Comprehensive Survey on Detection of Anomalies in Edge Computing

Network and Deep Learning Solutions

Sonali Jadhav

and Dr. Arun Kulkarni

Department of Computer Engineering, Thadomal Shahani Engineering College, University of Mumbai, Mumbai, India

Department of Information Technology, Thadomal Shahani Engineering College, University of Mumbai, Mumbai, India

Keywords: Edge Computing, Deep Learning, Edge Network Anomalies, DNN, GAN.

Abstract: Edge computing is an innovative computing model that plays an essential role in offering faster computation,

increased security, and lower transfer costs to the production applications that run on collection of IoT

devices, sensors, and the network. Today, IoT has become extensively utilized technology in a variety of

applications, including medical, industrial, agriculture, transport, manufacturing, surveillance and so on. Edge

computing involves processing a variety of data closer to its source, allowing for faster processing rates over

the large volumes and more effective outcomes in real time. However, with the increasing number and

complexity of IoT devices in Edge networks, as well as the never-ending accumulation of network data,

monitoring in Edge networks and identifying abnormal network behavior is getting increasingly challenging.

Anomaly detection is a significant issue that has received extensive attention across all range of disciplines

and application do-mains. Deep Learning is a branch of machine learning that deals with approaches inspired

by the structure and function of artificial neural networks which can be used to solve many real-world

problems. The objective of this research paper is to survey and illuminate the role of deep learning techniques

in tackling the edge computing anomalies and provides an extensive overview on deep Learning techniques

used for detecting them. The research work also covers the case study on anomaly detection over edge

computing using deep neural network (DNN) and generative adversarial network (GAN) models.

INTRODUCTION

The Internet of Things (IoT) consists of large number

of small devices or networks of sensors that

continuously generate large volumes of data. Due to

the limited computing and memory capabilities, raw

data is typically sent to centralized systems or the

cloud for analysis. Later, machine learning (ML) and

deep learning (DL) algorithms could be deployed on

edge devices, making it possible to bring intelligence

to the Internet of Things [28].

In Edge computing platform, various anomalies are

existing and represent a significant threat to the

security, performance, and reliability of

communication networks. The process of monitoring

business networks for unusual behavior is known as

network behavior anomaly detection [1]. As is widely

recognized, flows make up the data that is sent

through a network. There is always a temporal link

https://orcid.org/0009-0001-3731-3823

https://orcid.org/0009-0008-7699-8272

between flows, and anomalous traffic in the network

is no different [2]. Anomalies can arise from

various sources including malicious attacks, software

bugs, misconfigurations, and hardware failures.

Traditional anomaly detection methods such as rule-

based systems, statistical analysis, and machine

learning algorithms have limitations in effectively

identifying and mitigating these anomalies due to

their static nature and inability to adapt to evolving

threats, with the advent of deep learning techniques,

there has been a paradigm shift in anomaly detection

approaches. The ability of deep learning algorithms to

automatically derive hierarchical representations

from raw data, have shown promising results in

detecting complex and previously unseen network

anomalies. Deep learning uses a

complicated architecture or combination of non-

linear transformations to achieve high level

Jadhav, S. and Kulkarni, A.

Comprehensive Survey on Detection of Anomalies in Edge Computing Network and Deep Lear ning Solutions.

DOI: 10.5220/0013344100004646

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 1st International Conference on Cognitive & Cloud Computing (IC3Com 2024), pages 37-45

ISBN: 978-989-758-739-9

abstractions in data [3]. Therefore, this survey aims to

provide a comprehensive overview of network

anomalies and the application of deep learning for

their detection and mitigation. The objective of this

research paper is to recognize the behavioral patterns

of various network anomalies in regards to the

environments in which they exist, determine the

different types of network anomalies exists in the

edge computing network, contrasting over the

different edge computing network anomalies and

suitable deep learning techniques, proposing the

common architecture for data modelling using deep

learning and detection of edge anomalies using deep

learning techniques and finally proposing a case study

on anomaly detection in medical applications over the

edge network.

RELATED WORK

There is limited literature on implementations of edge

computing anomaly detection using deep learning is

available. Some of the important aspects in the

literature are explained as follows.

