1 Review Article Machine and Deep Learning for IoT Security and Privacy

2025-04-30 0 0 1.13MB 46 页 10玖币

Review Article

Machine and Deep Learning for IoT Security and Privacy:

Applications, Challenges, and Future Directions

Subrato Bharati1, Prajoy Podder2

1,2Institute of Information and Communication Technology (IICT), Bangladesh University of Engineering and Technology (BUET),

Dhaka-1205, Bangladesh

Correspondence should be addressed to Subrato Bharati; subratobharati1@gmail.com

Abstract: The integration of the Internet of Things (IoT) connects a number of intelligent devices with a minimum of human interference

that can interact with one another. IoT is rapidly emerging in the areas of computer science. However, new security problems were posed

by the cross-cutting design of the multidisciplinary elements and IoT systems involved in deploying such schemes. Ineffective is the

implementation of security protocols, i.e., authentication, encryption, application security, and access network for IoT systems and their

essential weaknesses in security. Current security approaches can also be improved to protect the IoT environment effectively. In recent

years, deep learning (DL)/ machine learning (ML) has progressed significantly in various critical implementations. Therefore, DL/ML

methods are essential to turn IoT systems protection from simply enabling safe contact between IoT systems to intelligence systems in

security. This review aims to include an extensive analysis of ML systems and state-of-the-art developments in DL methods to improve

enhanced IoT device protection methods. On the other hand, various new insights in machine and deep learning for IoT Securities

illustrate how it could help future research. IoT protection risks relating to emerging or essential threats are identified, as well as future

IoT device attacks and possible threats associated with each surface. We then carefully analyze DL and ML IoT protection approaches

and present each approach's benefits, possibilities, and weaknesses. This review discusses a number of potential challenges and

limitations. The future works, recommendations, and suggestions of DL/ML in IoT security are also included.

Keywords: Security applications, Deep learning, Internet of Things (IoT), Security and Privacy, Machine Learning, Applications

1. Introduction

Internet of Things (IoT) considers the interconnection

between several devices, i.e., industrial systems,

intelligent sensors, autonomous vehicles, mechanisms

and terminals, mechanical systems, etc. [1, 2].

Alternatively, it can be termed as a network of physical

things or objects that are connected with limited

communication, computation, and storage capabilities

along with embedded electronics (i.e., sensors and

actuators), connectivity of network, and software that

enables these things to exchange, analyze, as well as

collect data [3]. IoT relates to our everyday life,

extending from smart devices in the household, i.e.,

smart meters, IP cameras, smoke detectors, smart

adapters, smart refrigerators, smart bulbs, AC, smart

ovens, and temperature sensors, to more advanced

devices, for example, heartbeat detectors, radio-

frequency identification (RFID) devices, accelerometers,

IoT in automobiles, sensors in rooms, etc. [4]. Several

services and applications referred to by the IoT are

emerging in personal healthcare, home appliances,

critical agricultural infrastructure, and the military [1].

The massive scale of IoT networks introduces

latest issues, including the management of these devices,

the complete volume of data, communication, storage,

processing, as well as security and privacy concerns,

among others. There has been substantial research into

the various components of the IoT, such as architecture,

communication, applications, protocols, security, and

privacy, to name a few. The guarantee of security and

privacy and user satisfaction are the cornerstones of the

commercialization of IoT technology. The fact that the

IoT makes use of empowering technologies including

cloud computing (CC), software-defined networking

(SDN), and edge computing enhances the number of

dangers that attackers can encounter. As a result,

monitoring security in the development of IoT

infrastructure has become challenging and complex.

Solutions must consist of wide-ranging considerations to

fulfill the security challenges [5]. On the other hand, IoT

systems are frequently put to use in an unprepared state.

