IEEE Consumer Electronics Magazine - May/June 2022 - 32

AI Security for MEC
technique is activated at each SDN's edge controller
tomonitor and detect the flow rule attack.
Hassan et al.6 proposed a distributed snortbased
intrusion detection system (IDS) to monitor
andprotect the cloud computing network and control
system from denial of service (DoS) attacks.
IDSs are activated at each server and collaborate
with each other in a distributed manner to detect
the distributed DoS attacks. They proposed a set
of static rules related to the number of packets
sent and dropped that should be conformed by
themonitored nodes; when a node does not follow
the rules, its behavior constitutes a DoS attack.
Cai et al.7 developed a distributed signaturebased
attack detection to protect the metering
infrastructure against false data injection attacks.
They modeled the behavior of the attacks targeting
the smart grid metering system with a set of
signatures. The distributed IDSs based on attack
signatures are activated at edge servers to detect
the false data injection, while a reputation system
is used to evaluate the trust level of the smart
grid collector systems.
Detecting Attacks With Dynamic Malicious
Behaviors
In this kind of detection, a machine learning
algorithm is used to monitor the dynamic misbehaviors
executed by the cyber-attack with the
goal to detect attacks that have never been
detected before by the signature-based detection
technique, such as the zero-day threats.
Chen et al.8 developed an accurate attack
detection framework based on a deep belief neural
network to protect the transportation system
empowered by MEC from internal and external
attacks. The unsupervised learning algorithm
aims to determine the new attack features of
threats on the MEC servers in order to improve
the detection rate. The attack models used during
the experimentation phase correspond to
the network attacks that change their misbehaviors
frequently and dynamically.
Samy et al.9 proposed a robust defense system
based on a deep learning algorithm to protect
the IoT devices from several types of zeroday
attacks. The defense system is activated at
each edge server to monitor the behavior of the
distrusted IoT devices. According to their experimental
results, their defense system exhibits a
32
low reaction time and high accuracy detection.
However, they did not propose a security strategy
to protect the defense system against the
internal and external attacks.
Subramaniam and Kaur10 studied different
deep learning algorithms deployed at the edge to
secure a MEC network. According to their investigation,
the automated learning attack features
used by the deep learning algorithms could
improve the detection rate significantly as compared
to other machine learning algorithms such
as random forest and one-class support vector
machine. However, they did not evaluate the performance
of the algorithms with real data.
Wang et al.11 analyzed the performance of the
current IDSs based on machine learning algorithms
deployed at the Internet of Things (IoT)
edge. The performance of machine learning is
evaluated in terms of three main metrics: detection
accuracy, memory storage, and complexity.
The purpose of this study is to embed the
selected machine learning algorithm(s) with the
IDS at the real edge server to secure the IoT
network.
Table 1 summarizes the pros and cons of the
edge security frameworks5-10 based on the following
security and network metrics: detection
and false positive rates, computational overhead,
and memory storage. The detection and
false positive rates correspond to the detection
accuracy of security frameworks against the
known and unknown attacks. The computational
overhead represents the required cost of the
security framework to achieve a high level of
security, while the memory storage metric corresponds
to the number of attack signatures
and features required to detect the attacks
accurately.
Among the AI detection techniques that
could be leveraged in the cyber security context
for achieving high detection accuracy, i.e., high
attack detection with a low false positive rate,
the hybrid learning technique3 seems promising.
This technique combines the rules (signatures)
defined by the cyber security expert and attack
models obtained from the machine learning algorithm
to increase the detection accuracy. In this
context, there is a need to better understand the
human-machine interaction in addressing the
accuracy of attack detection with consideration
IEEE Consumer Electronics Magazine
IEEE Consumer Electronics Magazine - May/June 2022

Table of Contents for the Digital Edition of IEEE Consumer Electronics Magazine - May/June 2022