Shagufta Mehnaz Elisa Bertino [30], has explained

six categories of malicious attacks: Brute Force,

Denial-of-service (DoS), Web Attacks, Infiltration,

Botnet, and Distributed denial-of-service (DDoS). To

detect anomalies authors have explained deep neural

network (DNN), convolutional neural network

(CNN), recurrent neural network (RNN) techniques.

Ahmed Dawoud, Seyed Shahristani, Chun Raun [31]

has proposed a model that uses an autoencoder (AE)

to identify samples exhibiting anomalous behavior;

an attack classifier is then employed to categories

anomalies according to the sort of attack they

represent. Daniel Alejandro, Oscar Augusto et al [32]

has proposed the rule-based neural network model

with supervised machine learning technique to

perform optimal detection of any type of network

anomalies over Network Intrusion Detection (NIDS).

Zeyuan Fu [33] has proposed the rule-based neural

network model with supervised machine learning

technique to perform optimal detection of any type of

network anomalies over Network Intrusion Detection

(NIDS). Yuhuai Peng, Aiping Tan, Jingjing Wu,

Yuanguo Bi [34] has proposed Conceptual

frameworks of three main deep anomaly detection

approaches like Feature Extraction, Learning Feature

Representations of Normality, and End-to-end

Anomaly Score Learning. Mohab Sameh, Abdel-

Wahab et al. [35] has explained the different machine

learning and deep learning approaches that can be

used to detect network anomalies in Intrusion

Detection Systems. Haolong Xiang & Xuyun

Zhang[36] has proposed IDForest, and IoT- based

edge computing-powered anomaly detection

framework that uses insertion and deletion strategies

to update the tree structure of data.

METHODOLOGY

The methodology for proposed survey covers the

following aspects of solutions.

3.1

Edge Computing Network

Anomalies

Due to the distinctive features and difficulties of

distributed computing at the network edge, edge

computing is susceptible to network anomalies. The

primary contribution is the recommendation of an

edge computing architecture that can do anomaly

detection in multiple sources at numerous ends [4].

3.1.1 Communication Failures:

Due to unstable network connections or interference,

edge devices and nodes may occasionally have

connectivity problems or communication failures.

These errors might obstruct the flow of data and cause

service interruptions. Network Instability is also one

of the reasons for Communication Failure. It is

essential to build the edge network with redundancy,

prioritize key data, maximize network capacity,

implement strong error handling and recovery

mechanisms, and often monitor and maintain the

network architecture to reduce these communication

failures. when one node fails, there could be

cascading failures that have a negative influence on

the service's delivery and make it impossible to

achieve certain goals [5].

3.1.2 Latency Spikes:

Latency spikes in edge computing refer to sudden and

significant increases in the time it takes for data to

travel between edge devices and the processing or

storage resources. These spikes can have several

causes and can negatively impact the performance

and responsiveness of edge applications. Edge

computing includes processing data nearer to the data

source, at the network edge. However, latency spikes

can cause delays in data processing and response

times due to a lack of computing resources or

congestion in the edge network.

3.1.3 Scalability Issues:

Scalability issues can occur in edge computing due to

IC3Com 2024 - International Conference on Cognitive & Cloud Computing

various factors like Growing Number of Edge

Devices, Increased Data Volume, Resource

Limitations, Data Synchronization and Consistency,

Dynamic Edge Environment, Network Capacity.

Edge computing systems must scale dynamically in

response to shifting workload needs. When the system

struggles to handle abrupt increases in workload or

fails to allocate resources effectively, anomalies

might occur if the network architecture or edge nodes

are not appropriately setup for scalability.

3.1.4 Edge Network Congestion:

Edge computing edge network congestion difficulties

can emerge for a variety of reasons. Here are some

major causes of edge network congestion- Increased

Data Traffic, Bandwidth Limitations, Network

Infrastructure Bottlenecks, Inefficient Routing,

Bursty Traffic congestion, which can result in

performance deterioration, increased latency, and

packet loss. There are several strategies that can be

used to deal with edge network congestion concerns.