As a result, a fraudster can use wireless networks to

connect to IoT devices and gain physical access to

confidential data. Complexity and integrative

arrangements characterize IoT systems. In light of the

proliferation of connected devices, it might be difficult to

meet the ever-evolving security standards for the IoT. In

order to provide the necessary level of security, solutions

need to consider the system as a whole. However, most

IoT devices can function independently of human input.

Someone without permission may thus acquire physical

access to these devices [6-8].

Furthermore, The IoT system introduces novel

attack surfaces. The interconnected and interdependent

systems cause these types of attacks to surface.

Accordingly, the security of IoT systems is faced with a

higher risk than the security of other traditional

computing devices. The outdated computing systems

will be fruitless for these IoT schemes [9-11].

IoT systems ought to instantaneously consider

security, energy efficiency, IoT software applications,

and data analytics at the time of related tasks as a sign of

the wide-ranging application [7]. This expansion offers

an innovative scope for scholars from the

interdisciplinary research program to consider recent

challenges in the IoT schemes from various perceptions.

However, the large-scale as well as cross-cutting nature

of IoT devices and the many components engaged in

their implementation, have created new security issues.

The IoT devices characteristic presents various security

issues. Additionally, the stages of IoT provide a massive

number of useful information. If this information is not

analyzed and transmitted securely, a crucial privacy gap

may occur. Applying related security mechanisms, such

as authentication, encryption, application security,

network security, and access control, is inadequate and

challenging for enormous schemes with numerous

associated schemes. Every portion of the IoT platform

contains intrinsic vulnerabilities. Such as, a special kind

of botnet like ‘Mirai’ has newly affected extensively

distributed denial of service (DDoS) attacks by using IoT

systems [9, 12].

For extensive methods with multiple connected

devices and each module of the method having inherent

vulnerabilities, it is difficult and inadequate to apply

existing security protection mechanisms like encryption,

identity verification, application security, access control,

and computer security [9]. For example, the 'Mirai'

botnet has lately been responsible for large-scale DDoS

attacks by abusing IoT devices. For the IoT ecosystem,

existing security methods need to be improved.

However, the deployment of cryptographic functions

against a particular security issue is rapidly overtaken by

new categories of attack developed by the attackers in

order to bypass current remedies. Addresses spoofed

source IP are commonly applied in magnified DDoS

attacks to hide the location of attack's from the targeted

organization's security teams. As a result of the

vulnerabilities in IoT systems, more sophisticated and

catastrophic attacks such as Mirai might be predicted. On

account of the wide range of IoT scenarios and

applications, knowing which security solutions are best

for IoT systems is not easy. As a result, the focus of the

study should be on devising appropriate IoT security

methods [12, 13].

While security and privacy are interconnected,

security may exist without privacy, but privacy cannot

exist without security. Security safeguards the

availability of information, integrity, and confidentiality,

while privacy is more detailed about privacy rights in

relation to personal information. Regarding the

processing of personal data, privacy takes precedence,

but information security entails preventing illegal access

to information assets. Personal data may relate to any

information about a person, including names, credentials,

addresses, social security numbers, bank account data,

etc.

A number of ways have been suggested to address

the boundary between security and privacy concerns in

DL and ML. Homomorphic encryption, differential

privacy, trusted execution and secure multi-party

computing environment are the four most often used DL

and ML privacy technologies. This technique uses

differential privacy to prevent the adversary from

figuring out which instances were utilized to build the

target model. Training and testing data are protected by

safe multi-party computing and homomorphic

encryption. For sensitive data security and training code,

trusted execution environments leverage hardware-based

security and isolation. These approaches, on the other

hand, greatly increase the computing burden and need a

tailored approach for each type of neural network. DL or

ML privacy concerns have yet to be addressed in a way

that is accepted worldwide. To protect against

adversarial attacks, a wide variety of security measures

have been suggested, which may be divided into three

categories: input pre-processing, strengthening the

model's resilience and malware detection.

