dataset. These attacks include flooding,
scheduling, blackhole, grayhole, and flooding, all of which
have distinct challenges for detection. The LSTM model
achieves more than 0.90 as values of recall, precision, and
F1-score for both classes. The model performed well in
recognizing cases classified as "Normal," obtaining nearly
flawless precision, recall, and F1-score. With an accuracy of
0.95, recall of 0.91, and an F1-score of 0.93, the model
demonstrated a great performance even though it was
marginally less accurate in classifying "Anomaly"
occurrences.
Table 1: Classification report of results generated by LSTM
Precession
Recall
F1
-
Score
Accuracy
Normal
0.99 1.00 0.99
0.99
Anomaly
0.95 0.91 0.93
Macro Average
0.97 0.95 0.96
Weighted Average
0.99 0.99 0.99
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3387529
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/

VOLUME XX, 2017 9
Figure 4: Classification report of LSTM
The 99% overall accuracy with support of 74931 number of
actual instances highlights how well the framework works to
find intrusions in the simulated WSN environment. With
excellent results in terms of precision, recall, and F1-score
metrics, the weighted averages and macros further confirm
the models' resilience. These findings imply that the
proposed framework is a good fit for the job of anomaly
detection in WSNs of marine environment. It demonstrates
its dependability in recognizing typical network activity and
its ability to distinguish anomalies even when complex
attack techniques like flooding, scheduling, blackhole, and
grayhole attacks are present. The efficacy of the framework
in guaranteeing the security and integrity of maritime
wireless sensor networks is attributed to the amalgamation
of LSTM model.
Figure 5 shows the comparison of training and validation
loss. A satisfactory fit is indicated by a steady reduction and
plateau in loss during both training and validation. Since, the
validation loss closely tracks the training loss and does not
rise, there is no indication of overfitting. An additional
indication that the model is not overfitting is the convergence
of the training and validation loss curves.
Figure 6 shows the comparison of training and validation
accuracy. An encouraging sign is that the accuracy for both
training and validation is high and exhibits a similar trend.
The model performs well on the validation data in addition
to matching the training data well, indicating strong
generalization. At times, the validation accuracy even
marginally outperforms the training accuracy. There is a
regularization impact of dropout during training and the
same is not present during the validation set evaluation.
Figure 7 shows the confusion matrix generated by LSTM. To
identify the risky and normal situations correctly and
incorrectly, we divide the predictions in four categories i.e.,
TP, TN, FP and FN. The model detected 67733 instances as
TP, 6276 instances as TN, 300 instances as FP and 622
instances as FN. The high proportion of true positives shows
that the model detects normal cases with 99% precision. The
low percentage of normal instances being wrongly classified
as anomalies indicates the small number of false positives
(300). The model displays a greater quantity of false
negatives (622), signifying situations where real anomalies
are wrongly categorized as normal. Despite this, the overall
performance is strong, as evidenced by the 99% accuracy.
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3387529
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/

VOLUME XX, 2017 9
Figure 5: Training and validation loss
Figure 6: Training and validation accuracy
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3387529
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/

VOLUME XX, 2017 9
Figure 7: Confusion matrix generated by LSTM
2) ISOLATION FOREST (IF)
The IF model is used to detect the anomalies with Python
libraries and modules including Pandas and Sklearn. Table 2
shows the results in the form of classification report
generated by IF on a given dataset which is further visualized
in Figure 8. The classification report shows that a dataset
included by two classes, represented by the numbers "-1" and
"1." The IF model shows an accuracy of 0.39 for the
anomalous class ("-1"), meaning that only 39% of the cases
identified as anomalies were actual anomalies. This class has
a somewhat greater recall (0.72), meaning that 72% of real
anomalies were correctly detected by the model. For the
anomalous class, the F1-score is reported as 0.50, indicating
a trade-off between memory and precision. The model
exhibits a high precision of 0.97 for the normal class ("1"),
meaning that 97% of the occurrences predicted as normal
were in fact normal. With a recall of 0.88 for the normal
class, 88% of real normal occurrences were correctly
identified. The normal class F1-score is 0.92, indicating a
performance that strikes a balance between recall and
precision. The overall accuracy is stated to be 87%, and
metrics that are weighted and macro-averaged offer more
information about how well it performs. The related
weighted averages are 0.91, 0.87, and 0.89, but the macro-
averaged precision, recall, and F1-score are 0.68, 0.80, and
0.71, respectively. These findings imply that although the
model does a great job of classifying typical cases, it has the
ability do a better job of accurately identifying anomalies.