3.1.5 Resource Constraints:

Resource constraint issues can arise in edge

computing due to several factors inherent to the

edge environment like Power Limitations, Limited

Computational Power, Storage Capacity, Memory

Constraints, Storage Capacity, Network

Bandwidth, Heterogeneous Hardware,

Scalability Challenges. The compute, memory, and

energy resources of edge devices are frequently

constrained. When these resources are used up or

improperly maintained, anomalies can develop that

affect the edge network's overall performance.

3.1.6 Edge infrastructure Failures:

Edge infrastructure failures can occur in edge

computing due to various factors such as Power

Outages, Network Connectivity Issues, Hardware

Failures, Software and Firmware Issues,

Environmental Factors, Insufficient Capacity, Edge

computing depends on the edge servers, routers,

switches, and other networking hardware that make

up the network edge. Anomalies can happen because

of hardware malfunctions, power outages, or

environmental variables, which can cause service

interruptions or total system failure.

3.1.7 Edge infrastructure Failures:

Edge node overload issues can occur in edge

computing due to several factors such as resource

constraints, high workload demand, sudden spikes in

data or user activity, Inefficient task scheduling,

Insufficient load balancing, Device mobility,

Inadequate scalability planning. Compared to

centralized servers, edge nodes' processing power is

constrained. An edge node may get overloaded if the

workload is greater than its capacity, which could

lead to performance patterns, Resource Contentions.

Edge networks degradation and service disruptions.

3.3 Deep Learning Techniques

Anomaly detection methods using deep learning fall

into three categories: super-vised, hybrid, and

unsupervised [29]. The labels in the dataset mostly

determine the appropriate anomaly detection

technique. Applying deep learning techniques to

anomaly detection has several benefits. First and

foremost, these strategies are in-tended to handle

multivariate and highly dimensional data. This

eliminates the need to model anomalies independently

for each variable and average the results, making it

easier to integrate data from several sources. The

classification of various deep learning methods is

shown in Fig. 1.

Figure 1: Classification of Deep learning method.

Performance is yet another benefit. Deep learning

techniques provide the chance to model intricate,

nonlinear relationships inside data and use these

interactions for the task of anomaly identification.

Deep learning models are excellent for situations

involving large amounts of data because their

performance may scale as relevant training data

become available. Deep learning uses a complicated

architecture or combination of non-linear

transformations to achieve high level abstractions in

data [3]. Table 1 represents different deep anomaly

detection (dad) techniques.

Comprehensive Survey on Detection of Anomalies in Edge Computing Network and Deep Learning Solutions

Beyond specifying generic hyper parameters (number

of layers, units per layer, etc.), deep learning

approaches are also well- suited to jointly modeling

the interactions between multiple variables with

respect to a given task. Deep learning models only

need minor tuning to produce effective results. The

different Deep Anomaly Detection techniques are

listed in Table 1.

In edge computing contexts, network anomalies can

be detected using deep learning techniques. Here are

several methods that are frequently used:

3.3.1

Recurrent Neural Networks (RNNs)

RNNs, LSTM networks with long short-term

memory, are excellent at analyzing sequential data.

RNNs can examine time- series data from edge

devices and nodes in the context of edge computing

to find anomalies in network traffic, resource usage,

or other performance parameters. An extension of a

traditional feed-forward neural network is the

recurrent neural network (RNN). RNNs are effective

for simulating sequences because they have cyclic

connections; unlike feed forward neural networks

[3],[8]. The RNN-IDS model enhances intrusion

detection accuracy and offers a fresh approach to

intrusion detection research [9].

3.3.2 Generative Adversarial Networks

(GANs):

GANs can be used to discover the distribution of

typical network edge behavior. While the

discriminator network distinguishes between real and

created data, the generator network is trained to

produce artificial normal data. By evaluating the

discriminator's capacity to categorize fresh data

samples, anomalies can be found. Researchers are

dedicated to correctly and effectively identifying

aberrant photos in real-world applications and use

them in the field of anomaly detection [10].

3.3.3 Autoencoders

Autoencoders can be used to figure out how network

traffic typically behaves or how resources are used at

the edge. The autoencoder can reconstruct and

compare new data instances by being trained on

regular data. Recently, a deep learning technique

utilizing Autoencoder (AE) models has gained

acceptance for locating inappropriate characteristics

present in the huge network traffic samples [11]-[14].