Preprocessing's goal is to lessen the model's reliance on

immunity by doing operations such picture

transformation, randomization, and denoising that don't

often need model update or retraining. Introducing

regulation, feature denoising, and adversarial training as

well as other techniques to strengthen the model's

robustness via model retraining and change, fall under

the second group. Adaptive denoising and image

transformation detection are the examples of third-

category detection mechanisms that may be implemented

before the first layer of the model. To the best of our

knowledge, no defense strategy exists that can entirely

protect against adversarial cases despite the many

defensive mechanisms that have been offered. To

counter hostile instances, adversarial training is currently

the most effective technique. For poisoning attacks, there

are two basic means of defense. The first is an outlier

identification technique, which eliminates outliers from

the relevant set. The second step is to enhance the neural

network's ability to withstand contamination from

poisoned samples.

1.1. Motivation and Scopes

Deep learning (DL) and machine learning (ML) are

effective methods of data analytics as well as

investigation to realize ‘abnormal’ and ‘normal’

behavior following how IoT devices and components

interrelate with each other within the environment of IoT

[14]. The IoT systems' input data can be investigated and

collected to find out standard patterns of the interface,

thus detecting malicious manners at the initial stages.

Furthermore, DL/ML techniques can be significant in

identifying new threats, which are regular modifications

of existing threats, as they can highly detect upcoming

unknown threats by learning from previous attacks. As a

result, IoT systems need to be able to move from secure

communication between security-based intelligence and

devices via ML/DL techniques for safe and efficient

systems.

Several unique properties of IoT networks will

be discussed in the following paragraphs.

Heterogeneity: Each item in an IoT network has unique

features, communication protocols, and capabilities that

all function together. Different communication

paradigms and protocols (such as Ethernet or cellular), as

well as varied hardware resource limits, might be used

by the devices. On the one hand, this diversity allows

devices to communicate across platforms, but on the

other, it introduces additional obstacles to the network of

IoT.

Proximity communication: Additionally, IoT devices

may communicate with one other without trusting on a

central authority like base stations, which is an important

feature. Dedicated short range communication (DSRC)

and other point-to-point communication technologies are

used in Device-to-device communication (D2D).

Decoupling services and networks allows device-centric

and content-centric communication, broadening the IoT

service spectrum, whereas the conventional Internet's

design is more network-centric.

Massive deployment: Massive deployment is predicted

that the existing internet's capabilities would be exceeded

by the billions of devices linked to it and the Internet.

Massive IoT deployments are not without their own set

of difficulties. Storage and architecture networks for

intelligent devices, efficient protocols for data transfer,

and proactive detection and protection of IoT-based

devices from malicious attacks are only some of the

difficulties that need to be addressed. A worldwide

information and communication infrastructure that can

be retrieved from everywhere as well as at any time is

envisaged for IoT devices. How much is connection

reliant on the kind of IoT service and application

provided? For example, a swarm of sensors or a

connected automobile may have a local connection,

whereas critical infrastructure management and smart

home access through mobile infrastructure may have

global connectivity.

Low-cost and low-power communication: For optimal

network operations, low-cost as well as ultra-low-power

solutions are needed for the massive networking of IoT

devices. For modern and critical IoT connectivity, self-

healing and self-organization features are needed. Self-

organizing networks ought to be implemented in these

cases since relying on the network structure is not an

option.

Low Latency and Ultra-Reliable Communication

(LLURC): Remote surgical procedures, intelligent

transportation, and industrial process automation

systems all rely on the ability of IoT networks to reliably

and quickly respond to real-time demands.

Safety: As a result of the enormous number of IoT

devices linked to the internet, the private data exchanged

via these systems may be at risk. Privacy and device

security are also vital considerations. One of the most

exciting aspects of IoT is its ability to make timely and

intelligent choices based on the data it processes.

Dynamic changing network: An enormous number of

IoT devices need proper management. These devices

may behave dynamically. For example, when a device

goes to sleep or wakes up, it is determined by various

factors, including the software it is running.