Table 2: Classification report of results generated by IF
Precession
Recall
F1
-
Score
Accuracy
Normal (1)
0.97
0.88
0.92
0.87
Anomaly (
-
1)
0.39
0.72
0.50
Macro Average
0.68
0.80
0.71
Weighted Average
0.91
0.87
0.89
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3387529
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/

VOLUME XX, 2017 9
Figure 8: Classification report of IF
Figure 9: Confusion matrix generated by IF
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3387529
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/

VOLUME XX, 2017 9
Figure 9 shows the confusion matrix generated by IF. The
model accurately classified 24,906 cases as anomalies ("-1")
and 30,505 instances as part of the normal class ("1"). False
positive errors 9689 were made by it, identifying occurrences
as normal while in fact they belonged to the anomaly class.
False negative errors of 39,561 were found, showing cases
where normal occurrences were wrongly classified as
anomalies. The confusion matrix shows how well the model
differentiates between normal and anomaly cases.
The utilization of edge computing instead of centralized
servers for machine learning-based analysis itself presents
various benefits, such as instantaneous analysis, reduced
latency, maintenance of privacy, and effective resource
management. Since edge computing allows processing and
analysis of data directly at or near the source of data
generation, it provides immediate assessment of anomalies,
facilitating prompt decisions. The potential challenges in
implementing the proposed framework in live maritime
environments include cost overhead, hardware limitations,
data management, and real-time analysis through high-
performance computational resources. Edge devices
typically have limited computational resources compared to
centralized servers. Careful consideration of resource
utilization is required to handle large-scale data or varied
types of sensor inputs while maintaining adequate
performance.
VI. CONCLUSION AND FUTURE WORK
This study proposes a preventive framework for risk
management and anomaly detection in MWC. The
framework uses edge computing and machine learning to
handle the difficulties in safety presented by large and
usually inaccessible maritime environments. Data
processing is localized through the integration of edge
computing, which lowers latency and permits real-time
analysis at the information source. This is important when it
comes to making quick judgments in an emergency or
preventing mishaps. By filtering and aggregating data
locally, the framework maximizes bandwidth use and offers
a cost-effective solution for maritime locations with
restricted bandwidth. The incorporation of machine learning
models including LSTM and IF, gives the system the ability
to detect anomalies. The framework places a high priority on
cybersecurity by using edge computing to process sensitive
data locally, which lowers the possibility of data transfer
manipulation or unauthorized access. It recognizes
established lines of communication such as satellite
communication and VHF/UHF radio, guaranteeing
consistent and timely information sharing in maritime
operations. The usefulness of the proposed framework is
demonstrated by the experimental findings, which are based
on a publicly accessible dataset that simulates attacks on
WSN in marine communications. The capacity of proposed
framework is to identify anomalies and possible threats in
real-time and is demonstrated by the high precision, recall,
and F1-score metrics displayed by both the LSTM and IF
models. Overall, the results reveal that the LSTM model,
with an accuracy of 99%, outperformed the IF model.
In the future, we aim to explore algorithms with a hybrid
nature to improve performance in risk assessment and
anomaly identification with more advanced attack types. We
also intend to implement fog computing-based fuzzy logic
systems to optimize the performance of 5G communication
technology in the context of maritime communications.
ACKNOWLEDGMENT
This research was financially supported by the Deanship of
Scientific Research at King Khalid University under research
grant number (R.G.P.2/93/45).
REFERENCES
[1] Y. Shi, C. Long, X. Yang, and M. Deng, “Abnormal Ship
Behavior Detection Based on AIS Data,” Appl. Sci., vol.
12, no. 9, p. 4635, May 2022, doi: 10.3390/app12094635.
[2] G. Kodak and T. Acarer, “Evaluation of the effect of
maritime traffic regulations on the accident rate in the
strait of Istanbul,” Aquat. Res., vol. 4, no. 2, pp. 181–207,
2021, doi: 10.3153/AR21015.
[3] O. Eulaerts and G. Joanny, Weak signals in border
management and surveillance technologies. 2022. doi:
10.2760/784388.
[4] K. Wróbel, M. Gil, P. Krata, K. Olszewski, and J.