3.3.4

Attention Mechanisms

Deep learning models can incorporate attention

mechanisms to concentrate on facets or time intervals

of network input. Attention mechanisms can increase

the accuracy of anomaly detection in edge computing

environments by drawing attention to important

features or trends [15]. Neural networks can digest

input data while dynamically focusing on pertinent

portions of it through attention processes.

3.3.5

Variational Autoencoders (Vaes)

The VAEs are used in edge computing to capture the

distribution of network activity and resource usage.

Variational autoencoders (VAEs) are generative deep

learning models that can process a wide range of data

types, including images and text. VAEs represent

neural networks' posteriors and frequently beat

autoencoders and one-class support vector machines

in recognizing unknown threats [16- 18].

3.3.6

Reinforcement learning (RL)

Reinforcement learning has Deep Q- Networks

(DQNs) method that enhances the network

administration and detects the edge anomalies. For

improving the efficiency and security of edge

infrastructure, the RL agents are trained to allocate

resources, route traffic, and respond to the anomalies.

The Reinforcement learning supports variety of

patterns and data formats, including text and images

and used to address the issues related to resource

allocation [19].

3.3.7 Graph

Neural Networks (GNNs)

GNNs can identify abnormalities based on graph

patterns or irregularities in the edge network topology

by modelling the relationships and interactions

between edge devices, nodes, and their properties.

The suitable deep learning technique must be chosen

based on the unique properties of the edge network

Metho

Applicati

Superv

ised

Unsup

ervised

Hybrid

Model

Fraud Detection

Medical Anomaly

Detection Internet of

Things (IoT)

Industrial Damage

Detection Bi

-data

IC3Com 2024 - International Conference on Cognitive & Cloud Computing

data, the kinds of anomalies to be found, and the

resources and computational power available at the

edge devices or nodes. Anomaly detection in edge

computing environments can also be improved by

combining various approaches or hybrid methods

with conventional machine learning algorithms.

3.3.8

Convolutional Neural Networks (CNNs)

CNNs can be used to examine data gathered from

edge nodes about network traffic. CNNs can learn to

recognize patterns and anomalies in network traffic

by considering it as a sequential data stream or a time-

series, assisting in the identification of anomalous

behavior or assaults. The Convolutional Neural

Network (CNN) uses these fixed-size feature

matrices to detect insider threat after converting the

retrieved characteristics into them [7]. For

applications like image identification, object

detection, and image segmentation, CNNs are quite

effective.

4 ARCHITECTURE FOR DATA

MODELLING USING DEEP

LEARNING

The architecture for data modeling using deep

learning poses various stages. The block

diagram for anomaly detection and forecasting

using deep learning using various stages is

shown in Fig. 2. The diagram illustrates the

various components and steps involved in the

process.

Figure 2: Model for anomaly detection using deep

learning algorithms.

The different stages used in the above model are

explained as follows:

Data Collection: The process starts with the

collection of datasets from various sources,

including real-time transactions, credit card uses

patterns, and merchant transactions.

Preprocessing: To prepare for model training, the

obtained data is pre- processed, which includes

cleaning, normalization, and feature extraction.

Feature Selection: This stage entails picking the

most relevant characteristics from the pre-

processed data to serve as input for the deep

learning models.

Model Training: Deep learning models, such as

LSTM, Autoencoder, LSTM- Autoencoder,

Transformer, and GAN, are trained on the selected

features to learn the patterns and relationships in the

data.

Anomaly Detection: The trained models are then

used for anomaly detection, where they analyze

incoming data in real- time to identify any

deviations from normal patterns.

Feedback Loop: Anomaly detection results are fed

back into the system for evaluation and further

refinement of the m o d e l s t o i m p r o v e

t h e i r accuracy over time. Evaluation & Analysis

of results:

The performance of the deep learning models in

detecting anomalies and forecasting time series

data is evaluated using benchmark datasets and

metrics and perform the comparative analysis of

results from proposed solution with results from

existing machine learning and deep

learning

solutions. Validation of test results using statistical

techniques: Further the results of proposed model are

validated against statistical method like hypothesis

testing using Chi-Square & ANOVA Methods.