The commercialization of IoT services and

applications is heavily dependent on security and

privacy. Various sectors, such as health care and

business, have been impacted by security breaches that

range from basic hacking to well-coordinated intrusions

at the corporate level. Due to their restrictions and the

environment in which they operate, IoT devices and apps

face significant security issues. IoT security and privacy

concerns have been thoroughly considered from different

viewpoints, including communication privacy and

security, architecture data security, and identity

management as well as malware analysis [4]. Sections 3

and 5 explore more into the issues of security and the

threat model.

According to Fernandes et al. [15], security

challenges in IoT and conventional information

technology (IT) devices are comparable and different. In

addition, they also addressed privacy concerns. Software,

hardware, networks, and applications are some of the

most often cited points of comparison and contrast in this

debate. IoT and conventional IT have a lot of things in

common when it comes to security concerns. Despite

this, the real concern of the IoT is the lack of available

resources, which makes it challenging to implement

advanced security measures in IoT networks. Improved

algorithms and cross-layer architecture are also needed

to address IoT privacy and security concerns. As part of

an overall privacy and security method for IoT, current

security solutions will be nominated for consideration, as

well as new intelligent, resilient, scalable, and

evolutionary methods to handle IoT security concerns.

ML implies intelligent procedures that utilize

previous experiences or example data to understand how

to maximize performance criteria. Algorithms that use

machine learning to develop behavioral models on

massive datasets are known as ML. Because of machine

learning, computers can learn independently, even if no

instructions are provided. The newly included data is fed

into these models, which serve as a foundation for

generating predictions about the future. AI, optimization,

information theory, and cognitive science all have

origins in ML, so it is a multidisciplinary field of study

[16]. ML is no exception. Robotics, voice recognition,

and other areas where people are unable to apply their

skills, such as hostile environments, need the use of

machine learning [17]. It may also be used when the

answer to a particular issue evolves over time. To put this

into context, Google utilizes ML to identify risks to

mobile devices and apps running on the Android

platform. Infected mobile devices may be scanned and

cleaned using this tool. Macie, an Amazon tool that

applies ML to organize as well as categorize data stored

in Amazon's cloud storage, was also released recently.

False positives and true negatives may occur even when

ML methods are used correctly. As a result, if an

incorrect prediction is produced, ML approaches need

direction and change of the model.

Contrary to popular belief, with DL, a new kind

of ML, the model is able to establish its accuracy of

prediction. Prediction and classification tasks in novel

applications of IoT with customized and contextual

support might benefit from the self-service character of

DL models. Moreover, the complete volume of data

produced by IoT networks necessitates the use of DL and

ML methods to offer intelligence to the systems. In

addition, the IoT data created by DL and ML algorithms

can be effectively exploited to make educated and

intelligent choices by the IoT systems. The analyses of

privacy, security, malware, and attack detection are just

a few of the many applications for DL and ML. DL

methods can also be employed in IoT systems to conduct

identification tasks and complicated sensing to develop

new apps and services that take into account real-time

interactions among people, the physical environment,

and smart devices.

Real-world uses of ML in security include the following:

(i) Different handwriting styles are used for character

recognition in security encryption.

(ii) Recognition of faces in forensics: lighting, pose,

occlusion (beard, glasses), hairstyle, make-up, etc.

(iii) Software and apps that contain malicious code need

to be identified.

(iv) Behavior analysis is used to identify DDoS attacks

on infrastructure. On the other hand, there are several

difficulties associated with applying DL and ML in IoT

applications. For example, designing an appropriate

model for processing data from many IoT applications is

challenging. In the same way, correctly classifying input

data is likewise a complex undertaking. The use of little

marked data in the learning process is also tricky. Using

these models on IoT devices with limited processing and

storage resources presents further difficulties [18]. Like

essential infrastructure and real-time applications, DL

and ML algorithms produce anomalies. IoT security

solutions that use DL and ML must be thoroughly

analyzed in this context.