Montewka, “On the use of leading safety indicators in
maritime and their feasibility for Maritime Autonomous
Surface Ships,” Proc. Inst. Mech. Eng. Part O J. Risk
Reliab., vol. 237, no. 2, pp. 314–331, Apr. 2023, doi:
10.1177/1748006X211027689.
[5] N. Nomikos, P. K. Gkonis, P. S. Bithas, and P. Trakadas,
“A Survey on UAV-Aided Maritime Communications:
Deployment Considerations, Applications, and Future
Challenges,” IEEE Open J. Commun. Soc., vol. 4, pp. 56–
78, 2023, doi: 10.1109/OJCOMS.2022.3225590.
[6] T. Acarer, “Denizcilik Araştırmaları Dergisi : Amfora
Journal of Maritime Research : Amphora VHF Kısa
Mesafe Deniz Haberleşmesinin Data İletişimine
Dönüşmesinin Deniz İşletmelerinin Gemi Yönetimleri
İçin Temin Edeceği Olanaklar Possibilities to be Provided
by Maritim,” no. 3, pp. 15–31, 2023.
[7] C.-H. Yang, C.-H. Wu, J.-C. Shao, Y.-C. Wang, and C.-
M. Hsieh, “AIS-Based Intelligent Vessel Trajectory
Prediction Using Bi-LSTM,” IEEE Access, vol. 10, pp.
24302–24315, 2022, doi:
10.1109/ACCESS.2022.3154812.
[8] A. Siddiqa et al., “A survey of big data management:
Taxonomy and state-of-the-art,” J. Netw. Comput. Appl.,
vol. 71, pp. 151–166, Aug. 2016, doi:
10.1016/j.jnca.2016.04.008.
[9] Z. Ahmad, T. Acarer, and W. Kim, “Optimization of
Maritime Communication Workflow Execution with a
Task-Oriented Scheduling Framework in Cloud
Computing,” J. Mar. Sci. Eng., vol. 11, no. 11, p. 2133,
Nov. 2023, doi: 10.3390/jmse11112133.
[10] K. Tran, S. Keene, E. Fretheim, and M. Tsikerdekis,
“Marine Network Protocols and Security Risks,” J.
Cybersecurity Priv., vol. 1, no. 2, pp. 239–251, Apr. 2021,
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3387529
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/

VOLUME XX, 2017 9
doi: 10.3390/jcp1020013.
[11] A. A. Periola, A. A. Alonge, and K. A. Ogudo,
“Architecture and System Design for Marine Cloud
Computing Assets,” Comput. J., vol. 63, no. 6, pp. 927–
941, Jun. 2020, doi: 10.1093/comjnl/bxz169.
[12] S. A. H. Mohsan, Y. Li, M. Sadiq, J. Liang, and M. A.
Khan, “Recent Advances, Future Trends, Applications and
Challenges of Internet of Underwater Things (IoUT): A
Comprehensive Review,” J. Mar. Sci. Eng., vol. 11, no. 1,
p. 124, Jan. 2023, doi: 10.3390/jmse11010124.
[13] K. Sathish, R. C. Venkata, R. Anbazhagan, and G. Pau,
“Review of Localization and Clustering in USV and AUV
for Underwater Wireless Sensor Networks,” Telecom, vol.
4, no. 1, pp. 43–64, Jan. 2023, doi:
10.3390/telecom4010004.
[14] K. Saeed et al., “A Comprehensive Analysis of Security-
Based Schemes in Underwater Wireless Sensor
Networks,” Sustainability, vol. 15, no. 9, p. 7198, Apr.
2023, doi: 10.3390/su15097198.
[15] S. Fattah, A. Gani, I. Ahmedy, M. Y. I. Idris, and I. A.
Targio Hashem, “A Survey on Underwater Wireless
Sensor Networks: Requirements, Taxonomy, Recent
Advances, and Open Research Challenges,” Sensors, vol.
20, no. 18, p. 5393, Sep. 2020, doi: 10.3390/s20185393.
[16] R. Sawada, K. Sato, and T. Majima, “Automatic ship
collision avoidance using deep reinforcement learning
with LSTM in continuous action spaces,” J. Mar. Sci.
Technol., vol. 26, no. 2, pp. 509–524, Jun. 2021, doi:
10.1007/s00773-020-00755-0.
[17] F. Al-Quayed, Z. Ahmad, and M. Humayun, “A Situation
Based Predictive Approach for Cybersecurity Intrusion
Detection and Prevention Using Machine Learning and
Deep Learning Algorithms in Wireless Sensor Networks
of Industry 4.0,” IEEE Access, vol. 12, pp. 34800–34819,
2024, doi: 10.1109/ACCESS.2024.3372187.