4.3 Detection of Edge Anomalies Using

Deep Learning Techniques

There are many datasets are available for anomaly

detection in edge computing network, in each of the

dataset, data preprocessing and modeling is required

for getting efficient and optimal results. Based on

type of anomaly in the edge platform, different

datasets have been used by many authors in their

respective research. The various datasets and

examples of deep anomaly detection are explained as

follows. Table 2 demonstrate a list of possible

network anomalies in edge computing environments,

as well as the deep learning methods that can be

applied to spot those anomalies over the suitable

dataset [22]- [29].

Table 2: Detection of Edge Anomalies using Deep

learning techniques on a Suitable dataset

Comprehensive Survey on Detection of Anomalies in Edge Computing Network and Deep Learning Solutions

Furthermore, for this purpose, alternative deep

learning methods like Convolutional Neural

Networks (CNNs) or hybrid models that combine

CNNs and LSTMs could be investigated [20]- [21].

CASE STUDY

There are many real-world applications where

edge computing plays an important role for making

the computational speed of application faster along

with providing high privacy and security. Some of the

colossal benefits of edge computing platform are

enhanced scalability, lowers network latency, cost

savings, improved reliability, and real- time analytics.

But due to the different types of anomalous attacks

these benefits would get compromised resulting in

slow performance, high latency, compromised

security and connectivity. Because of which anomaly

detection plays an important role in making the edge

devices secured. Some of the real-world applications

of edge computing are Autonomous vehicles, Smart

cities for connected cars, cameras, and traffic signals,

healthcare, and telemedicine etc. In Healthcare and

Telemedicine, the edge computing is used for

processing of medical data in real time that help

doctors to monitor patient health remotely.

5.1

Anamoly Detection in Medical

Applications over Edge Network

In medical applications, edge computing plays an

important role in collecting processing and storing the

medical data at nearby edge locations where doctors

can easily monitor the patient health and provide

effective treatment. The doctors take the various

readings of the patient’s health records by using the

various sensory devices placed on the human body.

The readings taken from sensors further used for

getting the results related to diagnosis of disease

based on that they provide the accurate treatment. The

benefits of the edge computing for medical

applications are higher computational speed for

processing, speedy results of diagnosis, higher

storage for maintaining report records in various

formats, scalability, and easy accessibility. Therefore,

any compromise in the sensors data would result in

inaccurate disease diagnosis may result in wrong

treatment.

That is why the consistency must be maintained in the

data of the medical applications.

For example, suppose the sensory devices of IoT

edge have been used for collecting the pulse and

temperature data from the patient body. Any

inaccuracy or data inconsistency in the pulse or

temperature data may results in imposing the wrong

treatment for the patient or sometimes putting the

patience life in danger. The latency in edge may result

in slower generation of results of reports may delay

the treatment and the communication failure may

result in unavailability of reports may delay the

treatment of critical patient.

Therefore, it is necessary to get the accurate

sensory data over the edge and maintain integrity of

IC3Com 2024 - International Conference on Cognitive & Cloud Computing

it. The various stages of edge computing enabled

medical applications involves selecting the nearby

edge network devices and sensors, data acquisition

and processing, applying the deep learning algorithms

and get the automated results where accuracy is

important. The problem of anomaly detections comes

into Classification model of deep learning which

classifies the anomalous and normal data separately

from dataset.

To evaluate and validate the edge computing

enabled anomaly detection using deep learning for

medical applications, the experimentation has been

conducted on UNSW-NB15 dataset for detecting the

attacks that causes latency spike, communication

failure and the data inconsistency in the data [37]. For

experimentation, UNSW-NB15 dataset have been

used which has 2,57,673 Samples. About 1,64,673

samples having anomalous data. The different deep

learning models like Deep Neural Network (DNN)

and Generative Adversarial Networks (GAN) have

been used for classification of data incurring

anomalies over the labeled dataset that has sensor data

recorded at regular intervals. It verifies in each dataset

whether the anomalies exist or not. It is used for

validating the accuracy and F1-Score of the different

anomalies like latency spike, communication failure

or data inconsistency exist over the given data set

[38][39]. In this experiment we have trained a model

on number of samples 85699 for 25 epochs and

verified the accuracy with 29000 testing dataset and

same samples have been used with 25 epochs for

GAN model [39]. The result of classification models

like DNN and GAN generates the confusion matrix in

the classification that has four quadrant values like

True Positive (TP), True Negative (TN), False

Positive (FP) and False Negative (FN). Based on the

confusion matrix, evaluation metrics like Accuracy

(A), Recall (R), Precision (P) and F1-Score have been

calculated to test the anomalous behavior of the Edge

data. The formulas for various evaluation metrics for

classification are shown in Table 3.

Table 3. Formulas of evaluation metrics for classification.

Parameter Formula

Precision

(P)

P= TP/(TP+FP)

& TN/(TN+FN)

Recall (R)

R=TP/(TP+FN)

& TN/(TN+FP)

Score(F) F=(2*P*R)/(P+R)

Accuracy

(A)

A=(TP+TN)/(TP+TN+FP+

FN)

The results obtained from both DNN and GAN

models over the given dataset is given in Table 4.

Table 4: Results obtained from DNN and GAN models.

Models

Deep Neural Network

Generative

Adversarial Networks

Attacks

Precision Recall F1-

score

Precision Recall F1-

score

DOS

0.53 0.73 0.65 0.98 0.98 0.99

EXPLO

0.75 0.63 0.69 0.83 0.94 0.90

GENER

0.99 0.97 0.99 0.99 0.96 0.97

NORMA

0.98 0.98 0.99 0.99 0.9 0.88

ACCURACY

0.8325

0.9375

From above results, it has been observed that DNN

model gives the precision, recall, and F1-Score

average value of 0.8325, whereas the GAN model

gives the average value of 0.9375. That means GAN

models gives accuracy of 93.75% which is more than

DNN and has reduced loss than DNN.

The conclusion of above experiment is more the

loss in the attack samples lesser the accuracy. The

lesser accuracy results in more percent of attacks that

result in increase in increase of latency spike,

communication loss or sometimes data inconsistency.

The future scope for above experiment would be

use of fuzzy neural network which may give more

accuracy than the above two models.

CONCLUSION

The expanding number of IoT devices and the

continual collection of network data have highlighted

the relevance of anomaly detection in edge computing

networks. This research study discusses the difficulty

in monitoring edge networks and detecting

unexpected network behavior, emphasizing the

importance of having effective anomaly detection

tools and a relevant dataset. This research study has

quickly reviewed the various features of edge

computing anomalies that exist in different settings.

It also discusses the classification of various deep

learning methods and the relationships between

different types of data. The various datasets utilized

for edge computing anomaly detection have also been

examined. Finally, a model for using deep learning

techniques on edge computing data sets to detect

anomalies has been developed. Finally, the case study

on anomaly detection in medical applications over the

Comprehensive Survey on Detection of Anomalies in Edge Computing Network and Deep Learning Solutions

edge network has been discussed. It has covered the

importance of edge computing for medical

application along with detection of anomalies using

the two deep learning models like DNN and GAN. In

the results, the accuracy of GAN appears better than

DNN with lesser loss and seems better in detecting

anomalies than DNN. In the future scope, same could

be implemented using fuzzy neural classifier to get

better results than the DNN and GAN.

ACKNOWLEDGEMENTS

We gratefully acknowledge to Dr. Bhushan Jadhav

at Thadomal Shahani Engineering College who

helped and support during the research process. Their

contributions were instrumental in shaping the

direction and scope of this work. Additionally, we

acknowledge the reviewers whose thoughtful

comments and suggestions greatly enhanced the

clarity and depth of the manuscript. Lastly, we

express our appreciation to Springer and the editorial

team for their professionalism and assistance in the

publication process.

REFERENCES

Chalapathy, R., Chawla, S., 2019. Deep Learning for

Anomaly Detection: A Survey. arXiv:1901.03407.

Zhu, M., Ye, K., Wang, Y., Xu, C.Z., 2018. A Deep

Learning Approach for Network Anomaly Detection

Based on AMF- LSTM. In: 15th IFIP International

Conference on Network and Parallel Computing

(NPC), Muroran, Japan, pp. 137-141. Springer.

Kim, J., Kim, J., Thu, H.L.T., Kim, H., 2016. Long short-

term memory recurrent neural network classifier for

intrusion detection. In: Platform Technology and

Service (PlatCon), International Conference on, pp.

1-5. Springer.

Anantha, A.P., Daely, P.T., Lee, J.M., Kim, D.S., 2020.

Edge Computing-Based Anomaly Detection for Multi-

Source Monitoring in Industrial Wireless Sensor

Networks. In: ICTC 2020, pp. 1890- 1892.Springer.

Majeed, A.A., Kilpatrick, P., Spence, I., Varghese, B.,

2022. CONTINUER:Maintaining Distributed DNN

Services During Edge Failures. In: IEEE International

Conference on Edge Computing and Communications

(EDGE), 24 August 2022. Springer.

Hu, P., Chen, W., 2019. Software- Defined

Edge Computing (SDEC): Principles,

Open System Architecture and Challenges. In: IEEE

SmartWorld, Ubiquitous Intelligence &Computing

(SmartWorld/ SCALCOM/UIC/ATC/CB D

Com/IOP/SCI). Springer.

Yuan, F., Cao, Y., Shang, Y., Liu, Y., Tan, J., Fang,

B., 2018. Insider threatdetection with deep neural

network. In: International Conference on Computational

Science, pp. 43-54. Springer.

Kwon, D., Kim, H., Kim, J. et al., 2019. A survey of

deep learning-based network anomaly detection.

Cluster Computing, pp. 949–961.

Yin, C., Zhu, Y., Fei, J., He, X., 2017. A Deep Learning

Approach for Intrusion Detection Using Recurrent

Neural Networks. IEEE Access, vol. 5, pp. 21954-

21961.

10.

Zhang, L., Fan, F., Dai, Y., He, C., 2022.Analysis

and research of generative adversarial network in

anomaly detection. In: 7th International Conference on

Intelligent Computing and Signal Processing (ICSP),

IEEE.

11.

Zhang, B., Yu, Y., Li, J., 2018. Network intrusion

detection based on stacked sparse autoencoder and

binary tree ensemble method. In: Proc. IEEE Int. Conf.

Commun. Workshops (ICC Workshops), pp. 1–6.

12.

Yan, B., Han, G., 2018. Effective feature extraction via

stacked sparse autoencoder to improve intrusion

detection system. IEEE Access, vol. 6, pp. 41238–

41248.

13.

Al-Qatf, M., Lasheng, Y., Al-Habib, M., Al- Sabahi,

K., 2018. Deep learning approach combining sparse

autoencoder with SVM for network intrusion detection.

IEEE Access, vol. 6, pp. 52843–52856.

14.

Xu, W., Jang-Jaccard, J., Singh, A., Wei, Y., Sabrina,

F., 2021. Improving Performance of Autoencoder-

Based Network Anomaly Detection on NSL- KDD

Dataset. IEEE Access, vol. 9, pp. 140136-140146.

15.

Liang, Y., Chen, Z., Lin, D., Tan, J.,Yang, Z., Li,

J., Li, X., 2023. Three-Dimension Attention

Mechanism and Self-Supervised Pretext Task for

Augmenting Few-Shot Learning. IEEE Access, vol. 11,

pp. 59428- 59437.

16.

Kingma, D.P., Welling, M., 2013. Auto- encoding

variational Bayes. arXiv:1312.6114.

17.

An, J., Cho, S., 2015. Variationalautoencoder-

based anomaly detection using reconstruction

probability. SNU Data Mining Center, Seoul, South

Korea, Tech. Rep.

18.

Zavrak, S., Skefiyeli, M., 2020. Anomaly- Based

Intrusion Detection from Network Flow Features

Using Variational Autoencoder. IEEE Access, vol. 8,

pp. 108346-108358.

19.

Li, Q., Zhu, Y., Ding, J., Li, W., Sun, W., Ding, L., 2021.

Deep Reinforcement Learning based Resource

Allocation for Cloud Edge Collaboration Fault

Detection in Smart Grid. CSEE Journal of Power and

Energy Systems, pp. 1-10.

20.

Jozefowicz, R., et al., 2016. Exploring the limits of

language modeling. arXiv:1602.02410.

21.

Graves, A., Mohamed, A.R., Hinton, G., 2013. Speech

recognition with deep recurrent neural networks. In:

IEEE International Conference on Acoustics, Speech

and Signal Processing (ICASSP). IEEE.