1.2. Contributions

The main influences of this work are presented below:

 A review of various types of attacks with its example

is discussed.

 Comprehensive analysis of ML and latest

developments in IoT defense DL methods: the most

promising DL and ML algorithms are examined for

IoT protection schemes, and their benefits,

drawbacks, and implementations are addressed in

the security of IoT systems. In addition, compare,

and description tables are provided for DL and ML

approaches for learning lessons.

 A number of state-of-art applications of DL and ML

in IoT security and privacy are illustrated.

 We offer a taxonomy of the most recent IoT privacy

and security solutions based on deep learning and

machine learning techniques. Moreover, new

insights of ML and DL in IoT securities are

illustrated.

 Potential limitations, challenges, future directions,

and suggestions of DL and ML are appeared, how

they could help the recent and future research.

The work is presented as follows: Literature reviews

with their limitations and why this work is needed are

illustrated in Section 2. Next, applications of ML or DL

to IoT threats are illustrated in Section 3. Moreover,

Section 4 appears in the DL and ML models, where we

can find how to work with each ML and DL model in IoT

security, and solutions are also described in Section 5. In

section 6, we can see a number of new insights into Deep

and Machine Learning for IoT Security that can help

future research works. Section 7 discusses challenges,

limitations, and future directions. A number of

suggestions and recommendations are presented in

Section 8, and Section 9 shows the conclusion.

2. Literature Reviews

A number of surveys or reviews have covered IoT

security to offer some guidelines for future challenges.

Though several studies have looked at IoT security, none

have focused on DL or ML applications for IoT security.

Several works [19-25] have been reviewed for

motivating and organizing the challenges in access

control, authentication, application security, encryption,

and network security in IoT environments. The survey of

[26] provided a survey of IoT communication on security

issues with its solutions. Another paper [27] emphasized

IoT systems for intrusion detection.

Moreover, IoT frameworks for regulatory

approaches and legal issues can determine security and

privacy requirements [28]. The context of distributed IoT

has also covered privacy and security in [29]. These

works also concerned various challenges. Several issues

must be found out, and the researchers assert that the

distributed IoT method offers numerous advantages in

terms of privacy and security. The survey of [30]

described evolving threats and vulnerabilities in IoT

devices, for example, threats of ransomware as well as

security concerns. The authors of [31] concisely

indicated the context of IoT using ML techniques

concerning data security and privacy protection. This

survey also described three challenges with respect to

ML application in IoT environments (i.e.,

communication and computation overhead, partial state

consideration, and backup security justifications).

Numerous survey studies, including [31, 32], have

examined the use of data mining and ML methods in

cybersecurity to assist intrusion detection. Above all,

they reviewed anomaly detections and misuse in

cyberspace [32]. The methodology was based on several

classes of AI (artificial intelligence) methods from the

point of view of an IoT context, and the opportunities of

applying those approaches in IoT environments were

observed. However, that work did not emphasize the

implementation of DL an IoT perceptive. The review of

[3] also provided ML techniques in IoT security where

they offered future challenges and current solutions.

Another survey of ML methods for wireless sensor

networks (WSNs) was appeared in [33].

The motivation of that study was to review ML

methods in real-life WSN applications. i.e., clustering,

localization, routing and unrealistic aspects of quality of

service (QoS) and security. The framework of WSNs in

DL methods was described in the work of [34]. On the

other hand, this work emphasizes network configuration.