[18] L. Drazovich, L. Brew, and S. Wetzel, “Advancing the
State of Maritime Cybersecurity Guidelines to Improve the
Resilience of the Maritime Transportation System,” in
2021 IEEE International Conference on Cyber Security
and Resilience (CSR), IEEE, Jul. 2021, pp. 503–509. doi:
10.1109/CSR51186.2021.9527922.
[19] I. de la Peña Zarzuelo, “Cybersecurity in ports and
maritime industry: Reasons for raising awareness on this
issue,” Transp. Policy, vol. 100, pp. 1–4, Jan. 2021, doi:
10.1016/j.tranpol.2020.10.001.
[20] M. S. Karim, “Maritime cybersecurity and the IMO legal
instruments: Sluggish response to an escalating threat?,”
Mar. Policy, vol. 143, p. 105138, Sep. 2022, doi:
10.1016/j.marpol.2022.105138.
[21] M. Abbasi, E. Mohammadi-Pasand, and M. R. Khosravi,
“Intelligent workload allocation in IoT–Fog–cloud
architecture towards mobile edge computing,” Comput.
Commun., vol. 169, no. January, pp. 71–80, 2021, doi:
10.1016/j.comcom.2021.01.022.
[22] I. A. Alablani and M. A. Arafah, “EE-UWSNs: A Joint
Energy-Efficient MAC and Routing Protocol for
Underwater Sensor Networks,” J. Mar. Sci. Eng., vol. 10,
no. 4, p. 488, Apr. 2022, doi: 10.3390/jmse10040488.
[23] M. Jahanbakht, W. Xiang, L. Hanzo, and M. Rahimi
Azghadi, “Internet of Underwater Things and Big Marine
Data Analytics—A Comprehensive Survey,” IEEE
Commun. Surv. Tutorials, vol. 23, no. 2, pp. 904–956,
2021, doi: 10.1109/COMST.2021.3053118.
[24] S. Muthuramalingam, A. Bharathi, S. Rakesh kumar, N.
Gayathri, R. Sathiyaraj, and B. Balamurugan, “Iot based
intelligent transportation system (iot-its) for global
perspective: a case study,” in Intelligent Systems Reference
Library, vol. 154, 2019, pp. 279–300. doi: 10.1007/978-3-
030-04203-5_13.
[25] Z. Lv and W. Xiu, “Interaction of Edge-Cloud Computing
Based on SDN and NFV for Next Generation IoT,” IEEE
Internet Things J., vol. 7, no. 7, pp. 5706–5712, 2020, doi:
10.1109/JIOT.2019.2942719.
[26] F. M. Talaat, “Effective prediction and resource allocation
method (EPRAM) in fog computing environment for
smart healthcare system,” Multimed. Tools Appl., vol. 81,
no. 6, pp. 8235–8258, 2022, doi: 10.1007/s11042-022-
12223-5.
[27] L. U. Khan, I. Yaqoob, N. H. Tran, S. M. A. Kazmi, T. N.
Dang, and C. S. Hong, “Edge-Computing-Enabled Smart
Cities: A Comprehensive Survey,” IEEE Internet Things
J., vol. 7, no. 10, pp. 10200–10232, 2020, doi:
10.1109/JIOT.2020.2987070.
[28] N. Peppes, E. Daskalakis, T. Alexakis, E. Adamopoulou,
and K. Demestichas, “Performance of Machine Learning-
Based Multi-Model Voting Ensemble Methods for
Network Threat Detection in Agriculture 4.0,” Sensors,
vol. 21, no. 22, p. 7475, Nov. 2021, doi:
10.3390/s21227475.
[29] Q. Gang, A. Muhammad, Z. U. Khan, M. S. Khan, F.
Ahmed, and J. Ahmad, “Machine Learning-Based
Prediction of Node Localization Accuracy in IIoT-Based
MI-UWSNs and Design of a TD Coil for Omnidirectional
Communication,” Sustainability, vol. 14, no. 15, p. 9683,
Aug. 2022, doi: 10.3390/su14159683.
[30] I. Alrashdi, A. Alqazzaz, E. Aloufi, R. Alharthi, M. Zohdy,
and H. Ming, “AD-IoT: Anomaly detection of IoT
cyberattacks in smart city using machine learning,” in
2019 IEEE 9th Annual Computing and Communication
Workshop and Conference, CCWC 2019, 2019, pp. 305–
310. doi: 10.1109/CCWC.2019.8666450.
[31] V. Balasubramaniam, “Artificial Intelligence Algorithm
with SVM Classification using Dermascopic Images for
Melanoma Diagnosis,” J. Artif. Intell. Capsul. Networks,
vol. 3, no. 1, pp. 34–42, Mar. 2021, doi:
10.36548/jaicn.2021.1.003.
[32] Y. Song, S. Hyun, and Y.-G. Cheong, “Analysis of
Autoencoders for Network Intrusion Detection,” Sensors,
vol. 21, no. 13, p. 4294, Jun. 2021, doi:
10.3390/s21134294.
[33] G. Rjoub, J. Bentahar, O. Abdel Wahab, and A. Saleh
Bataineh, “Deep and reinforcement learning for automated
task scheduling in large-scale cloud computing systems,”
Concurr. Comput. , no. May, 2020, doi: 10.1002/cpe.5919.
[34] M. Zhang, P. Kujala, and S. Hirdaris, “A machine learning
method for the evaluation of ship grounding risk in real
operational conditions,” Reliab. Eng. Syst. Saf., vol. 226,
p. 108697, Oct. 2022, doi: 10.1016/j.ress.2022.108697.
[35] M. F. Elrawy, A. I. Awad, and H. F. A. Hamed, “Intrusion
detection systems for IoT-based smart environments: a
survey,” J. Cloud Comput., vol. 7, no. 1, pp. 1–20, 2018,
doi: 10.1186/s13677-018-0123-6.
[36] U. AlHaddad, A. Basuhail, M. Khemakhem, F. E. Eassa,
and K. Jambi, “Ensemble Model Based on Hybrid Deep
Learning for Intrusion Detection in Smart Grid Networks,”
Sensors, vol. 23, no. 17, p. 7464, Aug. 2023, doi:
10.3390/s23177464.
[37] M. A. Al-Shareeda, S. Manickam, S. A. Laghari, and A.
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3387529
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/

VOLUME XX, 2017 9
Jaisan, “Replay-Attack Detection and Prevention
Mechanism in Industry 4.0 Landscape for Secure
SECS/GEM Communications,” Sustainability, vol. 14, no.
23, p. 15900, Nov. 2022, doi: 10.3390/su142315900.
[38] A. Kumari et al., “AI-Empowered Attack Detection and
Prevention Scheme for Smart Grid System,” Mathematics,
vol. 10, no. 16, pp. 1–18, 2022, doi:
10.3390/math10162852.
[39] S. Kumar and R. R. Mallipeddi, “Impact of cybersecurity
on operations and supply chain management: Emerging
trends and future research directions,” Prod. Oper.
Manag., vol. 31, no. 12, pp. 4488–4500, Dec. 2022, doi:
10.1111/poms.13859.
[40] J. Singh, P. Singh, and S. S. Gill, “Fog computing: A
taxonomy, systematic review, current trends and research
challenges,” J. Parallel Distrib. Comput., vol. 157, no.
June, pp. 56–85, 2021, doi: 10.1016/j.jpdc.2021.06.005.
[41] T. ACARER, “ENDÜSTRİ’DEKİ GELİŞMELERİN
DENİZCİLİK İŞLETMELERİNE AİT GEMİLERİN
YÖNETİMİNDE TEMİN ETTİĞİ YENİ OLANAKLAR
VE İNSANSIZ GEMİLER,” Mersin Üniversitesi
Denizcilik ve Lojistik Araştırmaları Derg., Oct. 2023, doi:
10.54410/denlojad.1364567.
[42] M. Rodger and R. Guida, “Classification-Aided SAR and
AIS Data Fusion for Space-Based Maritime Surveillance,”
Remote Sens., vol. 13, no. 1, p. 104, Dec. 2020, doi:
10.3390/rs13010104.
[43] E. Markopoulos, J. Lauronen, M. Luimula, P. Lehto, and
S. Laukkanen, “Maritime Safety Education with VR
Technology (MarSEVR),” in 2019 10th IEEE
International Conference on Cognitive
Infocommunications (CogInfoCom), IEEE, Oct. 2019, pp.
283–288. doi:
10.1109/CogInfoCom47531.2019.9089997.
[44] S. Liu, L. Liu, J. Tang, B. Yu, Y. Wang, and W. Shi, “Edge
Computing for Autonomous Driving: Opportunities and
Challenges,” Proc. IEEE, vol. 107, no. 8, pp. 1697–1716,
Aug. 2019, doi: 10.1109/JPROC.2019.2915983.
[45] Y. Ren, F. Zhu, J. Qi, J. Wang, and A. K. Sangaiah,
“Identity Management and Access Control Based on
Blockchain under Edge Computing for the Industrial
Internet of Things,” Appl. Sci., vol. 9, no. 10, p. 2058, May
2019, doi: 10.3390/app9102058.
[46] J.-H. Huh, “Reefer container monitoring system using
PLC-based communication technology for maritime edge
computing,” J. Supercomput., vol. 76, no. 7, pp. 5221–
5243, Jul. 2020, doi: 10.1007/s11227-019-02864-z.
[47] T. ACARER and L. YILMAZ, GMDSS - Restricted Radio
Operator License - Global Maritime Distress and Safety
System. ACADEMY PUBLICATIONS , 2015. [Online].
Available: https://www.nadirkitap.com/gmdss-kisitli-
telsiz-operator-ehliyeti-kuresel-denizcilik-tehlike-ve-
guvenlik-sistemi-tayfun-acarer-levent-yilmaz-
kitap14280766.html
[48] NHL Stenden University of Applied Sciences, “Maritime
Cyber Attack Database (MCAD).” Accessed: Dec. 29,
2023. [Online]. Available:
https://www.nhlstenden.com/en/maritime-cyber-attack-
database
[49] W. M. S. Yafooz, Z. B. A. Bakar, S. K. A. Fahad, and A.
M. Mithon, Business Intelligence Through Big Data
Analytics, Data Mining and Machine Learning, vol. 1016.
2020. doi: 10.1007/978-981-13-9364-8_17.
[50] I. Almomani, B. Al-Kasasbeh, and M. Al-Akhras, “WSN-
DS: A Dataset for Intrusion Detection Systems in Wireless
Sensor Networks,” Journal of Sensors. Accessed: Nov. 01,
2023. [Online]. Available:
https://www.kaggle.com/datasets/bassamkasasbeh1/wsnd
s
[51] I. Almomani, B. Al-Kasasbeh, and M. AL-Akhras, “WSN-
DS: A Dataset for Intrusion Detection Systems in Wireless
Sensor Networks,” J. Sensors, vol. 2016, pp. 1–16, 2016,
doi: 10.1155/2016/4731953.
Dr. Abdulmohsen Algarni
received the Ph.D. degree from
the Queensland University of
Technology, Australia, in
2012. He was a Research
Associate with the School of
Electrical Engineering and
Computer Science,
Queensland University of
Technology, in 2012. He is
currently an Associate
Professor with the College of
Computer Science, King Khalid University. His research
interests include artificial intelligence, data mining, text
mining, machine learning, information retrieval, and
information filtering.
Dr. Tayfun Acarer received the
B.S. degree in electronics
engineering from Istanbul
Technical University, Istanbul,
Turkey, in 1980, the M.S. degree
from Istanbul University,
Istanbul, Turkey, in 1992, the
Ph.D. degree from Istanbul
University, Istanbul, Turkey, in
1995, and Asst. Prof. Dr. degree
from Bilgi University, Istanbul,
Turkey, in 2016. He has worked
at different companies in the ICT Sector. His last job was
Chair of the ICT Regulatory Body (ICTA). Since 2018, he
has been with the National Metrology Institute (UME),
Gebze, Turkey, where he is currently a Board member of the
National Metrology Institute. He currently lectures on
Information Technologies and Marine Communication and
Electronics Navigation Systems at six different universities.
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3387529
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/

VOLUME XX, 2017 9
Dr. Zulfiqar Ahmad has
completed his PhD in Computer
Science from Department of
Computer Science & Information
Technology, Hazara University,
Mansehra, Pakistan in 2022. He
has received his MSc (Computer
Science) degree with distinction
from Hazara University,
Mansehra, Pakistan in 2012 and
MS (CS) degree from
COMSATS University,
Abbottabad, Pakistan in 2016. He is the author of several
publications in the field of Fog computing, Cloud
computing, high performance computing, and Scientific
Workflows execution and management. His research areas
include Scientific Workflow Management in Cloud
Computing, Internet of Things, Fog Computing, Edge
Computing, Cybersecurity, and Wireless Sensor Networks
(WSNs).
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3387529
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/