IC3Com 2024 - International Conference on Cognitive & Cloud Computing

22.

Jayasinghe, M., Seneviratne, S., Seneviratne, A., 2017.

Deep learning for anomaly detection in network traffic

data. In: IEEE International Conference on Pervasive

Computing and Communications Workshops

(PerCom Workshops). IEEE.

23.

Banga, G., Dubey, K.K., Mishra, K.K., 2018. LSTM-

RNN based network anomaly detection. In:

International Conference on Computing, Power and

Communication Technologies (GUCON). IEEE.

24.

Chalapathy, R., Chawla, S., 2018. Deep Learning for

Anomaly Detection: A Survey. ACM Computing

Surveys (CSUR), vol. 51, no. 3.

25.

Yao, H., Zheng, S., Wen, H., Li, X., 2020.Deep

Learning for Anomaly Detection: A Review. ACM

Computing Surveys, vol. 53, no. 2.

26.

Pham, K.H., Bui, T.D., Venkatesh, S., 2019. Anomaly

Detection in Edge Computing Systems Using Deep

Learning. IEEE Transactions on Services Computing,

vol. 12, no. 1, pp. 114-127.

27.

Kiran, A., Basavaraju, T.G., 2021.Anomaly

Detection in IoT-Based Edge Computing Using LSTM

Autoencoder. In: International Conference on

Emerging Trends in Information Technology and

Engineering (ic-ETITE).

28.

Huc, A., Šalej, J., Trebar, M., 2023. Analysis of

Machine Learning Algorithms for Anomaly Detection

on Edge Devices. MDPI.

29.

Elsayed, M.S., Le-Khac, N.-A., 2020.Network

Anomaly Detection Using LSTM- Based Autoencoder.

In: Q2SWinet '20, ACM, Alicante, Spain.

30.

Mehnaz, S., Bertino, E., 2020. Privacy- preserving

Real-time Anomaly Detection Using Edge Computing.

In: IEEE 36th International Conference on Data

Engineering (ICDE), May 2020, pp. 469-480. IEEE.30.

31.

Dawoud, A., Shahristani, S., Raun, C., 2019. Deep

Learning for Network Anomalies Detection. In:

International Conference on Machine Learning and

Data Engineering (iCMLDE), IEEE, pp. 149-153.

32.

Li, R., Zhou, Z., Liu, X., et al., 2021.GTF:An

Adaptive Network Anomaly Detection Method at the

Network Edge. Security and Communication

Networks, vol. 2021, Article ID 3017797, pp. 1-12.

33.

Ferrari, P., Rinaldi, S., Sisinni, E., et al., 2019.

Performance evaluation of full- cloud and edge-cloud

architectures for Industrial IoT anomaly detection

based on deep learning. IEEE, pp. 420-425.

34.

Peng, Y., Tan, A., Wu, J., Bi, Y., 2019.Hierarchical

Edge Computing: A Novel Multi-Source Multi-

Dimensional Data Anomaly Detection Scheme for

Industrial Internet of Things. IEEE Access, vol. 7, pp.

111257-111270.

35.

Abdel-Wahab, M.S., et al., 2021. A Comparative Study

of Machine Learning and Deep Learning in Network

Anomaly- Based Intrusion Detection Systems. IEEE,

February 2021.

36.

Xiang, H., Zhang, X., 2022. Edge computing

empowered anomaly detection framework with

dynamic insertion and deletion schemes on data

streams. IEEE.

37.

Samariya, D., Ma, J., Aryal, S., Zhao, X., 2023.

Detection and explanation of anomalies in

healthcare data. Journal of Health Information

Science and Systems, Springer.

38.

Schneible, J., 2017. Anomaly Detection on the Edge.

In: MilCom 2017, IEEE Cyber Security and Trusted

Computing.

39.

Sharma, B., Sharma, L., Lal, C., Roy, S., 2023.

Anomaly based network intrusion detection for IoT

attacks using deep learning technique. Journal of

Computers and Electrical Engineering, Elsevier

Comprehensive Survey on Detection of Anomalies in Edge Computing Network and Deep Learning Solutions