Besides, it differs from the proposed survey that focuses

on DL/ML methods for ensuring IoT security. Some

traditional ML techniques [35] were considered with

advanced methods, including DL methods [36] for

processing big data. Above all, the relationship of several

ML techniques for signal processing approaches was

focused on investigating and processing relevant big

data. An overview of DL is offered on state-of-the-art

approaches [37]. The survey proposed the opportunities

and challenges of various existing solutions with their

uses and evolution. The essential principles of several

DL classifiers were evaluated with their procedures in

addition to developments of DL methods in several uses

[38, 39], for example, speech processing, pattern

recognition, and computer vision. In mobile advertising,

a review of improvement in DL methods was used for

recommendation systems, which show a crucial role

[40]. Various effective ML applications [41] were

similarly conducted in self-organizing networks. The

survey focused on the merits and demerits of various

methods and offered future opportunities and challenges

in expanding artificial intelligence and future network

design [42]. The significance of 5G in artificial

intelligence was highlighted. Intrusion detection using

data mining was covered in [43]. Application of

multimedia mobile was surveyed conducting DL

methods as well [44]. Recent DL techniques in mobile

security, speech recognition, mobile healthcare,

language translation, and ambient intelligence were

focused. Similar research was conducted on the most

advanced, state-of-the-art deep learning approaches used

in a diversity of IoT data analytics applications [45]. On

the other hand, our survey covers a complete review of

recent progress in deep learning approaches and cutting-

edge machine learning approaches for ensuring security

in IoT. This review compares and identifies the

advantages, prospects and weaknesses of different

DL/ML approaches for security in IoT. This paper also

discusses numerous future directions and challenges and

discuss the realized problems and future prospects based

on a study of possible DL/ML applications in the context

of IoT security, thus offering an effective guideline for

researchers or scholars to modify the security of IoT

environment from simply empowering a secure

communication between IoT modules to end on IoT

security on the basis of intelligence methods.

3. IoT Threat

Several heterogeneous sensing systems

communicating with one another through a local area

network (LAN) are referred to as the IoT [46]. The risks

in IoT are distinct from those posed by traditional

networks, owing in large part to the capabilities

accessible to end devices [47]. The traditional Internet

relies on powerful computers and servers with plenty of

resources, while the IoT relies on equipment with low

memory and computing power. That is why an IoT

device in the real world cannot continue employing

multifactor authentication and dynamic protocols like a

regular network. Wireless protocols applied by IoT

devices, for example, ZigBee and LoRa are less secure

than those used by traditional networks. A lack of a

standard operating system and particular features inside

IoT applications have resulted in various data contents

and formats in the systems, making the creation of a

uniform security protocol complicated [48]. There are

several security and privacy problems associated with the

IoT because of these flaws. As a network grows in size,

the risk of an attack rises. Since the IoT has no firewalls,

its network is more vulnerable than a traditional office or

company network. IoT systems that exchange data with

one another are frequently multi-vendor systems,

adhering to a wide range of spectra and protocols from

different manufacturers. The connection between such

devices is difficult, necessitating the use of a trustworthy

third party as a bridge [49]. Additionally, many reports

have posed concerns about how billions of smart devices

receive app updates [50, 51]. Since an IoT device has

small computing resources, its ability to cope with

advanced threats is harmed. To conclude, IoT

weaknesses may be classified as either essential or

widespread. For example, although vulnerabilities such

as battery drain attacks, insufficient standardization, and

insufficient confidence are exclusive to IoT systems,

vulnerabilities in internet-inherited systems may be

considered general. Numerous IoT risks have been

identified and classified in the past [32, 52-55]. We

address the most often identified challenges to the IoT

over the last ten years and try to categorize them into

privacy and protection classifications. Privacy and

security are basic principles that turn around network

availability [56-58]. On the internet of things, data may

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

1ReviewArticleMachineandDeepLearningforIoTSecurityandPrivacy:Applications,Challenges,andFutureDirectionsSubratoBharati1,PrajoyPodder21,2InstituteofInformationandCommunicationTechnology(IICT),BangladeshUniversityofEngineeringandTechnology(BUET),Dhaka-1205,BangladeshCorrespondenceshouldbeaddressedtoSu...

展开>> 收起<<

1 Review Article Machine and Deep Learning for IoT Security and Privacy.pdf

共46页,预览5页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

1 Review Article Machine and Deep Learning for IoT Security and Privacy

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: